An agency can automate client onboarding with AI in three linked steps: generate a short, tailored questionnaire from the signed proposal; collect account access through each platform's own sharing features, never by passing passwords around; and have the AI draft the kick-off agenda and pack from the answers for a person to check and send.
The common mistake is to start with the kick-off deck. The step that holds up every project is access: a campaign that cannot start because a login arrives on day nine costs more than any slide. So start with the questionnaire and the access steps, where AI saves most, and treat the kick-off pack as the output those steps feed.
One flow from signed proposal to kick-off call
Here is the shape, using an illustrative agency of eight people onboarding a client that runs a pest-control firm. The target is a kick-off call on day five after signature.
| Day | Step | Who or what does it |
|---|---|---|
| 0 | Proposal signed; account manager adds the client to the tracker | Person |
| 0 | AI drafts the tailored questionnaire from the proposal; manager checks it | AI, then person |
| 0 | Questionnaire and access checklist sent with a welcome note | Automation |
| 2 | Reminder for anything outstanding | Automation |
| 3 | AI summarises answers, flags gaps and contradictions | AI, then person |
| 4 | AI drafts the kick-off agenda and one-page brief; lead edits | AI, then person |
| 5 | Kick-off call; notes drafted, actions assigned | Person, AI for notes |
The tooling can be modest: a form tool, a spreadsheet or Notion database as the tracker, an email tool or automation platform to send and remind, and a chat assistant for drafting. For how to wire forms, contracts and emails together without code, automate client onboarding with AI covers the general mechanics; the steps below are the agency-specific ones.
The questionnaire: let the proposal shorten it
A generic 60-question form tells a client you did not read their proposal. A better approach is a core of questions every client answers plus a tailored set generated from what they bought. Give the assistant the scope section of the proposal, with names removed, and your standard question bank, and ask for a form that skips anything already answered.
I run a small marketing agency. Below is the scope of work from a signed proposal and my standard onboarding question bank.
Create a client onboarding questionnaire of at most 18 questions that takes a client about 20 minutes.
Rules:
- Group the questions under 4 headings: Business basics, Audience and offer, Brand and approvals, Access and tools.
- Skip questions already answered in the scope.
- Add 3 to 5 questions specific to the services in the scope (e.g. paid social, email, website).
- Plain, friendly wording, no jargon. Mark each question as required or optional.
- Do not ask for passwords, security codes or card details anywhere.
SCOPE: [paste]
QUESTION BANK: [paste]The last rule is non-negotiable and worth repeating in the form itself.
An illustrative result for the pest-control client, whose scope is paid social and a landing page, includes questions such as "Which services earn you the most per job, and which do you want more of?", "Do you serve any areas you would prefer not to advertise to?", "Who approves ads and how quickly can they respond?" and "Which tools do you use to take bookings (phone, form, booking software)?" The last one is a tailored question, since it affects how the agency will track conversions. The manager's review changes one thing: the AI asked "What is your annual revenue?", which the client may not wish to share this early. It becomes optional, with the note "a range is fine".
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowAccess: grant it through the platform, not through a message
Access is where agencies create their biggest security exposure. A shared password in an email, in a chat thread and in a spreadsheet is three places for it to leak and no record of who used it. Platforms provide proper alternatives, and the checklist should name them:
- Google Ads: the agency sends a link request from its manager account using the client's customer ID, and someone with admin access to the client's account accepts it, either from the email invitation or under Admin, Access and security, Managers. Google's help pages describe the steps, and nobody shares a login.
- Meta ads and pages: someone with full control of the client's business portfolio adds the agency as a partner in Meta Business Suite settings, using the agency's business portfolio ID, then picks the assets to share (such as the Page and ad account) and full or partial access. Meta's business help pages explain it, and staff keep their own logins. Put your portfolio ID in the welcome note so the client is not left hunting for it.
- Google Analytics: the client adds named agency users to the account or property with a role such as Viewer, Analyst or Editor.
- Website and email platforms: a named user account for each person at the agency, with the lowest role that does the job.
- Anything without partner features: a password manager's sharing function, with the client sharing a specific item, not a spreadsheet or a message.
Platform menus change, so the checklist should link to the vendor's own help page for each system and tell the client what to look for, not rely on screenshots from a year ago. Ask the AI to draft the client-facing instructions from the help page text you paste in, in plain steps, and have a colleague follow them on a test account before they go out.
A filled-in access checklist
For the pest-control client, the AI turns the signed scope into this checklist. Illustrative:
| System | Level the agency needs | How the client grants it | Status |
|---|---|---|---|
| Meta business portfolio | Partner access to ad account and page | Client adds the agency as a partner using its portfolio ID | Waiting |
| Google Analytics | Analyst role for two named agency users | Client adds the users | Done |
| Website CMS | Editor role for one named user | Client invites by email | Waiting |
| Booking software | Read-only reports | Client creates a report user | Not started |
| Brand assets | Logo files, fonts, photos | Shared folder with the client's own share link | Done |
Each status moves to Done only after an agency person has logged in with their own credentials and confirmed they can do what the project needs. That final check is easy to skip and it prevents a surprise on launch day. If the access request has been outstanding for two days, the automation sends a short reminder naming only the open items.
What stays out of every AI tool
Write a short rule and put it at the top of the access checklist. First, no passwords, one-time codes, recovery codes or card numbers go into a questionnaire, a chat assistant, an email or a note. Second, the AI sees the checklist of statuses, not credentials. Third, anything the client typed into the form that looks like a secret is deleted from the answers before summarising, and the client is told so politely. Illustrative note for that: "Thanks for the details. For your security we don't keep logins in forms, so we've removed that line and sent you the safe way to share access instead."
It is also worth checking what your contracts say about AI use with client material. AI clauses for agency contracts covers the wording, and if you have clients with confidentiality terms, check that your AI tools and settings match what you have promised.
From answers to a kick-off pack
Once the form is back, paste the answers (with any personal or secret details removed) into the assistant and ask for three outputs: a summary of the answers in a paragraph, a list of gaps and contradictions, and a draft kick-off agenda. A prompt that works:
Below are a client's onboarding answers and the scope of work.
1. Summarise the client's goals, audience and constraints in 120 words.
2. List gaps (questions unanswered) and contradictions (answers that conflict with each other or with the scope).
3. Draft a 45-minute kick-off agenda with timings, who leads each item, and the decisions needed from the client.
4. Draft a one-page kick-off brief: goals, scope, deliverables, key dates, approvers, access status.
Use only what is in the material. Where something is unknown, write "to confirm on the call".An illustrative gap the assistant might flag for the pest-control client: the form says the client wants "as many bookings as possible" while the scope budget is $1,500 a month, and the booking software answer says call-outs are capped at eight a day. The contradiction is a good use of the first ten minutes of the call. Another flag: the approver named on the form differs from the one in the contract. The manager checks the real position before the call, which is the kind of catch that a template-only process misses.
A sample agenda from the same run, trimmed and edited by the account lead:
- 0:00 Introductions and what success looks like in 90 days (account lead, 10 minutes)
- 0:10 Audience and offer: confirm the summary, fix errors (strategist, 10 minutes)
- 0:20 Capacity and conversions: how many call-outs can you handle, and how are bookings tracked? (strategist, 10 minutes)
- 0:30 Approvals and communication: who approves, how fast, which channel (account manager, 5 minutes)
- 0:35 Access: what is done, what is waiting (account manager, 5 minutes)
- 0:40 Actions, dates, next meeting (account lead, 5 minutes)
Notes and actions after the call
After the call, the assistant can turn a transcript or notes into a recap with actions, owners and dates. Keep it to what was agreed; instruct it to mark anything uncertain as "to confirm". Send the recap within a working day, and put the actions in the tracker. A good recap contains no surprises for the client, because every line came out of the call, and the clearest sign of a poor one is a commitment that nobody remembers making.
A second example: an online clothing store client
The same flow adapts to a different client. A store selling clothes online has bought email marketing and a refreshed product-page template. The tailored questions change: "Which email platform holds your customer list, and how many subscribers have opened an email in the last six months?", "What is your returns policy in a sentence?", "Which products drive your repeat orders?" The access checklist changes too: admin access to the email platform for a named user, an invitation to the store's analytics, and read-only access to the product catalogue. The kick-off agenda spends fifteen minutes on list hygiene and sending frequency, not on audience definition, because the proposal already covered the audience. The AI did not know any of this in advance; the manager's scope paste and the question bank produced it, and that is the point of generating the form from the proposal.
The welcome note: before and after
The email that carries the form sets the tone for the relationship, and it is the easiest place for AI to produce something flat. Before, the agency's standard message read: "Please find attached our onboarding questionnaire. Kindly complete and return at your earliest convenience." After a pass with the assistant and a human edit, it reads:
Hi [first name], welcome aboard. To make Friday's kick-off useful, here is a short form (about 20 minutes) and a checklist of the accounts we'll need access to. Please don't send passwords: the checklist shows how to give us access safely, and I'm happy to do it with you on a short call. If you can finish the form by Wednesday, I'll bring a first draft of the plan to the call. [your name]
The edits that mattered: a date with a reason attached, an explicit offer to help, and a sentence about passwords that protects the client. The assistant's first version lacked all three, and added a closing line about being "thrilled to partner", which the manager cut.
When a client cannot or will not grant access
Some clients cannot do it themselves, because the person who owns the account is a former agency, a freelancer or an IT contractor. Plan for three cases. If the owner of record is somebody else, ask the client to confirm who it is, and offer a short email the client can forward, drafted by the AI from your access checklist. If the client is not technical, offer a 20-minute screen-share where they click through the sharing screen while you watch. If the client is reluctant on security grounds, treat that as a good sign and give them the least access that works, for example read-only analytics to begin with, upgrading later when trust and a need exist.
Keep the status in the tracker honest. "Waiting on client" is a different state from "blocked by third party", and the second needs the account manager to chase the third party directly. Ask the assistant to draft the polite chaser, but name the specific person and request in it.
Least access, and taking it back
Every access grant should be the lowest role that does the work, held by a named person, and removed when the engagement ends. Put this in the tracker: for each system, the agency person holding access and a review date. Then add an offboarding step to both ends: when a staff member leaves the agency, their access across client accounts is removed the same day, and when a client leaves, the agency's access is removed and the client confirms it in writing. Ask the assistant to produce the list of systems and users for a given client from the tracker, so the clean-up is a checklist, not a memory exercise.
Illustrative: the agency has eight staff and fifteen clients, so there may be dozens of individual grants. A quarterly review, listing grants by person and comparing with who actually works on the account, will typically find a few that nobody remembers. Removing them costs a few minutes and shrinks the exposure if a staff member's own account is ever compromised.
Reminders that sound human
Automated chasers are where agencies lose goodwill. Keep them short, specific and signed by a person. A day-two reminder for the pest-control client might read: "Hi [first name], two things still open on our list: Google Analytics access (it takes two minutes, steps are on the checklist) and the booking software report. Shall I ring you at 10 tomorrow to do them together? [your name]". It names the open items, offers help and proposes a time. A generic "just checking in" tells the client nothing and gets ignored. Have the assistant draft three versions in your agency's voice, pick the best, and keep it as the template with merge fields for the open items.
Time and cost
For an agency starting from nothing, plan roughly a day to build the question bank and the access checklists for your main services, a few hours to set up the form, tracker and reminders, and an hour to test on a dummy client. After that, each new client costs perhaps thirty to forty-five minutes of the account manager's time for the questionnaire check, the summary review and the pack edit, against several hours of back-and-forth when done by hand. Those figures are illustrative; time one real onboarding both ways before you rely on them. The tool cost is small: a form tool, an automation tool and a chat assistant, many of which have free or low-cost tiers. Check the vendors' pricing pages and your data settings before you put client details into any of them.
What breaks, and how it shows up
- Clients ignore the form. Typically because it is too long or arrives without a human introduction. Send it with a two-line personal note from the account manager, and offer a 15-minute call to complete it together.
- Half the access arrives, half does not. The tracker shows a mix of Done and Waiting for a week. Escalate by phone on day three, since the person who has to act is often a busy owner who has not seen the email.
- The AI summary hides a disagreement. If the summary sounds smooth but the answers were split, re-run with an instruction to quote the conflicting lines. Read the raw answers for anything about budget, authority or deadlines.
- The agenda is generic. If every kick-off looks the same, you are not feeding the proposal's specifics in. Add the "decisions needed from the client" line to the prompt and check it is concrete.
- Over-automation. A client who gets four automated emails in two days feels processed. Cap reminders at two and keep the first and last touch personal.
Measuring whether it works
Track three numbers for the first five clients through the flow: days from signature to the kick-off call, days from signature to full access, and the number of questions the team had to ask again after the form. Illustratively, if signature to kick-off moved from twelve days to five and re-asked questions dropped from six to two, the flow is doing its job. If access takes longest, tighten the checklist wording or offer to join a screen share for the hardest platform. If re-asked questions stay high, the form is too long or the questions are not plain enough. Then watch the early relationship: a client who feels well started is more likely to renew, and reviewing account health to spot at-risk retainers is the next step once the work is running.
For social-media-specific onboarding, a template and AI welcome pack goes further on that sector, and if the agency's longer-term plan is to standardise how it works across clients, a lightweight agency operating system in Notion is where the answers from this flow would live.
Questions about AI in agency onboarding
Should the AI ever see client passwords?
No. Passwords and recovery codes should never go into an AI tool, a form, a chat or an email. Ask clients to grant access through each platform's own sharing or partner features, or through a password manager's sharing function, and let the AI work only from the checklist of what has been granted.
How long should an onboarding questionnaire be?
Aim for fifteen to twenty questions that a client can answer in twenty minutes, grouped by theme. Longer forms get half-answered. Use the kick-off call for anything that needs discussion, and let the AI shorten the form by skipping questions already answered in the proposal.
What should be automated and what should stay personal?
Automate the sending, reminders, chasing, filing and first drafts. Keep the welcome call, the review of strategic answers and any conversation about goals or budget with a person. The client should feel a human has read what they wrote.
How do I know the kick-off pack is right before I send it?
Check each fact in the pack against the client's own answers and the signed proposal: names, goals, scope, dates and responsibilities. Treat anything the AI could not find as a question for the call, not a gap to fill.
Further reads
- How to Mention AI Use in Client Contracts and Proposals — Mention AI use in contracts and proposals clearly.
- AI Rollout Plan for a Ten-Person Marketing Agency — Roll AI out across a small agency team.
- AI Content Workflow for Agencies: From Brief to Approved Draft — The delivery flow that follows the kick-off.
- AI Automation for Financial Advisers: Onboarding Without Re-Keying — A similar flow for advisers, with identity checks.
- AI Automation for Freelancers: Lead Capture to Onboarding in One Flow — A solo version, from lead capture to onboarding.
- How to Write Business Proposals Faster With AI — Write the proposal that feeds this flow.
- PPC Audit Template: Run Google Ads Audits for New Clients With AI — A five-section Google Ads audit template, the AI prompts that summarise the exports, and a filled sample with a one-page prospect summary.
- PR Coverage Report: Build One for Each Client With AI — A coverage report structure, the clippings log columns, AI prompts for the summary, and a sample monthly report with the checks a person must do.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Google Ads Help (manager account links), Meta's business help pages (partner access to business assets) and Google Analytics Help (user access), all checked 4 November 2026. The flow, forms and figures are illustrative.