Agency automation guide
25 Automations Every Agency Should Have
Agency automation means setting up software rules to run your own agency's repeat work: replying to new leads, sending invoices, setting up new clients, chasing renewals. It is not the same as hiring an automation agency. This guide lists 25 automations for small marketing, creative and digital agencies, in the order the work happens, from the first enquiry to the monthly profit report. Each one has its own one-page guide with a trigger, the steps and the result. We mark which step a plain rule does, which step Claude AI does and where a person approves. 2027 is the planning year, so you do not need all 25 at once. Pick a few, build them first and add the rest as your agency grows. You do not need to code. You do need to know how your own agency works, and each guide tells you what to have ready.
How to read each one-pager
Every one-pager opens with a trigger, the event or schedule that starts it. Then come the steps and the result you get. Each step in the diagram shows who does the work. A Rule is a fixed instruction, such as wait three days or add up the hours. Claude AI reads or writes text. A person approves or decides. An app, such as Gmail or Slack, does its own job.
Stage 1 of 6
Win the lead
Answer new enquiries fast, learn who they are and get them to a call.
Speed to Lead Automation: Reply to Every Enquiry in a Minute
Speed to lead automation sends a personal first reply about a minute after a form submission, day or night. n8n catches the form, HubSpot saves the lead, Claude drafts the reply and Gmail sends it.
The problem
A new enquiry lands on Friday at 4:50 PM. Nobody sees it until Monday. By then the lead has booked a call with another agency.
How it works
A website form submission starts it. n8n catches the form, saves the lead in your CRM, and Claude drafts a short reply that repeats the request in plain words. Gmail sends it from a named person's mailbox with your booking link, Slack tells your team, and a sheet logs the reply time.
Where AI fits
Claude writes the opening line so it refers to what the lead actually asked, and it flags vendor pitches and spam. A fixed template with the lead's name and chosen service also works for a first version, so AI is an upgrade, not a requirement.
What you need
n8n (Zapier or Make work in its place), a Claude (Anthropic API) key, a Webflow or WordPress form, HubSpot, Gmail, Slack and Google Sheets, plus your booking link and 3 to 5 real past replies as tone examples.
Step by step
Catch form submissionRulen8n
n8n catches the form submission through a webhook and answers right away, so the form does not time out.
Save to CRMAppHubSpot
HubSpot creates or updates the contact by email, so a repeat enquiry updates one record instead of making a duplicate.
Draft the replyClaude AIClaude
Claude reads the message and a one-page summary of your services, then returns a subject and a short body with no prices and no promised dates.
Send the replyAppGmail
Gmail sends the reply from a named person's real mailbox with your booking link, or saves it as a draft during the first two weeks.
Alert the teamAppSlack
Slack posts the lead summary, what was sent and the CRM link to your sales channel.
Log reply timeAppGoogle Sheets
Google Sheets stores when the lead submitted and when you replied, so you can see your real response time.
Build time
Our estimate: 3 to 5 hours. Basis: 1 hour for the form webhook and a test, 1 hour for the CRM and Slack steps, 1 to 2 hours for the prompt and tests on 10 real past enquiries, and 1 hour for Gmail sign-in and a draft-mode trial.
Time saved
Our estimate: about 1 hour a week. Basis: 10 enquiries a week at 5 minutes each to write and send a first reply is 50 minutes, and the bigger gain is reply speed, not hours.
Watch out
Auto-replies to bot submissions and double sends damage your sender reputation, so add a honeypot or CAPTCHA, drop fake emails, ignore repeats within 10 minutes, and send from a real mailbox on your own domain with SPF and DKIM set up, never a no-reply address. Check first: which Webflow or WordPress plan lets your form send a webhook, and test the whole flow on one lead.
The number to watch
Track the median minutes from form submit to first reply (aim for under 5) and the share of enquiries that go on to book a call.
Questions about this automation
Should the AI reply send automatically or wait as a draft?
Start with drafts. The Gmail step can save a draft, and a Slack ping tells a person to look. Switch to auto-send after you have read about 20 drafts in a row that needed no edits.
Will an instant reply feel robotic?
Not if it names the specific request, is signed by a real person and says a human will follow up. One line saying it is an automatic first reply keeps it honest.
Do I need AI to reply to leads automatically?
No. A fixed template with the lead's name and chosen service works for a first version. Claude adds a reply that refers to what the lead actually wrote.
Related reading
Want every enquiry answered in a minute, even at 4:50 PM on a Friday? Book a free call and we will build it on your form.
Lead Enrichment Automation: Know the Company Before You Call
Lead enrichment automation fills in company facts on every new business lead within about a minute. Apollo supplies industry and headcount, and Claude adds a short summary of what the company does and likely needs.
The problem
Your sales call starts with a Google search. The form only gave you a name, an email and one vague line. You spend minutes on research before you know if the lead deserves a call.
How it works
A new lead starts it, from the form workflow or the CRM. n8n skips free-mail addresses, then asks Apollo about the company's domain. Claude summarises the company's website, HubSpot stores the fields, and weak matches get a tag so a person takes a look.
Where AI fits
Claude turns the company website into a short summary and a likely need. Facts such as headcount and industry come from Apollo, not from the AI, so they are not guessed.
What you need
n8n (Zapier or Make work in its place), an Apollo API key, a Claude (Anthropic API) key and HubSpot with custom fields for industry, size, tech, summary and enrichment status.
Step by step
Get the new leadRulen8n
n8n picks up the new lead, either from the same form workflow or from a schedule that pulls recently created HubSpot contacts.
Skip free emailRulen8n
A rule skips company lookups for gmail.com, outlook.com and similar domains and marks the lead as personal email.
Look up companyAppApollo
This step needs a custom HTTP request to Apollo's Organization Enrichment, which returns industry, estimated employees, revenue and technologies at 1 credit per company.
Summarise the companyClaude AIClaude
Claude, limited to searching the company's own domain, returns what the company does, a likely need and the source URL.
Update the CRMAppHubSpot
HubSpot stores the fields as custom properties, so the data sits where your team already works.
Flag the missesRulen8n
If nothing matched, the record gets an enrichment failed tag, so a person looks instead of trusting blanks.
Build time
Our estimate: 4 to 6 hours. Basis: 1 hour for the API key and test calls, 1 hour for the free-mail filter and field mapping, 2 hours for the Claude prompt and JSON schema, and 1 to 2 hours for CRM properties and the failure path.
Time saved
Our estimate: about 1.5 hours a week. Basis: 15 leads a week at 6 minutes of manual company research each is 90 minutes.
Watch out
The wrong company can get matched, for example from a personal email, a consultant on their own domain or a forwarding domain, so skip free-mail domains, store the source of every field, and leave a field blank with a check tag when the match is weak. Check first: always send Apollo the domain, and test whether your HubSpot step can write custom properties, because an HTTP request to HubSpot's API is the fallback.
The number to watch
Track the enrichment hit rate, meaning the share of business-email leads that come back with industry and company size filled, plus the cost per enriched lead.
Questions about this automation
Is it legal to enrich leads with company data?
Company-level data is usually low risk, but personal data about people in the EU or UK falls under GDPR, so cover it in your privacy policy and ask a lawyer about your case. Apollo's own docs say it will not reveal personal emails for people in GDPR-compliant regions.
What does lead enrichment cost per lead?
Apollo's docs list 1 credit per company lookup, so your cost depends on your plan. Claude web search costs $10 per 1,000 searches plus normal token costs (as of Nov 2026, check current pricing), so two searches add about 2 cents in search fees.
Will the AI make up facts about the company?
Headcount, industry and revenue come from Apollo, not from Claude. Claude only writes the summary from the company's own website and returns the source URL, and a weak match is left blank and tagged for a person to check.
Related reading
Tired of Googling every new lead? Book a free call and we will build enrichment into your CRM.
Lead Scoring Automation: Rank Every Enquiry Before You Reply
Lead scoring automation gives each new enquiry a score from 0 to 100, a tier and a one-line reason. Your rules set the number, and Claude only reads the message to label intent and red flags.
The problem
Every enquiry gets the same slow treatment. The biggest retainer waits behind the lead who is only browsing. You have no quick way to tell them apart.
How it works
A lead finishes enrichment, and that starts it. n8n adds points for budget, service fit, company size, timeline and role, while Claude reads the message and labels intent. n8n combines both into a tier, HubSpot stores it, and Slack sends top-tier leads to the owner at once.
Where AI fits
Claude reads the free-text message, which fixed rules cannot do, and labels intent and red flags. The final number comes from your rules, so you can explain every score, and for a form with only fixed fields plain rules are enough.
What you need
n8n (Zapier or Make work in its place), a Claude (Anthropic API) key, HubSpot with custom properties for score, tier and reason, Slack, and 20 to 40 past won and lost deals to set the weights.
Step by step
Receive enriched leadRulen8n
A webhook or an enrichment-complete flag hands the lead's form answers and company data to n8n.
Score with rulesRulen8n
A Code or Set step adds points for budget band, service fit, company size, timeline and role, using weights you set from past won and lost deals.
Read the messageClaude AIClaude
Claude returns the lead's intent (ready to buy, researching, vendor pitch, job seeker or spam), scope clarity, red flags and a one-line reason.
Combine and tierRulen8n
Rules in n8n combine the points with small adjustments based on Claude's labels, for example A at 70 or more, B at 40 to 69 and C below 40, with spam forced to zero.
Write score to CRMAppHubSpot
HubSpot stores the score, tier and reason as properties, so lists and reports can use them.
Route by tierAppSlack
Slack pings the owner for A-tier leads, B-tier leads land in a same-day queue and C-tier leads trigger a polite self-serve email.
Build time
Our estimate: 4 to 8 hours. Basis: 2 hours reviewing past deals to set weights, 1 hour for the rules step, 2 hours for the Claude schema and tests on 20 old leads, and 1 to 2 hours for CRM properties and Slack routing.
Time saved
Our estimate: about 2 hours a week. Basis: 10 leads a week at 12 minutes of manual triage each is 120 minutes, and this depends heavily on your volume.
Watch out
The score drifts away from reality when nobody checks it against closed deals, so once a month compare tier with outcome, change the weights if A-tier leads do not close more often than C-tier leads, and keep personal traits such as name, gender or nationality out of the score. Check first: the 70 and 40 cut-offs are an example, not a benchmark, so set yours from your own deals.
The number to watch
Track win rate by tier. A-tier leads should close clearly more often than C-tier leads, and if they do not, change the weights.
Questions about this automation
Do I need AI for lead scoring?
No. Rules on budget, size and service fit cover most of it. AI helps only when the lead writes a free-text message, to spot intent, spam and vendor pitches.
Does my CRM already do lead scoring?
HubSpot's lead scoring tool is in Marketing Hub and Sales Hub Professional and Enterprise, per HubSpot's docs. Pipedrive says custom scoring needs Premium or higher, and its calculation page says scores are for deals, so check lead support in your own account. On lower plans, n8n fills the gap.
How do I choose the weights and tiers?
Look at 20 to 40 past won and lost deals and note the source, budget, company size and service of each. Give more points to what the won deals share, then test the score on 20 old leads before you trust it.
Related reading
Want your best leads to get the first reply, with a reason on every score? Book a free call and we will build it.
Calendly Follow Up Email Automation: Chase Leads Who Do Not Book
Calendly follow up email automation nudges leads who got your booking link but never booked. It sends a short reply at 24 and 72 hours, checks Calendly first, and stops once the lead books.
The problem
You send a Calendly link and the lead says nothing. The thread dies because nobody remembers to chase it. Nobody owns the chase, so it never happens.
How it works
It starts when a booking link goes out to a lead. n8n logs the link and watches Calendly for a booking. If none arrives in 24 hours, Claude writes a short nudge and Gmail sends it in the same thread. A second nudge goes at 72 hours, then the lead is tagged no booking, and anyone who books is dropped from the chase.
Where AI fits
Claude makes each nudge refer to what the lead asked. A fixed three-line template also works, so AI is optional here.
What you need
n8n (Zapier and Make both have Calendly triggers), Calendly with a personal access token or OAuth app, Gmail, Google Sheets and a CRM tag for no booking, plus a Claude (Anthropic API) key if you want nudges written to fit each lead.
Step by step
Log link sentAppGoogle Sheets
Google Sheets stores the lead's email, original message, link sent time and a nudge count of 0 when the link goes out.
Catch the bookingAppCalendly
The Calendly trigger fires on each new booking, and the workflow marks the matching row as booked.
Check every hourRulen8n
n8n runs on a schedule every hour and finds rows that are not booked and are 24 or 72 hours past the link time with the right nudge count.
Confirm no bookingRuleCalendly
This step needs a custom HTTP request that asks Calendly for active events by the lead's email, so a booking you would otherwise miss is caught.
Write the nudgeClaude AIClaude
Claude writes three short sentences that refer to the original request and re-offer the link, with no pressure and no invented details.
Reply and tagAppGmail
Gmail replies in the original thread, then n8n raises the nudge count and, after the second nudge, tags the CRM record no booking.
Build time
Our estimate: 3 to 5 hours. Basis: 1 hour for the Calendly connection and a test booking, 1 hour for the tracking table, 1 to 2 hours for the schedule and nudge logic, and 1 hour for tests with a test email address.
Time saved
Our estimate: about 1 hour a week. Basis: 6 unbooked leads a week at 2 chase emails of 5 minutes each, including remembering who to chase, is 60 minutes.
Watch out
A lead who books under a different email than the one on the form looks like a no-show, so ask Calendly for active events by invitee email before every nudge and hold the nudge for human approval when the lead is high value. Check first: webhook access differs by Calendly plan, so test on your own plan, and check how you pre-fill the lead's email, because Calendly documents it only for the JavaScript embed.
The number to watch
Track the booking rate within 7 days of the link being sent, meaning links sent versus bookings made.
Questions about this automation
Can Calendly send these follow-ups by itself?
Calendly Automations can email or text after a booking, before it starts, after it ends and after a no-show. Every listed trigger starts from a booked event, so none fires for a person who never books.
Do Calendly webhooks need a paid plan?
Calendly's developer FAQ says webhooks are on all plans including Free, but routing form submission, contact and meeting recap events need a paid plan. Make's Calendly app page still says a paid subscription is required, so test on your own plan first.
How many follow-ups does it send?
Two: one at 24 hours and one at 72 hours. Then it stops and tags the lead no booking, and you can change the wording and the send hours.
Related reading
Want every booking link chased politely, and nobody nagged twice? Book a free call and we will build it.
Automated Meeting Reminders That Survive Cancels and Reschedules
Automated meeting reminders email your client 24 hours and 1 hour before a call and stay correct after any cancel or reschedule. Your host also gets a short prep brief in Slack 30 minutes before.
The problem
Clients forget the call or show up unprepared, because the only reminder was the invite sent two weeks ago. Your host also has to look up the client's notes by hand before each call.
How it works
A Calendly booking starts it. n8n saves the meeting in a sheet and checks it every 15 minutes. Just before it sends, it confirms the meeting is still active, then Gmail emails the client 24 hours and 1 hour ahead. 30 minutes before the call, Claude turns CRM notes and form answers into a five-line brief and Slack posts it to the host.
Where AI fits
The reminders need no AI: a fixed template with the time, join link and reschedule link is enough. Claude writes only the host's prep brief, and you can skip that part.
What you need
n8n (Zapier or Make work in its place), Calendly with a personal access token or OAuth app, Gmail, Slack, Google Sheets and HubSpot, plus a Claude (Anthropic API) key if you want the host brief.
Step by step
Catch the bookingAppCalendly
The Calendly trigger passes the client's email, start time, timezone, cancel and reschedule links and form answers to n8n.
Save the remindersAppGoogle Sheets
One row per meeting stores start time, email and status, so a cancel or a reschedule updates the row instead of leaving a stale reminder.
Check every 15 minutesRulen8n
n8n runs every 15 minutes and finds active meetings that start in about 24 hours or about 1 hour and have no reminder sent yet.
Confirm still activeRuleCalendly
This step needs a custom HTTP request to Calendly that checks the event is still active just before sending.
Send the reminderAppGmail
Gmail sends a short email with the time in the client's timezone, the join link and the reschedule link, then marks the row as sent.
Brief the hostClaude AIClaude + Slack
30 minutes before, Claude turns CRM notes and booking answers into a five-line prep brief and Slack posts it to the host.
Build time
Our estimate: 4 to 6 hours. Basis: 1 hour for the Calendly connection, 1 hour for the tracking table with cancel and reschedule handling, 1 hour for the schedule and email send, 1 hour for templates, and 1 to 2 hours for the host brief.
Time saved
Our estimate: about 45 minutes a week. Basis: 10 meetings a week at 2 minutes to remind plus 2 minutes to look up the client is 40 minutes, and the real gain is fewer missed calls, though we claim no no-show rate.
Watch out
Reminders can go out for canceled or moved meetings because a timer does not know the booking changed, so handle both the canceled and created events (a move sends a cancel marked as rescheduled), update the stored row and re-check the event in Calendly right before sending. Check first: text reminders sent from n8n are untested here, so start with email and test with a booking you then move and cancel.
The number to watch
Track the no-show rate, meaning the share of booked meetings where the invitee does not attend, before and after, using Calendly's no-show marks.
Questions about this automation
Does Calendly already send reminders?
Yes. Calendly Automations send emails and texts before and after events on the Professional, Standard, Standard Plus, Teams, Teams Plus and Enterprise plans, so check your plan. Use n8n only when you need meetings outside Calendly, a Slack prep brief or CRM-aware messages.
Should I use the n8n Wait node for reminders?
You can, but a stored row plus a 15-minute schedule is easier to fix after a reschedule. Wait node times use the n8n server time, not the workflow timezone, and waits over 65 seconds are saved to the database.
Can I send text message reminders?
Calendly's own text reminders work, but they need invitee opt-in, cap each text at 180 characters, are not sent during a trial and are not supported in Singapore, Russia or Iran. Texts sent from n8n through another service are not checked here, so test them before you rely on them.
Related reading
Want reminders that stay right after every reschedule, and hosts who walk in prepared? Book a free call and we will build it.
Stage 2 of 6
Calls, proposals and contracts
Follow up after the call, write the proposal and send the contract.
Follow up email after no show: send it the same day
Send your follow up email after a no show the same day. One Slack click marks the missed sales call, Claude fits your approved template to the lead, and the rep's own Gmail sends it.
The problem
A prospect books a sales call and never joins. Nobody has time to chase it the same day, so the lead goes cold. The follow-up is always the next thing to do, and it slips.
How it works
Fifteen minutes after each call ends, n8n asks the rep in Slack whether the prospect showed, and a No click marks the no-show in Calendly. That fires a webhook, n8n finds the lead in HubSpot, Claude fills your template, and Gmail sends the email from the rep's own address. Three days later n8n sends one last short nudge, but only if there is no reply and no new booking.
Where AI fits
Claude personalizes a template you approved, using one detail from the booking form, and keeps the email to one short paragraph. The timing, the stop rules and the sending are plain rules, not AI.
What you need
n8n (Zapier or Make through a generic webhook catch), Claude, Calendly with a personal access token and webhook signing key, HubSpot or another CRM, the rep's Gmail or Outlook, Slack, and your approved no-show template with a booking link.
Step by step
Mark the no-showA personSlack
n8n asks the rep in Slack 15 minutes after each call ends, and a No click marks the no-show in Calendly (reps can also mark it in Calendly themselves).
Catch the webhookRulen8n
An n8n Webhook node catches Calendly's no-show event and checks the signature header against your signing key.
Look up the leadAppHubSpot
n8n finds the contact and deal in HubSpot by email and reads the booking-form answers (test first that the no-show event carries them).
Write the emailClaude AIClaude
Claude fills your approved template in the rep's voice, using one detail from the booking form, and returns a subject and a body.
Send from the repAppGmail
The email goes out from the rep's own mailbox, is logged on the deal, and the deal is tagged 'No-show, touch 1'.
Wait, check, nudgeRulen8n
A plain 3-day wait in n8n checks for a reply or a new booking and sends one last short nudge only if both are empty.
Build time
Our estimate: 3 to 5 hours. Basis: one API call to create the webhook subscription, about 6 steps, one email template and one Slack prompt, plus test runs with a fake booking.
Time saved
Our estimate: about 1 hour a week. Basis: 4 no-shows a week, at about 15 minutes each to find the lead, write the email and set a reminder to chase.
Watch out
Calendly does not detect no-shows by itself and reps forget to mark them, so the fix is to have n8n ask the rep in Slack 15 minutes after each call ends and mark the no-show through Calendly's API when the answer is No. Check your Calendly plan before you build, because sources disagree on which plans allow no-show events, and test it on one lead first.
The number to watch
Share of no-shows who rebook within 7 days.
Questions about this automation
How soon should I send a follow-up email after a no show on a sales call?
The same day is the usual advice, ideally within an hour or two while the missed slot is fresh. Keep it short, make no accusation, and put one booking link in it.
Does Calendly mark no-shows automatically?
No. Calendly's help center describes marking a no-show as a manual step on the meeting. You can mark it yourself, or have n8n ask the rep in Slack and mark it through Calendly's API.
Does Claude write the whole email?
Claude fills a template you approved, using one detail from the booking form. The offer and the tone come from your template, and the timing and sending are plain rules.
Related reading
Want every no-show chased the same day? Book a free call and we build it on your Calendly and inbox.
Meeting notes automation: summaries, CRM notes and tasks
Meeting notes automation turns every recorded client call into a short summary, decisions, and action items with owners. It posts them to Slack, saves them on the CRM record, and drafts the tasks for one-click confirm.
The problem
After every client or sales call someone has to write up notes, update the CRM and chase action items. On a busy day that work gets skipped. Some of it never gets done.
How it works
When your recorder (Fireflies or Fathom) finishes processing a call, it sends a webhook to n8n. n8n fetches the transcript, and Claude turns it into a fixed structure: summary, decisions, action items with owner and due date, and client concerns. The notes are saved to the contact and deal in HubSpot, posted to Slack, and each action item becomes a task marked 'AI draft, confirm'.
Where AI fits
Claude reads the transcript and returns a fixed structure of summary, decisions, action items with owners and risks, in your own format. Your recorder already transcribes the call, and the webhook, matching and posting are plain rules and apps, not AI.
What you need
n8n (Zapier or Make are options), Claude, a Fireflies or Fathom account with API access, HubSpot, Slack, a task tool such as ClickUp, and the summary format you want.
Step by step
Call ends, webhook firesAppFireflies
Fireflies or Fathom posts to an n8n Webhook node as soon as the transcript is ready.
Fetch the transcriptRulen8n
n8n fetches the transcript: for Fireflies a GraphQL call with the meeting ID, while Fathom can send the summary and transcript inside the webhook itself.
Extract the factsClaude AIClaude
Claude reads the transcript and returns a fixed structure of short summary, decisions, action items with owner and due date, client concerns, and any budget or timeline mentions (a check step should confirm the structure is valid before anything is saved).
Save to the CRMAppHubSpot
n8n matches attendees by email and attaches the summary as a note on the contact and deal in HubSpot (check that the n8n HubSpot node can add notes, or use the HubSpot API).
Post and create tasksAppSlack
The summary goes to the client's Slack channel, and each action item becomes a ClickUp or Asana task flagged 'AI draft, confirm'.
Build time
Our estimate: 4 to 6 hours. Basis: one webhook, one transcript fetch, one Claude prompt with a fixed output structure and three output steps, and most of the time goes into tuning the prompt on 3 or 4 real transcripts.
Time saved
Our estimate: about 5 hours a week. Basis: 10 calls a week at about 30 minutes each to write notes, update the CRM and create tasks, minus a few minutes of review.
Watch out
Claude can assign an action item to the wrong person or invent a due date when the transcript is vague, so the fix is to require owner and due date to be quoted from the transcript, leave them blank when unclear, and mark every created task 'AI draft, confirm'. Check your Fireflies or Fathom plan for API and webhook access, and check your local recording-consent rules (this is not legal advice), before you switch it on.
The number to watch
Share of calls with a summary posted within 15 minutes of ending.
Questions about this automation
Can I automate meeting notes without a note-taker bot joining the call?
Only if your recorder supports it. Fathom and Fireflies both normally join as a bot, so tell clients and check your local recording-consent rules before you switch this on.
Which tool do I need for the transcript, Fireflies, Fathom or Otter?
Fireflies and Fathom both document webhooks and an API that n8n can call. Otter's public API was reported as Enterprise-only by third-party sources, so check it before you plan around it.
Does the AI turn the call audio into text?
No. Your recorder makes the transcript. Claude only reads it and puts it into your format, and each task waits for a one-click confirm.
Related reading
Tired of typing up calls? Book a free call and we build the summary flow on your own recorder and CRM.
AI proposal generator for agencies: a first draft in minutes
An AI proposal generator for agencies drafts the scope from your discovery call and builds the document from your template. Claude writes the narrative, a lookup sets every price from your rate card, and the owner approves.
The problem
Writing each proposal from scratch can take about two hours of a senior person's day (our estimate). The quality also changes with who writes it.
How it works
When the owner moves a deal to 'Proposal needed' in HubSpot, n8n gathers the deal, contact, call summary and booking-form answers into one brief. Claude drafts goals, deliverables and phases and picks services by ID from your rate card, then a plain lookup adds the prices from your Google Sheet. n8n fills a Google Doc from your template, and Slack sends the owner the link with Request changes and Approve buttons.
Where AI fits
Claude writes the narrative parts, which are goals, approach and phases, and chooses services from your rate card. It never sets prices: a plain lookup does, so a number cannot be invented.
What you need
n8n (Zapier or Make are options), Claude, HubSpot, Google Docs and Google Sheets, Slack, a proposal template with placeholders, a rate card sheet, and 3 to 5 past winning proposals as style examples.
Step by step
Deal hits stageAppHubSpot
n8n's HubSpot trigger fires when a deal changes stage and a filter keeps only your 'Proposal needed' stage (n8n's docs say only one HubSpot trigger can be active at a time, so check your other workflows first).
Gather the briefRulen8n
n8n pulls the deal, contact, company, the call summary from automation 7 and the booking-form answers into one text block.
Draft the scopeClaude AIClaude
Claude returns goals, deliverables, phases and the service IDs it chose from your rate card, and it never writes prices.
Price from the sheetRuleGoogle Sheets
A plain code step looks up each service ID in your rate card and adds the totals, so every number comes from your sheet.
Build the documentAppGoogle Docs
n8n copies your proposal template, fills its placeholders and saves the document to the client folder (the built-in Google Docs node may not fill placeholders, so this step may need a custom HTTP call to the Google Docs API).
Ask for a reviewA personSlack
Slack sends the owner the link with a Send and Wait approval, and the deal moves to 'Proposal in review'.
Build time
Our estimate: 8 to 12 hours. Basis: turning a real proposal into a template and a clean rate card takes most of it, plus a few rounds of prompt tuning against past deals.
Time saved
Our estimate: about 6 hours a month. Basis: 4 proposals a month, each taking about 2 hours by hand and about 30 minutes to review and edit.
Watch out
A model that sees only a thin brief can invent scope your agency cannot deliver, so Claude can choose only from rate-card service IDs, a code step rejects unknown IDs and blank fields, and a person reviews every draft before it reaches the client. Check your HubSpot plan for API access to deals before you build, and treat the time figures here as estimates, not measurements.
The number to watch
Minutes from stage change to a reviewed draft, and the share of drafts sent with light edits.
Questions about this automation
Will an AI proposal generator make up prices?
It can if you let it. Keep prices out of the prompt: Claude picks service IDs and a lookup against your rate card adds the numbers.
Do I need PandaDoc to do this?
No. A Google Doc template works for drafts. PandaDoc helps when you want pricing tables and e-signature in one place, and its API creates documents from a template with your tokens and recipients. n8n has no built-in PandaDoc node, so that route uses a custom HTTP call.
Can the proposals follow my own style?
Yes. You give it your template and 3 to 5 past winning proposals as style examples. The owner reviews each draft before it goes out.
Related reading
Want proposals that start as a draft? Book a free call and we build it around your template and rate card.
Business proposal follow up email: send it on a timer
A business proposal follow up email should go out on a schedule, not on memory. n8n waits 3 days, checks the client has not replied, Claude drafts a short nudge, and the rep approves it before it sends.
The problem
Proposals go out, the owner gets busy, and the follow-up that could have won the deal never gets sent. The deal goes quiet and nobody notices.
How it works
When a deal moves to 'Proposal sent' in HubSpot, n8n waits 3 days and checks the inbox and the deal for any movement. If the client has not replied, Claude drafts a short email that points to one specific part of the proposal, and the rep approves it in Slack before Gmail sends it. A second, different nudge goes out on day 10, and after that n8n creates a call task instead of another email.
Where AI fits
Claude varies the wording and picks one real detail from the proposal for each nudge. The schedule, the stop checks and the sending are plain rules.
What you need
n8n (Zapier or Make are options), Claude, HubSpot or another CRM, the rep's Gmail or Outlook, Slack, a proposal summary stored on the deal, and your follow-up gaps in days.
Step by step
Proposal marked sentAppHubSpot
n8n's HubSpot trigger fires when the deal moves to 'Proposal sent' and n8n stores the deal ID and send date (n8n's docs say only one HubSpot trigger can be active at a time).
Wait 3 daysRulen8n
A Wait node pauses the run until the first follow-up date, so check how long your n8n setup keeps a waiting run before you rely on a 10-day wait.
Check nothing changedRuleGmail
n8n searches the inbox for a reply from the client's domain and re-reads the deal stage, and it stops the whole run if either shows movement (check that your Gmail step can search by sender domain).
Draft the nudgeClaude AIClaude
Claude writes 3 to 5 sentences in the rep's voice, points to one line from the proposal summary, and offers a 15-minute call.
Review and sendA personSlack
The draft goes to the rep in Slack with a Send and Wait approval, and on approve it is sent from Gmail and logged on the deal.
Repeat, then escalateRulen8n
n8n repeats once at day 10 with a different angle, then creates a call task instead of another email.
Build time
Our estimate: 3 to 5 hours. Basis: one trigger, two Wait steps, a reply check, one prompt and an approval step, plus testing with a dummy deal.
Time saved
Our estimate: about 1.5 hours a week. Basis: about 6 open proposals, each needing a check and a short email about every 5 days at roughly 15 minutes.
Watch out
The follow-up can go out after the client already replied by phone or signed, which looks careless, so re-check the deal stage and inbox right before every send and cap the sequence at 2 emails plus a call task. The 3-day and 10-day gaps are common practice, not a measured rule, so test the schedule on one dummy deal and tune it to your sales cycle.
The number to watch
Share of sent proposals that get a reply within 10 days.
Questions about this automation
How long should I wait before following up on a business proposal?
Many teams use about 3 days for the first nudge and about a week later for the second. That is common practice, not a measured rule, so tune it to your sales cycle.
Should the follow-up send itself or wait for my approval?
Start with approval, since a bad send to a client is costly. Switch to auto-send for the second nudge once you trust the drafts.
Will every follow-up sound the same?
No. Claude points to one real line from your proposal summary and writes in the rep's voice, and the day-10 email takes a different angle. The rep still reads each draft before it goes.
Related reading
Want every proposal followed up on time? Book a free call and we build it for your CRM and inbox.
Automate contract sending: from closed won to signature request
To automate contract sending, move the deal to Closed won. n8n checks the legal name, address, signer email and price, builds the contract from your PandaDoc or DocuSign template, and sends it. No AI is needed.
The problem
A client says yes, and then the contract sits for days. Someone copies details into a template and hunts for the right email address.
How it works
When the deal moves to 'Closed won' in HubSpot, n8n loads the deal, company and signer, and stops with a Slack alert if any required field is blank. If everything is filled, n8n builds the contract from your template through a custom HTTP call, waits for the draft, and sends it for signature. Slack tells the owner it is out, and the document link is saved on the deal.
Where AI fits
None: this flow has no Claude step. A template with merge fields and a validation check do the work, and adding AI to a legal document adds risk without saving time.
What you need
n8n (Zapier or Make are options), HubSpot or another CRM with legal name, address, signer name and email, price and start date, a PandaDoc or DocuSign template with tokens and signer roles plus API access, and a Slack channel for alerts.
Step by step
Deal marked wonAppHubSpot
n8n's HubSpot trigger fires on the deal stage change and n8n loads the deal, company and signer contact.
Validate the fieldsRulen8n
An IF step checks legal name, address, signer name, signer email, price and start date, and if anything is blank it stops and pings the owner in Slack.
Build from templateAppPandaDoc
n8n has no built-in PandaDoc or DocuSign node, so a custom HTTP call posts your template ID, the recipients and your tokens (DocuSign uses a template ID and template roles instead).
Wait for draftRulen8n
PandaDoc creates the document in the background, so n8n checks its status until it reads as a draft.
Send for signatureAppPandaDoc
A second custom HTTP call sends the document with a subject and message (DocuSign sends when the envelope status is set to sent).
Log and notifyAppSlack
Slack posts the sent notice, n8n writes the document link to the deal and checks status on a schedule to log when it is signed (test first that the status check reports a signed state).
Build time
Our estimate: 4 to 8 hours. Basis: about 6 steps, but most of the time goes into setting up the template tokens, signer roles and API credentials, and DocuSign adds setup time.
Time saved
Our estimate: about 1 hour a month. Basis: 3 contracts a month at about 20 minutes each of copying and sending. The bigger gain is a shorter time to signature, which we have not measured.
Watch out
A blank or wrong signer email or legal name sends the contract to the wrong person or with gaps in it, so stop at a validation step before the API call, send only when every required field passes, and have a person open the draft first for the first month. Check your PandaDoc or DocuSign plan for API access before you build, and have a lawyer approve the template and the e-signature rules for your clients' countries (this is not legal advice).
The number to watch
Hours from 'yes' to contract sent.
Questions about this automation
Can I automate contract sending without paying for PandaDoc or DocuSign API access?
The API route needs API access on your plan, and the pages we read did not state which plans include it. Check your plan first, or let n8n build the draft and keep the final send as one manual click.
Should an automation send the contract without anyone looking at it?
For a standard agency agreement with a fixed template, many owners allow it after a trial period. Until then, have n8n create the draft and ask for one approval click in Slack before it sends.
Does this flow use AI?
No. A template with merge fields and a check for blank fields do the work. We leave AI out of a legal document because it adds risk without saving time.
Related reading
Want the contract out minutes after the yes? Book a free call and we build it on your template.
Not sure which of the 25 to build first?
Tell us how your agency works on a free call. We will point to the two or three automations that fit and say what each needs from you.
Stage 3 of 6
Get paid
Send the invoice when the contract is signed and chase it until it is paid.
Automatic invoice after contract signed: how it works
An automatic invoice after contract signed means the deposit invoice is created and emailed within minutes of the client's signature, so nobody rebuilds it from the contract by hand. Plain rules do the work. Claude helps only with free-text terms.
The problem
A contract gets signed on Tuesday and the deposit invoice goes out the following week, because someone has to remember to build it. Until it goes out, your deposit sits unpaid.
How it works
When the client signs in PandaDoc, n8n catches the signed-contract event and reads the client, the total and the payment terms. It creates a draft invoice in Stripe and sends it, then posts to Slack and moves the deal to Won in your CRM. The final invoice still goes out at the milestone you set.
Where AI fits
Small. Plain rules read the PandaDoc tokens and pricing table. Claude is used only for free-text contracts: it pulls the deposit percent, net days and amount into a structured form and flags low confidence for a person to approve.
What you need
n8n, a PandaDoc account with an API key and webhook key, a Stripe account, a Slack channel for alerts, and a written deposit rule (percent, net days and which PandaDoc tokens hold the amount), with Claude as an option.
Step by step
Catch signed contractRulen8n
n8n has no built-in PandaDoc node, so a Webhook node catches the signed-contract event, checks its signature and skips repeat deliveries.
Read payment termsRulen8n, Claude optional
Plain rules read the client, total and tokens from the contract, and Claude steps in only when the payment terms are free text.
Create invoiceAppStripe
n8n has no Stripe invoice resource, so a custom HTTP call creates a draft invoice and adds the deposit line, using the document ID so a repeat cannot create a second invoice.
Send to clientAppStripe
A second HTTP call finalizes the invoice, and Stripe emails it to the client with payment instructions.
Log and notifyAppSlack
n8n saves the document ID and invoice ID in a data table, posts to Slack, and moves the deal to Won in your CRM.
Build time
Our estimate: 3 to 5 hours. Basis: one webhook, three Stripe calls and a test pass with a sample contract. We have not measured it.
Time saved
Our estimate: about 10 to 15 minutes per signed contract, or roughly 1 hour a month at 4 contracts. Basis: the manual open, copy, build and send steps. The bigger gain is cash arriving sooner.
Watch out
PandaDoc's docs say failed webhook deliveries are not retried, so if n8n is down at signing no invoice is made, and a manual resend can create a duplicate. The fix is to use the document ID as the Stripe idempotency key, store the document ID to invoice ID pair in an n8n table, and run a daily check of completed documents against it. Check first: which PandaDoc plans include API and webhook access, how PandaDoc signs its webhooks (our sources disagree), and the exact call that lists completed documents for the daily check.
The number to watch
Track the minutes from signature to invoice sent, and the share of signed contracts invoiced within one business day.
Questions about this automation
Should the invoice send automatically or wait for my approval?
Start with a draft and a Slack approve button for your first 10 contracts, then switch to auto-send once the amounts match every time. Stripe finalizes an invoice as soon as you send it, and you cannot edit amounts on a finalized invoice.
Can it invoice a deposit now and the balance later?
Yes. Create the deposit invoice at signature and store the balance as a milestone row. A second workflow creates the final invoice when the project task is marked done.
Does this work with DocuSign instead of PandaDoc?
DocuSign can fire a Connect event when an envelope is completed, so the same flow can work. We have not verified DocuSign's webhook signing details or plan requirements, so check them before you build.
Related reading
Want contracts that invoice themselves? Book a free call and we will build it around your contract and invoicing tools.
How often to send invoice reminders (and how to automate it)
How often to send invoice reminders: 3 days before the due date, on it, then 3, 7, 14 and 30 days after. That is our recommended schedule, not a published standard. n8n runs it weekdays and Claude writes the wording.
The problem
Chasing late invoices is awkward, so it happens late, unevenly, or not at all. With 20 to 30 open invoices, nobody keeps track of who got which nudge.
How it works
Every weekday at 9:00, n8n lists your open invoices and works out how many days each one is from its due date. A rule picks the reminder stage, Claude drafts the wording, and Gmail sends it. Stages from day 14 on are saved as drafts for the owner to approve.
Where AI fits
Claude drafts the wording and tone for each stage from the invoice details. Timing, stage rules and stop conditions are plain code, so a model never decides to skip or add a reminder.
What you need
n8n, an invoice source with due dates and status (Stripe, QuickBooks or Xero), Gmail or Outlook send access, Claude, an n8n data table for the reminder log, and the billing contact for each client.
Step by step
List open invoicesAppStripe
A scheduled run asks Stripe for open invoices that are sent by email, 100 at a time, and QuickBooks can do the same job.
Work out days lateRulen8n
A code step compares each due date to today and reads the last reminder stage from an n8n data table.
Pick the stageRulen8n
A rule turns the days from the due date into one of six stages: 3 days before, the due date, then 3, 7, 14 and 30 days after.
Draft the messageClaude AIClaude
Claude writes a short reminder in the right tone from the invoice number, amount, payment link and contact name, and the day 14 and later reminders become Gmail drafts for the owner.
Send and logAppGmail
Before sending, n8n re-fetches the invoice, skips it if it is no longer open, sends the email, logs the stage, and posts escalations to Slack.
Build time
Our estimate: 4 to 6 hours. Basis: one scheduled workflow with a Switch step, a prompt per stage, a send log and a dry run on past invoices. We have not measured it.
Time saved
Our estimate: about 1 hour a week with 20 to 30 open invoices. Basis: 2 to 3 minutes per manual chase, for 20 to 30 chases a week.
Watch out
The risk is reminding a client who has already paid because the bank transfer is not matched yet. The fix is to re-fetch the invoice right before every send, skip anything not open, and keep a disputed flag that pauses the sequence for that client. Check first: Stripe's invoice list shows no due-date filter, so the build filters by due date in a code step, and you should re-check that against the API version you use.
The number to watch
Track the average days from due date to paid, and the share of invoices paid before the day 14 stage.
Questions about this automation
How often should I send invoice reminders?
A workable schedule is 3 days before the due date, on the due date, then 3, 7, 14 and 30 days after. This is our recommendation, not a published standard, so adjust it to your payment terms and stop the moment the invoice is paid or disputed.
Can Stripe, QuickBooks or Xero send reminders without n8n?
Yes, all three have built-in reminders, but the timings and limits differ, so check them in your own account. Add n8n when you want your own tone, an owner approval at day 14, Slack alerts or a per-client pause, which the built-ins mostly lack.
Should reminders send automatically or wait for approval?
Our recommended setup sends the early reminders on its own and holds day 14 and later as drafts for the owner. Run a dry run on past invoices first, so you can see who would have been reminded.
Related reading
Want overdue invoices chased on the right day without you doing it? Book a free call and we will build it around your invoicing tool.
Stage 4 of 6
Onboard the client
Collect what you need, book the kickoff and set up the channel and task board.
Client onboarding form template: how to automate it
A client onboarding form template of about ten questions can build your project brief, Drive folder and ClickUp project the moment a client submits it. Claude writes the brief and flags gaps. Plain rules do the routing.
The problem
New-client details arrive in pieces across email, DMs and calls, so the team starts the project with gaps. Someone then has to chase whatever is missing.
How it works
The client submits one form after signing, and n8n catches the answers. Claude turns them into a one-page brief and a list of gaps, then n8n builds the Drive folder, the ClickUp project and the CRM record and emails the client. It then starts the kickoff booking and Slack workflows, so this is the first link of the onboarding chain.
Where AI fits
Claude summarizes the answers into a brief and flags vague or missing items, for example 'brand assets: will send later', and anything that looks like a password. Routing the fields to each tool is plain rules.
What you need
n8n, a form tool such as Tally or Typeform, Claude, Google Drive, ClickUp or Asana, Gmail and your CRM, plus about 10 form fields, a Drive parent folder ID and a project template ID.
Step by step
Client submits formAppTally
n8n has no built-in Tally node, so a Webhook node receives the answers from Tally as soon as the client submits and replies right away.
Write the briefClaude AIClaude
Claude turns the answers into a one-page brief with a summary, the risks, and a list of missing or vague answers.
Create project spaceAppGoogle Drive
n8n creates the client folder in Google Drive and the ClickUp or Asana project from your template, and puts the brief in the first task.
Update the CRMAppHubSpot
The deal is marked onboarded in your CRM, and the billing contact and approver are saved to the record.
Confirm to clientAppGmail
Gmail sends the client a thank-you that lists any gaps and what happens next.
Start next stepsRulen8n
A plain rule passes the client record to the kickoff booking and Slack channel workflows.
Build time
Our estimate: 3 to 5 hours. Basis: one form, one webhook workflow, one Claude prompt and three app steps. We have not measured it.
Time saved
Our estimate: about 30 minutes per new client, or roughly 2 hours a month at 4 new clients. Basis: chasing answers and copying them into the folder, task and brief by hand.
Watch out
Tally expects a reply within 10 seconds, so a slow run can miss it and Tally will retry later, creating a duplicate project. Set the n8n Webhook node to respond immediately, do the work after, and skip any response ID you have already processed. Check first: confirm the response ID field name in a test payload, copy file uploads into Drive in the same run because we have not checked how long Tally upload links last, and test how well Claude flags password-like text.
The number to watch
Track the share of forms that arrive complete with no chasing, and the days from signature to a complete form.
Questions about this automation
What should a client onboarding form ask?
Keep it to about ten questions: legal and billing details, who approves work, goals and success measures, audience, deadline, brand assets, examples they like, preferred way to talk, and who to invite to Slack. Leave out anything you can learn on the kickoff call, because every extra field is another reason to put the form off.
Which form tool works best with n8n?
Tally and Typeform both send a webhook on submit, and Tally says webhooks are free for all users. n8n also has a Typeform Trigger and a built-in Form Trigger, and Google Forms can be routed through Google Sheets with the Sheets Trigger. Check Typeform's webhook plan requirements first, because we could not confirm them.
Should the form ask clients for passwords?
No. Never ask for passwords in a form. Ask the client to share logins through a password manager instead, and say so in the form.
Related reading
Want one form that sets up every new client? Book a free call and we will build it.
Kickoff call agenda: how to automate booking and prep
A kickoff call agenda built from the client's own onboarding answers lands in both inboxes 24 hours before the call, after a one-time Calendly link books it. Claude drafts the agenda. Plain rules handle booking, timing and nudges.
The problem
Booking the kickoff takes a thread of 'does Thursday work', and the call starts with no agenda and no shared idea of what the client wants.
How it works
When the onboarding form is done, the client gets a one-time Calendly link by email, and n8n nudges them if it sits unbooked. When they book, Claude drafts an agenda from their form answers. n8n holds it until 24 hours before the call, then emails it to client and team, as the second link of the onboarding chain.
Where AI fits
Claude drafts the agenda and a short list of questions from the onboarding brief. Booking, timing and nudges are plain rules, with no AI.
What you need
n8n, a Calendly account with an API token and a kickoff event type, Gmail, Claude, the onboarding brief from the form, and a ClickUp task or Slack channel to receive the agenda.
Step by step
Create one-time linkAppCalendly
n8n has no Calendly action node for this, so a custom HTTP call creates a one-time booking link for your kickoff event type.
Send the linkAppGmail
Gmail sends the client the link with a three-line prep checklist.
Nudge if unbookedRulen8n
A daily check finds clients with a link and no booking, sends one reminder after 2 days, and pings the account owner on Slack after 5.
Catch the bookingAppCalendly
n8n's Calendly Trigger fires on a new booking and matches the invitee email to the client record.
Draft the agendaClaude AIClaude
Claude builds an agenda from the brief: goals, scope, open questions, and what the team needs to hear.
Share before the callAppGmail
A Wait step holds until 24 hours before the start, then Gmail sends the agenda to client and team and it is added to the project task.
Build time
Our estimate: 2 to 4 hours. Basis: two small workflows, one API call and one prompt. We have not measured it.
Time saved
Our estimate: about 20 minutes per new client, or roughly 1 to 2 hours a month at 3 to 5 new clients. Basis: the scheduling emails and agenda writing you now do by hand.
Watch out
The client may book with a different email than the one on the form, so the booking cannot be matched to the client. Match on email first, then on name or company domain, and send unmatched bookings to a Slack review queue instead of dropping them. Check first: our sources conflict on which Calendly plan you need, so check your Calendly plan before you build, and test how long a one-time link stays valid.
The number to watch
Track the days from the onboarding form to a booked kickoff call.
Questions about this automation
What should a kickoff call agenda include?
A 45-minute starting point: intros and roles (5 minutes), goals and success measures (10), scope and what is out of scope (10), timeline and milestones (5), access and assets (5), communication and approvals (5), next steps and owners (5). Cut anything the onboarding form already answered.
Do Calendly webhooks need a paid plan?
Calendly's current API spec says the booking-created and booking-canceled events are available on all plans including Free, while some older help pages say a paid plan is needed. Check your Calendly plan before you build, and test on it first.
Who gets the agenda, and when?
Both the client and your team get it by email 24 hours before the call, and it is added to the project task. That way everyone has read the same plan before the call starts.
Related reading
Want kickoff calls booked without the email thread? Book a free call and we will build it.
Automate Slack channel creation for a new client
To automate Slack channel creation for a new client, n8n names the channel, adds your team, posts a welcome message and sends a Slack Connect invite. Claude writes only the welcome post. Pilot the client invite with one client first.
The problem
Setting up a client channel means naming it, adding the team, writing a welcome post and inviting the client, and every account gets it done a little differently. The result is different names and different welcomes for every client.
How it works
When the onboarding workflow passes a new client record, n8n builds a channel name such as client-acme and creates the channel. It adds your project team, Claude writes the welcome post, and a custom API call sends the client a Slack Connect invite. This is the third link of the onboarding chain, after the form and the kickoff booking.
Where AI fits
Light. Claude drafts the welcome message and topic from the brief. Naming, inviting and sharing are plain rules.
What you need
n8n, a paid Slack plan on your side, your own Slack app with the bot scopes channels:manage, groups:write, channels:write.invites, groups:write.invites, chat:write and conversations.connect:write, a lookup table of role to Slack user ID, the client's contact email, Claude and ClickUp.
Step by step
Build channel nameRulen8n
A code step builds a channel name that follows Slack's rules: lowercase letters, numbers, hyphens and underscores, 80 characters or fewer.
Create the channelAppSlack
The Slack node creates a public or private channel per your rule, and n8n saves the returned channel ID.
Add the teamAppSlack
The Slack node adds the project lead, designer and account manager by Slack user ID from a lookup table.
Post welcome messageClaude AIClaude
Claude writes the welcome post and topic from the brief, and the Slack node posts the message and sets the topic.
Invite the clientAppSlack
A custom HTTP call to Slack's inviteShared method sends the client a Slack Connect invite, one email per call.
Log and notifyAppClickUp
n8n saves the channel ID on the ClickUp project task and messages the account owner that the invite is pending.
Build time
Our estimate: 3 to 5 hours, plus 30 to 60 minutes to set up the Slack app and scopes. Basis: five Slack calls, one prompt and a test client. We have not measured it.
Time saved
Our estimate: about 15 minutes per new client, or roughly 1 hour a month at 4 new clients. Basis: creating, naming, inviting and posting by hand.
Watch out
The Slack Connect invite stays pending until the client accepts, and their Slack admin may need to approve it, so a first post can sit unseen. Keep the first message team-only, have the workflow chase after 2 days, and if the client side is blocked, resend the invite from Slack's own screen. Check first: we have not confirmed that a bot token can invite a brand-new outside client by email, so pilot it with one real client before you rely on it.
The number to watch
Track the minutes from the onboarding form to a ready channel, and the share of Slack Connect invites accepted within a week.
Questions about this automation
Do I need a paid Slack plan for Slack Connect?
On your side, yes: the invite method returns a not_paid error on free workspaces, and Slack says each organization needs a paid plan to work in channels together. Slack's help articles disagree on free-plan clients, so check with your client's admin. We have not tested a live free workspace.
Can a bot create the channel and invite the client without anyone clicking?
A bot token with channels:manage can create the channel and add your team without a click. Inviting a client by email may work the same way, but we have not confirmed it for a client who has no shared channel with you yet, so pilot it with one real client. Build retries for name_taken, restricted_action and ratelimited.
What if the channel name is already taken?
Slack returns a name_taken error when the name exists. Add a short suffix, such as the project name, and try again.
Related reading
Want every new client channel set up the same way? Book a free call and we will build it, and pilot the client invite with you first.
ClickUp client onboarding template: build each board automatically
A ClickUp client onboarding template can set up each new client's board for you. When the onboarding form arrives, n8n clones your folder template, ClickUp sets the dates, and the board link lands in Slack.
The problem
Every new client means someone duplicates a folder, renames it, shifts the dates and assigns tasks by hand. Each board ends up slightly different.
How it works
This is the last link in the onboarding chain: form, kickoff booking, Slack channel, then the task board. The same form submission reaches n8n, a rule picks the template for the client's package, and ClickUp copies the lists, tasks and dates. n8n waits until the copy is complete, Claude writes a short first-task brief, and the board link lands in the client's Slack channel.
Where AI fits
Light and optional. Plain rules pick the template, clone the board and check the copy. Claude only summarizes the form answers into a first-task brief and flags missing information.
What you need
n8n, a ClickUp Workspace with one folder template per service package and an API token, Slack, a client sheet or CRM, and an Anthropic API key for the optional brief.
Step by step
Onboarding form landsRulen8n
The same n8n webhook that carries the onboarding form receives the client name, package, start date and account lead.
Pick the templateRulen8n
A rule maps the package, such as SEO retainer or web build, to the right ClickUp folder template.
Clone the templateAppClickUp
ClickUp copies the lists, tasks and views and moves the dates to fit the start date.
Wait until builtRulen8n
n8n checks the new folder until its task count matches the template, because ClickUp can answer before the copy is finished.
Brief and assignClaude AIClaude
Claude turns the form answers into a short brief for the first task and lists anything missing, while a rule sets the account lead on the kickoff tasks.
Post the board linkAppSlack
n8n posts the board link in the client's Slack channel and writes the folder ID back to your client sheet or CRM.
Build time
Our estimate: 3 to 5 hours if the template already exists, plus 2 to 4 hours to tidy the template itself. Basis: one Switch node, two HTTP Request calls, one polling loop and one Slack node.
Time saved
Our estimate: about 30 minutes per new client, so 1 to 2 hours a month at 2 to 4 new clients. Basis: duplicating a folder, renaming, shifting dates and assigning by hand takes roughly 20 to 40 minutes. The bigger win is that every board is identical.
Watch out
ClickUp can answer before the folder is fully built, so a step that assigns tasks finds an empty folder. Fix: check the new lists until the task count matches the template before doing anything else. Check this first: we have not confirmed whether folder templates need a specific ClickUp plan, or whether setting the account lead as assignee works through n8n's ClickUp node.
The number to watch
Minutes from onboarding form to board live (aim for under 10), and the percentage of new clients with a board on day one.
Questions about this automation
Can I create a ClickUp client onboarding board from a template automatically?
Yes. ClickUp's public API has a Create Folder from Template endpoint, and n8n can call it with the HTTP Request node because the built-in ClickUp node has no template operation. A publicly shared template must be added to your Workspace first.
Does the same idea work in Asana or Monday?
Asana has an instantiate-project endpoint for project templates. It runs as an asynchronous job and needs date values if the template has dates. Monday has a duplicate_board mutation with a limit of 40 mutations a minute.
Can I use Zapier or Make instead of n8n?
They are options. n8n is our default because its HTTP Request node can call ClickUp's template endpoint directly. We have not checked whether Zapier or Make have a ClickUp folder template action, so plan on an HTTP or webhook step there.
Related reading
Want every new client's board built before the kickoff call? Book a free call and we will build it for your agency.
Not sure which of the 25 to build first?
Tell us how your agency works on a free call. We will point to the two or three automations that fit and say what each needs from you.
Stage 5 of 6
Deliver and watch
Report every week, catch late work and notice a client who is drifting.
Weekly client status report template: automate the first draft
A weekly client status report template can fill itself. Every Tuesday, n8n pulls the numbers and calculates the changes, and Claude writes the summary. Your account manager edits a ready draft in minutes and sends it.
The problem
Account managers lose hours every week copying numbers from analytics and ad platforms into a report. Clients skim it quickly.
How it works
Every Tuesday at 8 am, n8n reads your client sheet, pulls last week and the week before from GA4, and lists the ClickUp tasks that were finished. A rule calculates every percentage change, so Claude never does the arithmetic. Claude then writes the summary from those facts, and a draft waits in Gmail for the account manager to edit and send.
Where AI fits
Claude writes the narrative only from numbers and tasks that rules have already calculated, and says 'cause not yet known' when the data does not explain a change. Fetching, arithmetic and sending are plain rules and a person.
What you need
n8n, a client sheet, read access to each client's GA4 property, ClickUp, a Google Docs template, Gmail, a KPI target per client, and an Anthropic API key.
Step by step
Tuesday scheduleRulen8n
A weekly schedule starts the run on Tuesday at 8 am and reads one row per client from your Google Sheet.
Pull channel numbersAppGoogle Analytics 4
n8n pulls last week and the week before from GA4, while ad spend and clicks from Google Ads or Meta need a custom API call, so check that first.
Pull work doneAppClickUp
n8n lists the ClickUp tasks finished last week and the tasks due next week.
Calculate the changesRulen8n
A rule works out every percentage change and target gap, so the AI never does arithmetic.
Draft the summaryClaude AIClaude
Claude receives only those facts and writes a headline, what changed, next week's plan and one ask for the client.
Draft for reviewA personGmail
n8n fills a Google Docs template and saves a Gmail draft, and the account manager edits and sends it.
Build time
Our estimate: 8 to 14 hours for the first client including the template and the prompt, then about 30 minutes per extra client. Basis: three data sources, one calculation step, one prompt and one document step.
Time saved
Our estimate: about 3 to 5 hours a week at 10 clients. Basis: 30 minutes per client to pull, copy and write by hand, against about 5 minutes to review a draft.
Watch out
GA4 data can take 24 to 48 hours to finish processing, so a Monday run reports an unfinished Sunday and the numbers change later. Fix: run on Tuesday, print the date range in the report, and store the numbers you reported so next week compares against the same figures. Check this first: Meta Ads and Google Ads metrics need a custom call, and we have not confirmed how the Google Docs node replaces placeholder text.
The number to watch
Reports delivered on schedule (aim for 100 percent) and edits per report, meaning how many sentences the account manager changes.
Questions about this automation
What should a weekly client status report include?
Keep it to one page: a one-sentence headline, three to five numbers against last week and target, work completed, work planned for next week, and one thing you need from the client. Clients read the top and skim the rest.
Can Looker Studio send the weekly report by itself?
Looker Studio can send scheduled emails of a dashboard, but it cannot write the commentary. Use n8n and Claude for the narrative, and link to the live dashboard.
Can Zapier or Make run this instead of n8n?
They are options, and n8n is our default. Supermetrics can supply ad data if you do not want a custom call. Looker Studio can still handle the charts.
Related reading
Want Tuesday's report drafted before your team opens Slack? Book a free call and we will build it for your agency.
Overdue task notification: tell your team before the client asks
An overdue task notification for ClickUp needs a scheduled check, because ClickUp has no overdue webhook. Every weekday at 9 am, n8n finds late tasks and alerts the assignee, then the account lead, then the owner.
The problem
Tasks slip past their due date quietly. The owner finds out when the client asks where the work is.
How it works
Every weekday at 9 am, n8n asks ClickUp for every open task past its due date, drops tasks that wait on the client, and groups the rest by person and client. A log sheet makes sure each task is alerted once per stage: the assignee on day 1, the account lead on day 3 and the owner on day 5. Slack sends one grouped message per person, and for late client-facing tasks Claude drafts a short heads-up for the account manager to confirm.
Where AI fits
None for finding late tasks: plain rules do that. Claude is optional and only drafts the client heads-up, which a person approves.
What you need
n8n, ClickUp with an API credential, Slack with a user ID for each assignee, a Google Sheet for the alert log, and an Anthropic API key if you want the optional draft.
Step by step
Weekday 9am checkRulen8n
A weekday 9 am schedule starts the check, because ClickUp has no overdue webhook event and n8n has to look every morning.
Fetch overdue tasksAppClickUp
n8n asks ClickUp for every task with a due date before now, including subtasks, 100 tasks at a time.
Filter and groupRulen8n
A rule drops tasks in waiting-on-client statuses, works out days late, and groups tasks by person and client.
Check alert historyRuleGoogle Sheets
n8n reads a log sheet so each task is alerted once per stage, then adds the new alerts.
Send the nudgeAppSlack
Slack sends one grouped message per person with task names, client and days late, not one message per task.
Draft client noteClaude AIClaude
For client-facing tasks late three days or more, Claude drafts a three-sentence heads-up with a new date, and the account manager confirms it.
Build time
Our estimate: 3 to 5 hours, plus about 1 hour for the optional Claude draft. Basis: one paginated API call, one grouping step, one log lookup and one Slack message.
Time saved
Our estimate: about 1 to 2 hours a week for a project manager. Basis: 15 to 20 minutes a day scanning boards for slipped tasks, five days a week. The main gain is catching slips the same day.
Watch out
Tasks that wait on the client look late and teach the team to ignore alerts. Fix: exclude waiting-on-client statuses or a tag, alert once per stage instead of daily, and run a separate weekly list of open tasks with no due date, because a due date filter never matches those. Check this first: we have not confirmed that one ClickUp call with no list filter returns every task in your Workspace, so filter by list if in doubt.
The number to watch
Overdue tasks at 9 am each Monday, plus on-time delivery rate: tasks finished on or before their due date.
Questions about this automation
Does ClickUp have a webhook for overdue tasks?
No. ClickUp's webhook events include a due date change, but its docs list no overdue event. A scheduled check is the way to find tasks that have passed their due date.
How often should an overdue task check run?
Once each weekday morning is enough for most agencies. ClickUp's docs allow 100 requests a minute per token on the Free, Unlimited and Business plans, and a daily run with paging uses a small part of that.
Do I need AI for overdue task alerts?
No. Plain rules find, filter and group the late tasks. Claude is optional, and it only drafts the client heads-up for a person to approve.
Related reading
Want late tasks to reach your team at 9 am and not your client's inbox? Book a free call and we will build it.
Client health scorecard: score every client out of 100 each week
A client health scorecard scores every active client out of 100 each Friday from six signals. Plain rules do the math, Claude only reads email tone, and Slack warns you when a client turns red.
The problem
Unhappy clients rarely complain first. They go quiet, pay late, and then cancel, and you see it only at the end.
How it works
Every Friday at 3 pm, n8n loops over your active clients and reads the signals from your tools: payment, replies, late tasks, hours used and results against goal. Rules turn each signal into points out of 100, Claude reads the tone of the last five client emails, and n8n adds a row to your scorecard sheet. If a client is red, or falls 15 points in two weeks, Slack tells you the score, the weakest signals and a suggested check-in.
Where AI fits
Claude only reads the tone of recent client emails and writes a one-line reason, worth 10 of the 100 points. Every other signal, and the score itself, comes from plain rules.
What you need
n8n, a client sheet with the ID for each tool, ClickUp, a time tracker such as Harvest, QuickBooks or Stripe, Gmail, GA4, Google Sheets, Slack, and an Anthropic API key.
Step by step
Friday scheduleRulen8n
A Friday 3 pm schedule starts the run and loops over active clients from your client sheet.
Collect six signalsRulen8n
n8n reads overdue invoices, late tasks, hours used, days since the last client reply and the main KPI against target.
Score each signalRulen8n
A rule scores each signal full, half or zero points using the weights, then adds them into a score out of 100.
Read message toneClaude AIClaude
Claude reads the last five client emails as data and returns positive, neutral or negative with one short reason.
Log the scorecardAppGoogle Sheets
n8n adds this week's score and signal breakdown to a Google Sheet so you can see the trend.
Alert on redAppSlack
If a client is red or fell 15 points in two weeks, Slack sends you the client, score, weakest signals and a suggested next step.
Build time
Our estimate: 10 to 16 hours. Basis: five data connections at 1 to 2 hours each, plus the scoring step, the tone prompt and the alert.
Time saved
Our estimate: about 2 hours a month of owner review at 12 clients. Basis: roughly 10 minutes per client to check five tools by hand. The real value is a weekly early warning instead of monthly or never.
Watch out
Missing data scores as zero, so a healthy client with no invoice this month shows red and the alerts lose trust. Fix: score no data as half points, mark the signal n/a, and only alert when at least four of six signals have data. Check this first: the weights and bands are our suggested starting design, not an industry standard, and we have not confirmed Gmail filters, Stripe overdue invoice lookups or Harvest budget fields in n8n.
The number to watch
Number of red clients each week, and how many return to green within 60 days after a check-in. Also watch clients who left while green, because that means a signal is missing.
Questions about this automation
What signals should a client health scorecard include?
A simple version uses six signals with weights out of 100: payment 20, responsiveness 20, on-time delivery 20, results vs goal 15, budget burn 15 and message tone 10. Score green at 80 and up, amber at 55 to 79, and red under 55.
How do I know my scorecard weights are right?
These weights are a starting point, not a published standard. After 90 days, compare scores with the clients who complained or left, then move weight to the signals that warned you first.
Can Zapier or Make run the scorecard instead of n8n?
They are options, and n8n is our default. HubSpot or Pipedrive can supply the last contact date, and Toggl can replace Harvest through an API call.
Related reading
Want a weekly warning before a client goes quiet? Book a free call and we will build your scorecard.
Stage 6 of 6
Keep and grow
Renew retainers, offer more, ask for reviews, case studies and referrals, and check your profit.
Retainer renewal reminder: start the renewal talk 60 days out
A retainer renewal reminder starts the renewal conversation weeks early. n8n checks contract dates daily: 60 days out brings internal prep, 30 days a draft client email, and 14 days an owner alert. A person sends every email.
The problem
Retainers end because nobody raised the renewal in time, not because the client wanted to leave. The end date arrives and the first conversation happens too late.
How it works
Every day at 8 am, n8n reads your contract sheet and counts the days left on each active retainer. At 60 days the account lead gets an internal brief, at 30 days a draft renewal email with a results recap waits in Gmail, and at 14 days with no signed renewal the owner is alerted. A person reviews and sends every message.
Where AI fits
Claude drafts the internal brief and the client email only from facts you supply, and it must not invent results or pricing. Date checks, stages and reminders are plain rules, and a person sends every message.
What you need
n8n, a contract sheet with end date, notice period and auto-renew, your weekly report and health sheets, Gmail, Slack, ClickUp or your CRM, and an Anthropic API key.
Step by step
Daily date checkRulen8n
A daily 8 am schedule reads your contract sheet and works out the days left on every active retainer.
Match renewal stageRulen8n
A rule matches each retainer to a stage, 60, 30 or 14 days, using days left at or below the stage and a stage column so no reminder repeats.
Gather resultsAppGoogle Sheets
n8n pulls the client's recent weekly report summaries and health scores from the sheets your weekly report and health check fill.
Draft renewal noteClaude AIClaude
Claude writes the internal brief and a draft client email from those facts: what was delivered, what moved and the proposed next term.
Create drafts and tasksA personGmail
n8n saves the email as a Gmail draft and creates a task for the account lead, but never sends it, so a person reviews and sends it.
Ping the ownerAppSlack
Slack tells the account lead, and the owner at 14 days, with a link to the draft, and n8n updates the stage column.
Build time
Our estimate: 4 to 6 hours, faster if the weekly report and health scorecard flows already exist. Basis: one sheet read, one stage rule, one prompt, one draft step and one Slack step.
Time saved
Our estimate: about 1 hour per renewal, so 1 to 3 hours a month at 1 to 3 renewals. Basis: around 30 minutes to assemble a results recap plus 30 minutes to draft the email.
Watch out
The real deadline is often the notice date, not the end date, so a contract that renews unless cancelled 30 days before the end needs reminders counted from that notice date. Fix: store both dates, key the timeline off the notice date when auto-renew is on, and use days left at or below the stage so a day when n8n was down does not skip a reminder. Check this first: notice rules vary by contract and need your own contract or lawyer, and we have not checked custom end-date fields in HubSpot or Pipedrive.
The number to watch
Renewal rate (retainers renewed divided by retainers due) and the percentage of renewal conversations started at least 30 days before the end.
Questions about this automation
How many days before a retainer ends should I send a renewal reminder?
A workable timeline is 60 days for internal prep, 30 days for the client proposal and 14 days for the owner follow-up. Check your contract's notice period first, because the notice date may come before the end date.
Can Stripe send retainer renewal reminders for me?
Not well. Stripe's invoice.upcoming event is sent a few days before renewal, set by the Upcoming renewal events setting in the Dashboard, and only for subscriptions charged automatically. That is too late for a renewal conversation, so drive reminders from the contract end date and use Stripe for payment events.
Does Claude send the renewal email?
No. Claude drafts the email from facts you supply, and n8n saves it as a Gmail draft. A person reviews and sends every message.
Related reading
Want every renewal raised 60 days early? Book a free call and we will build it for your agency.
Upsell email to existing clients: draft it when the moment is right
An upsell email to existing clients works best when a result has just landed and every invoice is paid. This automation checks each active client every morning, drafts one offer with Claude, and sends it only after you approve.
The problem
You only think about selling more to a client when you are quiet or short of cash. The best moments to offer a next step pass without anyone noticing.
How it works
Every morning n8n checks each active client against simple rules: budget nearly used, no unpaid invoice, no offer in the last 90 days. When a client passes, Claude drafts one short offer. You read it in Gmail and press Approve, and then it goes out and a deal is created in your CRM.
Where AI fits
Claude writes the email from the client's real results and picks one offer from your fixed menu. Plain rules decide who gets an offer and when, and a person approves every email before it is sent.
What you need
You need n8n, a Claude API key, Harvest for project budgets, Stripe for invoice status, Gmail, HubSpot or Pipedrive, and a Google Sheet that holds your offer menu and a log of offers sent.
Step by step
Daily signal checkRulen8n + Harvest
A daily n8n schedule reads each active project's budget in Harvest and the open invoices in Stripe. This is a plain rule step with no AI.
Apply the rulesRulen8n
Plain rules keep a client only if a signal fired, no invoice is open or overdue, and no offer went out in the last 90 days.
Draft the offerClaude AIClaude
Claude reads the client's real results and your fixed menu of offers, picks one offer, and writes a short plain email.
You approveA personGmail
You get the draft in Gmail with Approve and Decline buttons. Nothing is sent until you click.
Send and trackAppGmail + HubSpot
On Approve, Gmail sends the email, your CRM gets a deal tagged as an upsell, and the date goes in your Sheet so the 90 day cooldown works.
Build time
Our estimate: 6 to 10 hours. Basis: five integrations, one Claude prompt, one approval loop and a mapping sheet that links each client across Harvest, Stripe and your CRM. It is faster if you have built n8n workflows before.
Time saved
Our estimate: about 1 to 2 hours a week. Basis: checking budgets and invoices by hand for 8 to 15 active clients at about 10 minutes each, plus drafting the emails.
Watch out
An offer sent to a client who is behind on payment or unhappy burns trust, so the rules step checks for open or overdue invoices, your own do-not-pitch flag and a 90 day cooldown read from your Sheet. Check this first: Harvest only reports budget remaining on projects that have a budget set, money budgets are visible only to admins and project managers with rate permission, and n8n's Stripe node has no invoice operation, so invoices are read with a custom HTTP Request call that we have not tested.
The number to watch
Track upsell acceptance rate (offers accepted divided by offers sent) and expansion revenue per month from upsell deals.
Questions about this automation
When should I send an upsell email to an existing client?
Send it right after a delivered result or when the current budget is nearly used, because that is when the next step is obvious to the client. Do not send it while an invoice is overdue or a complaint is open.
Does the AI decide which clients get an upsell email?
No. Plain rules in n8n decide who gets an offer and when, because that part must be predictable. Claude only writes the email and picks one offer from your menu, and you approve it.
Can I use Zapier or Make instead of n8n?
They are options, but we have not checked their connector coverage for each app in this flow. Check that before you commit to either one.
Related reading
Want upsell offers that go out at the right moment and only after you approve them? Book a free call and we will build it for your agency.
Google review request automation that asks every client the same way
Google review request automation sends every client the same short request 7 days after a project ends. It never asks only happy clients and never offers a reward, which keeps your agency inside Google's rules.
The problem
Happy clients rarely leave reviews unless asked. Asking by hand gets forgotten, feels awkward, and is easy to do in ways that break Google's rules.
How it works
When a project is marked complete in ClickUp or Asana, n8n waits 7 days and checks plain rules. Claude then writes a short personal email with your review link, and Gmail sends it and a Sheet logs it. When a review appears on your Business Profile, Slack tells you, Claude drafts a reply, and you approve it.
Where AI fits
Claude drafts the personal request email and a reply to each new review. A person approves every reply, and plain rules choose who is asked and when, so no step looks at sentiment.
What you need
You need n8n, a Claude API key, ClickUp or Asana, Gmail, a Google Sheet, Slack, your Google review link, and Google Business Profile API access if you want the new review alert.
Step by step
Milestone reachedRuleClickUp + n8n
A completed project in ClickUp or Asana starts the flow, and n8n waits 7 days so the client has seen the results.
Check the rulesRulen8n
Plain rules check that the client has not been asked in the last 12 months, has a valid email and has not asked to be left alone, then pick the platform by a fixed rule, never by how the client feels.
Write the requestClaude AIClaude
Claude writes a short personal email that names the project and includes your review link. It is told not to mention stars, suggest what to write, or offer anything in return.
Send and logAppGmail + Sheets
Gmail sends the same request with the same link to every client, and a Google Sheet logs the date, client and platform.
Watch new reviewsA personGoogle Business Profile
A person replies to reviews. When a new review appears on your Google Business Profile, Slack tells you, Claude drafts a reply, and you approve it before it is posted.
Build time
Our estimate: 3 to 5 hours for the request flow. Basis: three nodes and one prompt. Add the wait for Google Business Profile API approval if you want review alerts, because the docs do not say how long approval takes.
Time saved
Our estimate: about 1 to 1.5 hours a month. Basis: 4 to 6 completed projects a month at 10 to 15 minutes each to remember, write and log a request. The bigger gain is that every request actually gets sent.
Watch out
Review gating breaks Google's rules: do not ask how a client felt first and send the public link only to happy ones. The fix is to send the same request with the same links to every client at the same milestone, spread over days. Check this first: Google Business Profile API access needs an application and shows a quota of 0 until approved, and we have not confirmed whether an agency can hold a G2 listing, so only use a G2 link if you have one.
The number to watch
Track reviews posted divided by requests sent, per platform, and new reviews per quarter.
Questions about this automation
Can I offer a discount or freebie for a Google review?
No. Google's policy bans any incentive such as payment, discounts, or free goods or services in exchange for a review, and says such reviews will be removed. The safest rule is no incentives on any platform.
Can I ask only my happiest clients for reviews?
No. Google's policy says merchants must not selectively solicit positive reviews or discourage negative ones, and the FTC's guide says not to ask only customers you think will be positive. Ask every client at the same milestone and reply to all reviews, good or bad.
Does Claude read how the client feels before the request goes out?
No. Plain rules choose who is asked and when, and no step looks at sentiment. Claude only writes the email, and a person approves replies to reviews.
Related reading
Want review requests that go to every client the same way, with a person approving each reply? Book a free call and we will build it.
Case study request email template, sent while the win is fresh
A case study request email works best right after a win. This automation collects your real results, drafts the request and a fact-only outline with Claude, and sends it after you approve, with one follow-up.
The problem
You have strong client results sitting in your delivery tools. Nobody asks the client for a case study while the win is fresh, so you never publish one.
How it works
When your account lead flags a project as a win, n8n gathers the scope, the before and after numbers and the client contact. Claude drafts a short request and an outline that uses only those facts. You approve the email, it is sent, and one follow-up goes out 5 days later if there is no reply.
Where AI fits
Claude drafts the request email and a fact-only outline. It never creates numbers, a plain Code step rejects any draft that has a number not found in the source data, and a person approves the email.
What you need
You need n8n, a Claude API key, ClickUp, Gmail, a Google Sheet with your before and after results, and HubSpot or Pipedrive for the client contact.
Step by step
Win flaggedRuleClickUp + n8n
A win flag in ClickUp, set by your account lead, or a result that passes a threshold in your Sheet starts the flow.
Collect the factsRulen8n
Plain rules read the project scope, the before and after numbers and the client contact, and tie every number to its source cell.
Draft request and outlineClaude AIClaude
Claude drafts a short request that names the real result and asks for a 15 minute interview, plus an outline that uses only the supplied facts and marks any gap as CONFIRM.
You approveA personGmail
You approve the email in Gmail before anything goes out.
Send and follow upAppGmail + CRM
Gmail sends the request, waits 5 days, sends one polite follow-up if there is no reply, and your Sheet and CRM log the outcome and any approval.
Build time
Our estimate: 4 to 6 hours. Basis: one trigger, one data gather, one Claude prompt with a number check, one approval loop and a follow-up wait.
Time saved
Our estimate: about 2 to 3 hours per case study. Basis: finding the facts, writing the request and writing a first outline by hand takes an afternoon, and this reduces it to a review.
Watch out
Publishing a client's name, logo or numbers without written approval, or letting the AI invent a number, is the main risk. The fix is an approval step in the request, no publishing until an approved cell is set in your Sheet, and a Code step that blocks any draft with numbers not in the source data. Check this first: consent rules for publishing names, logos and numbers vary by contract and country, so check your contracts, and we did not re-check the ClickUp or Asana triggers and custom fields.
The number to watch
Track case studies approved divided by requests sent, and days from win to approved case study.
Questions about this automation
What should a case study request email say?
Name the specific result, ask for one small thing such as a 15 minute call or three written answers, say how you will use it, and promise the client will approve the final text. Keep it under 100 words.
Can the AI make up numbers in a case study?
It is told to use only the facts it is given and to mark gaps as CONFIRM. A plain Code step also blocks any draft that has a number not found in your source data.
Do I need the client's permission to publish a case study?
Yes, get written approval before you publish a name, logo or number. In this flow nothing is published until an approved cell is set in your Sheet, and you should check your contracts because the rules vary by country.
Related reading
Want a case study request that goes out the week of every win, with every number checked? Book a free call and we will build it.
How to ask for referrals from clients at the right moment
To ask for referrals from clients, ask right after a documented win, when invoices are paid and no complaint is open. This automation finds that moment, drafts the ask with Claude, and sends it after you approve.
The problem
Referrals are your cheapest new business, but you ask at random. You ask too late or never, and not at the moment a client is happiest.
How it works
Every day n8n scores each client from simple signals and fires only when the score is high, a win was logged in the last 14 days, and no ask went out in 180 days. Claude drafts a short note with a blurb the client can forward. You approve it, and n8n sends it and tracks any introduction in your CRM.
Where AI fits
Claude writes the personal ask and the forwardable blurb. Health scoring and timing are plain rules, because a score you can explain is better than one a model made up, and a person approves every ask.
What you need
You need n8n, a Claude API key, Stripe, Harvest, Google Sheets for the wins log and ask log, Gmail, Slack, and HubSpot or Pipedrive with a lead source field that includes Referral.
Step by step
Score client healthRulen8n + Stripe + Harvest
A daily n8n run adds up points from rules you set: invoices paid on time in Stripe, hours within budget in Harvest, tasks delivered on time in ClickUp, and a win in your Sheet.
Check the momentRulen8n
A plain rule check passes a client only with a high score, a win in the last 14 days, no open or overdue invoice, no open complaint and no ask in 180 days, and it caps the run at 3 asks a week.
Draft the askClaude AIClaude
Claude writes a short personal note that names the win, asks for one kind of introduction, and adds a three line blurb the client can forward.
You approve and sendA personGmail
You read the draft in Gmail and approve it. On Approve, n8n sends it and logs the date.
Track and thankAppHubSpot or Pipedrive
When a new deal in HubSpot or Pipedrive has the source Referral, n8n posts to Slack so you can thank the client within a day.
Build time
Our estimate: 5 to 8 hours. Basis: the health score rules and the mapping of each client across systems are most of the work.
Time saved
Our estimate: about 30 to 60 minutes a week. Basis: scanning 10 to 20 clients by hand to decide who to ask. The real gain is asks that would otherwise never be sent.
Watch out
Asking at the wrong time, such as with an overdue invoice or an open complaint, or asking everyone in one burst, wastes goodwill, so step 2 enforces no open invoice, a 180 day cooldown and a cap of 3 asks a week. Do not copy this selection logic into review requests, because choosing happy clients is fine for referrals but not allowed for reviews. Check this first: we did not check HubSpot's built-in health score, so the score is calculated in n8n, and we did not check Zapier or Make connector coverage.
The number to watch
Track referral conversion: asks sent, referred leads received, referred deals won, and the share of new business sourced from referrals in your CRM.
Questions about this automation
When is the best time to ask a client for a referral?
Right after a documented win or delivered result, when invoices are current and no issue is open. Ask for one specific introduction and write the forwardable note for them so it takes a minute.
Should I pay clients a referral fee?
You do not need to, and a personal thank-you is often enough. If you do pay, check the tax, contract and disclosure rules where you operate first, because they vary.
Does the AI send the referral email by itself?
No. Claude drafts the note and you approve it before n8n sends it. Plain rules decide who is asked and when.
Related reading
Want referral asks that go out at the right moment, only after you approve each one? Book a free call and we will build it.
Marketing agency profit margin: see profit per client every month
Marketing agency profit margin is easiest to improve when you see it per client. This automation works out revenue minus time cost minus tool costs each month, using plain code, and Claude writes the summary.
The problem
You know revenue per client but not profit. So you keep serving clients who cost you more in time than they pay, and you find out months late.
How it works
On the third business day of each month n8n pulls paid invoices, hours per client and tool costs. Plain code works out profit and margin per client and flags anyone under your target. Claude writes a plain English summary from those finished numbers, and it lands in Slack and a Looker Studio page.
Where AI fits
Claude only writes the plain English summary from numbers n8n has already calculated. All the arithmetic is done in code, never by the model, and the summary is checked against the data.
What you need
You need n8n, a Claude API key, a billing tool such as Stripe, a time tracker such as Harvest, a Google Sheet for tool costs and your client mapping, Looker Studio, and Slack.
Step by step
Pull revenueAppBilling tool
n8n reads last month's paid invoices per client from Stripe, QuickBooks or Xero.
Pull time costAppTime tracker
n8n reads hours per person per client from your time tracker, then multiplies by each person's loaded cost rate.
Add tool costsAppGoogle Sheets
A Google Sheet holds your tool and contractor costs, and a fixed rule such as share of hours assigns them to clients.
Calculate profitRulen8n
Plain code does all the arithmetic: profit and margin per client and in total, compared with your target and the last 3 months.
Write the summaryClaude AIClaude
Claude receives the finished numbers and writes the top three problems and three suggested actions. A Code step checks that every number in the text exists in the data.
Deliver the reportAppSheets + Slack
n8n adds the rows to a Google Sheet that feeds a Looker Studio page, and posts the summary to Slack.
Build time
Our estimate: 10 to 16 hours. Basis: three data pulls with HTTP Request nodes, the client mapping sheet and the number check make this the heaviest of the five.
Time saved
Our estimate: about 3 to 5 hours a month. Basis: exporting from three systems and matching clients in a spreadsheet for 10 to 20 clients. The bigger gain is seeing profit monthly instead of yearly.
Watch out
Time that was never logged makes a client look more profitable than they are, so send a weekly timesheet reminder and flag anyone whose logged hours are well under their weekly capacity. Check this first: n8n has no Toggl action node and reads Xero and QuickBooks reports through the HTTP Request node, so each needs a custom HTTP call that we have not tested, we did not re-verify that Looker Studio reads from Google Sheets, and on n8n Cloud Starter an execution is capped at 5 minutes, so split large pulls or self-host.
The number to watch
Track profit margin per client, flag any client under your target, and track the share of clients below target month over month.
Questions about this automation
What is a good profit margin for a marketing agency?
Promethean Research, in a third-party summary of a survey, reports that the average digital agency earned a 13% net margin after tax in 2025 and about 15% on average since 2015. Agencies under 10 people averaged 19% and those with 10 to 24 people averaged 12%. Use these as a reference, not a rule.
How do I calculate profit per client?
Take the month's revenue, subtract hours worked times each person's loaded cost rate, and subtract the tool and contractor costs you can assign to that client. Margin is profit divided by revenue.
Does the AI do the maths in the report?
No. Plain rules in n8n compute every number. Claude only writes the summary from the finished numbers, and a Code step checks that each number in the text is in the data.
Related reading
Want to see profit per client on the third business day of every month? Book a free call and we will build the report for your agency.
Where to start: our view
This is our opinion, not a rule. If you only build five, we would build these. Each one fixes a gap that costs a small agency money or time.
Instant lead reply
A lead that waits a weekend may have asked another agency, so a fast first reply is the cheapest win.
Payment reminders
Chasing late invoices by hand is easy to put off, and plain rules do it on the right day every time.
Client onboarding form
One onboarding form feeds the kickoff, the channel and the task board, so it makes three other automations easier.
Weekly client report
A weekly client report writes itself from numbers you already track, and a person checks it before it goes out.
Renewal reminders
Retainers are your steadiest income, and a reminder before the end date means no renewal is missed.
How much of this needs AI?
Claude AI is good at reading and writing text. It can summarise a call, read a lead's message, write a first draft of a proposal or turn a client's answers into a project brief. It is a poor choice for counting, dates and money, because those need to be exactly right every time.
Plain rules do that work better. They add up the hours, check the date, wait three days and send the message. Contract send needs no AI at all, and invoice on signature needs almost none. For a few others, such as booking link follow-up, meeting reminders, task board setup and missed deadline alerts, the AI step is optional. A person approves what matters, and prices always come from your rate card, never from AI. When you start, we suggest saving client emails as drafts first, so someone reads each one before it goes out.
Which tools do you need?
Our default is n8n, a workflow tool that connects your apps and runs the rules. Zapier and Make work too, and each guide lists them as options. Where a tool has no built-in connection, the guide uses a custom HTTP call, which n8n, Zapier and Make all support. Features and limits depend on your plan, so test each automation on your own plan before you rely on it.
Questions about agency automation
What is agency automation?
Agency automation means automations an agency runs on itself, for its own leads, clients and billing. Examples are a reply that goes out when a lead fills in your form, or a reminder before a retainer ends. It is not the same as hiring an automation agency to build things for others.
What automations should a marketing agency have?
Start with the steps every client goes through: lead reply, proposal, contract, invoice, onboarding, weekly report and renewal. This guide covers 25 of them in that order. You do not need all 25, so pick the ones where work is slow or gets forgotten.
How do I automate a marketing agency?
Write down the steps one client goes through, from first enquiry to renewal. Pick the step where work waits or gets forgotten, then build one automation for it. Each guide here gives the trigger, the steps and the result, so you can see what is involved before you start.
How much does agency automation cost?
It depends on the build, the tools you use and how many automations you want. We quote after a free call, once we know your process. We do not give a number before that, because any number would be a guess.
Can AI run my agency?
No. AI can read text and draft a first version, but it should not decide what you charge, what you promise a client or who you work with. In these guides, plain rules handle the counting and timing, and a person approves what matters. AI is one helper in a process you still run.
Do I need to code to automate my agency?
No. Most of these are built by connecting apps you already use in a tool like n8n, Zapier or Make. A few steps use a short custom HTTP call, which the guide spells out. If you would rather not build it yourself, you can book a free call.
How do I set up client onboarding automation for an agency?
Start with a client onboarding form. When the client sends it, the automation can write a one-page brief, create the project folder, book the kickoff call, open the Slack channel and set up the task board. Automations 13 to 16 in this guide cover each step.
Is it safe to let AI write to my clients?
It is safe when the AI only drafts and a person decides what goes out. We suggest saving client emails as drafts when you start, then sending automatically only for simple notices once you trust them. Money amounts and dates always come from plain rules, not from AI.
Want these running in your agency?
Book a free call. We look at your process, pick the automations that fit and tell you what the build involves. There is no price on this page, because it depends on what you need.
Book your free call
Tell us a little about your agency. After you send it, you pick a time on the calendar.
Build times and time saved are estimates, tool plans and features change, so test every automation on your own plan before you rely on it.
