Most UK service businesses spend between 8 and 11 hours onboarding a single new client. Contracts, welcome packs, intake questionnaires, system access, first briefing calls — each one handled manually, each one a context-switch that pulls you away from paid work. One properly configured AI agent can compress that to under an hour. This tutorial walks you through exactly how to build it.
Why Client Onboarding Is the Perfect First Automation
When businesses ask us where to start with AI agents, onboarding is almost always our first recommendation. Here is why: it is high-frequency, high-effort, and almost entirely templated. The process for onboarding client A looks almost identical to the process for onboarding client B — the same documents, the same questions, the same system setup. That pattern recognition is exactly what AI agents are designed to exploit.
The second reason is that the ROI calculation is clean. If you onboard four new clients a month and each one takes 10 hours, that is 40 hours of operational overhead before you have done a single hour of billable work for those clients. Automate that to 1 hour per client and you recover 36 hours every month — permanently.
Third: onboarding is almost entirely a coordination and administration task. You are collecting information, generating documents from templates, sending communications on schedule, and updating systems. These are exactly the jobs AI agents do well. Unlike tasks that require genuine judgment — strategy, creative work, client relationship management — onboarding is systematisable. If you have ever thought "why am I doing this again?", an agent can probably own it.
If the process for onboarding client A looks identical to the process for onboarding client B, an AI agent should own that process — not you.
The Five Stages of Client Onboarding (and Where AI Fits)
Before you can automate anything, you need to map the process. Most service business onboarding — whether you are a consultant, agency, coach, accountant, or recruiter — follows five stages. Here they are, with an honest assessment of what an AI agent can own versus what still needs a human in the loop.
Stage 1: Initial Data Collection
The client has said yes. Now you need their contact details, company information, billing details, and the specific information relevant to your engagement. This stage is 100% automatable. A form linked to your agent triggers an intake workflow — the agent extracts the structured data, creates a CRM record, and moves to stage two. A human only needs to see the output, not manage the process.
Stage 2: Document Generation
Contracts, service agreements, welcome packs, project briefs — these are template documents with variables filled in from stage one. Name, company, scope, start date, pricing. An AI agent can generate all of these in seconds. What used to take 45 minutes of document editing now takes 30 seconds.
Stage 3: Document Delivery and Signature Chase
Once documents are generated, they need to be sent, signed, and chased if not returned. An AI agent sends the initial email, checks for signatures on a schedule, and follows up after 48 hours if nothing comes back. It flags to you if something has stalled at day five. Your involvement is exception handling, not routine chasing.
Stage 4: System Setup and Access
This varies by business, but typically involves creating a project folder, setting up a shared workspace, granting tool access, and adding the client to your project management system. Some steps require genuine judgment. Others — folder creation, CRM record completion, calendar invite for kick-off call — are fully automatable via MCP-connected tools. As we covered in the MCP explainer, each tool your agent needs to interact with can be connected through a published server. Google Drive, Notion, HubSpot, Asana — all have MCP servers available now.
Stage 5: Kick-Off Preparation
The day before the first client meeting, your agent prepares a briefing document: a summary of what was collected at intake, the key deliverables agreed in the contract, background research on the client's business. This is the kind of preparation that used to take 30 minutes of review time. With an agent doing it automatically, you walk into the kick-off meeting already prepared — without spending the prep time yourself.
Building the Agent: What It Actually Needs
An onboarding agent is not a single monolithic script — it is a set of triggers, conditions, and actions that run in sequence or in parallel depending on what has happened. Here is the minimal stack you need.
An intake trigger. The event that starts the workflow. A form submission, a CRM status change (deal moved to "Closed Won"), a manual trigger when a client signs, or an email matching a specific pattern. The trigger tells the agent: onboarding has started, here is the client data.
Document templates with variable slots. Before you build anything, create clean template documents with clear placeholder variables: {{client_name}}, {{company}}, {{service_scope}}, {{start_date}}, {{monthly_fee}}. The cleaner your templates, the more reliably the agent fills them.
Connected tools via MCP. Your agent needs to reach your Google Drive to create folders, your CRM to update records, your email to send documents, and your e-signature tool to dispatch contracts. Each connection is an MCP server configured once and reused across all your agent workflows — not just this one.
A state tracker. Onboarding is multi-step and takes days. Your agent needs somewhere to track where each client sits in the process — what has been sent, what has been signed, what is outstanding. A simple database table or a CRM custom field works. This is what allows the agent to resume correctly after a 48-hour wait for a signature.
An exception escalation path. Define the conditions under which the agent stops and flags to you: contract not signed after five days, intake form incomplete, client email bounced. Everything else runs without your involvement. You only touch exceptions.
Build for exceptions, not for the normal path. On the normal path, the agent runs start to finish without you. You appear only when something unusual happens.
Real Numbers: What to Expect
Based on the onboarding automations we have built for UK service businesses over the past 12 months, here is what the numbers typically look like:
- Pre-automation time per client: 8–11 hours (average 9.5 hours)
- Post-automation time per client: 25–60 minutes (agent runs automatically; human reviews exceptions)
- Time saving per client: approximately 9 hours
- Break-even point: First or second client after deployment (build time is typically 8–16 hours)
- One-time setup cost: £1,500–£3,500 depending on complexity and number of connected tools
- Monthly maintenance: Near-zero once stable — the agent runs, you review the exception log
For a consulting firm onboarding two new clients per week, the time recovery is around 18 hours per week, every week. Over a year, that is roughly 900 hours — the equivalent of a part-time operations manager, at a fraction of the cost.
We documented a similar result in the Yorkshire accountancy firm case study: 22 hours of weekly admin recovered through a three-agent system, with client onboarding forming a significant part of the automation. The numbers are consistent across different business types — the savings are real and they compound.
The Stack That Makes It Work
You do not need to build this from scratch. For the onboarding automations we deploy, we use three components.
OpenClaw as the agent orchestration layer. It handles the multi-step workflow logic, the state tracking, and the trigger management. When the intake form fires, OpenClaw routes the event to the right agent, manages the sequence, and handles the retry logic for the signature chase.
MCP servers for tool connectivity. Google Drive MCP for folder creation and document storage. Gmail MCP for sending and monitoring emails. A CRM MCP server for updating client records. Each connection is a configuration exercise, not a build project.
Gemini or Claude for the intelligence layer — the parts that require reading, summarising, drafting, or making conditional decisions. Kick-off briefing generation, email drafting, intake data extraction, contract variable population. Language model tasks that plug into the workflow wherever judgment is needed.
If you have already read the 90-day transformation playbook, onboarding automation fits naturally into the first 30-day sprint — self-contained, high-value, and a good first test of your agent stack before you take on more complex workflows.
Getting Started This Week
You do not need a perfect system to start. Here is the minimum viable version you can build this week:
- Document your current process. Write out every step you currently take when onboarding a new client — the tools you use, the time each step takes, how often you do it.
- Clean up your templates. Take your contract, welcome pack, and intake questionnaire. Add proper variable placeholders and remove anything hard-coded to a specific client.
- Pick one trigger. CRM deal won, form submission, email confirmation — pick one and start there.
- Automate one stage first. Start with document generation — the highest-leverage, lowest-risk step. Get that working reliably before moving to the next stage.
- Add the signature chase. Once documents generate reliably, automate delivery and follow-up. This single step recovers most of the routine time.
A useful principle from the first agent guide: start with the process you do most often and dislike the most. Onboarding usually qualifies on both counts.
If you want help mapping your current onboarding process and identifying exactly where an agent can take over, get in touch. We scope and design an onboarding automation in a single 90-minute session — most clients leave with a clear build plan they can execute in two to three weeks.