Seventy per cent of UK service business enquiries never convert — not because the prospect was not interested, but because follow-up stopped. A proposal goes out. No reply. Three days pass. The business owner means to chase, does not get around to it, and the prospect books with whoever got back to them first. An AI follow-up agent changes that. It monitors every open opportunity in your pipeline, sends the right message at the right time, and alerts you only when a prospect needs a human conversation. Here is exactly how to build one for your service business.
The Follow-Up Gap: Why Prospects Go Cold
Research by Velocify found that responding to an enquiry within five minutes increases conversion by 900% compared to waiting thirty minutes. Most UK service businesses do not respond within five minutes. Many do not follow up a second time at all.
The problem is not motivation — it is system. Follow-up competes with delivery. When you are busy serving existing clients, chasing prospects slips. And when a pipeline grows to ten, fifteen, twenty open opportunities, the cognitive load of remembering who to contact, when, at what point in the sequence, with what message, becomes genuinely unmanageable without a dedicated system.
Manual CRM tracking helps, but it still depends on a human to act on every reminder. An AI follow-up agent removes the dependency on human memory entirely. It monitors your pipeline continuously, identifies which prospects need contact and when, generates the right message for the right stage, and sends it — or queues it for your approval if the message needs personalisation. You spend your time on the calls that need you, not on chasing the ones that just need a well-timed email.
The average UK consultant sends 1.2 follow-up messages per prospect. Research consistently shows that four to eight touchpoints are required before a buying decision is made. The gap between what is sent and what is needed is where most revenue leaks out.
What the AI Follow-Up Agent Does
The agent runs three core functions: pipeline monitoring, message generation, and escalation routing. Each handles a distinct part of the follow-up workflow.
Pipeline Monitoring
The agent connects to your CRM — HubSpot, Pipedrive, or Airtable — and checks every open deal on a configurable schedule. It reads the deal stage, the last contact date, the deal owner, and notes from the most recent interaction. It applies your follow-up rules to identify which prospects are due for contact: a first follow-up 48 hours after a proposal is sent, a second follow-up seven days after that, a final check-in after fourteen days of silence.
You define the rules once. The agent executes them without exception, without forgetting, and without the mental overhead of maintaining a follow-up schedule alongside everything else you are doing.
Message Generation
For each prospect due a follow-up, the agent generates a draft message. It reads the deal notes, the proposal sent, the prospect's company and role, and the stage of the conversation. It produces a short, direct message in your voice — not a template, but a context-aware draft that references where the conversation actually stands.
For a warm prospect who requested time to review a proposal: a brief, friendly nudge that references the specific proposal and offers a call to answer questions. For a cold prospect who went quiet after an initial call: a lighter touch that removes pressure and opens the door without demanding a response.
The drafts go to a dedicated Slack channel or email inbox for your review. You approve, edit, or skip each one. As your trust in the outputs builds, you can set specific message types to send automatically without review.
Escalation Routing
The agent flags two escalation conditions directly to you. First: a prospect who has not responded after the full sequence — so you know when a deal is cold and can decide consciously whether to re-engage or close it out. Second: an inbound reply that requires a human response — a prospect asking a detailed question, pushing back on pricing, or ready to proceed. The agent reads the email thread and routes these to you in Slack with context and a suggested response for you to personalise.
You are involved at the moments that need you. The routine touchpoints run without you.
The Technical Stack
This agent uses the same stack deployed across all QF builds:
- n8n Cloud (£55/month) — workflow orchestration, scheduling, CRM and email integrations
- Anthropic Claude Haiku — message generation and reply classification at low cost per run
- HubSpot, Pipedrive, or Airtable — your CRM as the source of truth for pipeline data
- Gmail or Outlook via API — for sending approved messages and reading replies
- Slack — for approval notifications and escalation alerts
Total running cost: £60–75/month, including Claude API usage at typical service business volume (25–80 open opportunities at any time).
If you are not yet using n8n, the planner-executor pattern post covers how to structure the orchestration layer in more depth. If you are new to AI agents entirely, the first AI agent guide gives you the foundational concepts before you start building.
Building the Agent Step by Step
Here is the full build sequence.
Step 1: Connect Your CRM
In n8n, create a new workflow with a Schedule Trigger set to run twice daily — 9am and 2pm works well for most service businesses. Add a CRM node configured to pull all open deals where the stage is "Proposal Sent" or "In Negotiation" and the last-contact date is more than 48 hours ago.
Map the following fields to n8n variables: deal_name, contact_name, contact_email, last_contact_date, deal_stage, deal_value, deal_notes, and proposal_sent_date. These feed into the message generation step.
Step 2: Apply Follow-Up Rules
Add an IF node to classify each deal into one of four sequences:
- Sequence A — First follow-up: 48 hours after proposal sent, no reply received.
- Sequence B — Second follow-up: Seven days after the first follow-up, still no reply.
- Sequence C — Final check-in: Fourteen days after the second follow-up, still no reply.
- Sequence D — Cold, escalate: More than 30 days since last contact, full sequence completed.
The IF node checks days since last contact and the number of follow-ups already sent — stored as a field in your CRM and updated by the agent on each send — to assign the correct sequence for each deal in the run.
Step 3: Generate the Message with Claude
Feed each classified deal into an Anthropic API node in n8n. Your system prompt should include your name, your firm's name, your preferred communication style, and the sequence context. The user message should include the mapped deal data from Step 1.
A solid system prompt for this node: "You write follow-up emails for [Your Name] at [Firm Name], a UK [type] consultancy. Emails are direct, warm, and short — never more than four sentences. Reference the specific context of the conversation. Never use hollow filler phrases. End with a clear, low-pressure next step. Output only the email body — no subject line, no sign-off."
Claude Haiku handles this well at low cost. Each email generation costs under £0.001 at current Anthropic API pricing — the entire month's message generation for a typical service business pipeline runs under £5.
Step 4: Route for Approval via Slack
Add a Slack node to send the generated draft to a private channel. Format the message to include the prospect's name, deal value, days since last contact, sequence number, and the draft email text. Include two buttons: "Approve & Send" and "Skip this one."
Wire the Approve button to a Gmail or Outlook node that sends the email and updates the CRM — incrementing the follow-up count and setting the last-contact date to today. Wire the Skip button to update the CRM only, logging that the follow-up was reviewed and deferred.
If you have already built the human-in-the-loop approval workflow in n8n, the Slack button pattern here is identical — reuse it directly.
Step 5: Monitor Inbound Replies
Create a second n8n workflow triggered by Gmail or Outlook webhooks. When a reply lands in your inbox from a prospect in your CRM, run it through a Claude node with a classification prompt:
"Classify this email as one of: READY_TO_PROCEED, NEEDS_INFORMATION, TIMING_OBJECTION, PRICE_OBJECTION, COLD_NO_INTEREST, or GENERAL_REPLY. Output only the classification label."
Route READY_TO_PROCEED and NEEDS_INFORMATION to a Slack alert with the full email context and a suggested response. Route TIMING_OBJECTION and PRICE_OBJECTION to a flag for a direct call. Route COLD_NO_INTEREST to a CRM update marking the deal as lost and archiving it. The whole classification step costs fractions of a penny per email.
The Numbers to Expect
Based on the service businesses this has been deployed for, the typical before-and-after looks like this:
- Follow-up consistency: From sporadic (average 1.2 touches per prospect) to systematic (four-touch sequence, every time, for every open deal).
- Proposal-to-client conversion: Typical improvement of 8–15 percentage points within 90 days. For a firm converting at 25%, that is a jump to 33–40% on the same volume of enquiries.
- Time on follow-up tasks: From 3–5 hours per week to 20–30 minutes — reviewing Slack notifications rather than manually writing and sending emails.
- Running cost: £60–75/month for the full agent including n8n and Claude API costs.
The return calculation is straightforward. If your average client is worth £3,000 in fees and the agent improves your conversion rate by three additional clients per quarter, the agent generates £9,000 of additional annual revenue for a £900 annual running cost. The ratio improves further as pipeline volume grows.
The agent does not close deals. That is your job. It makes sure every deal gets a fair chance — that no prospect goes cold simply because you were busy doing excellent work for existing clients. The revenue was always in the pipeline. The agent makes sure it does not stay there.
Common Questions Before You Build
Will prospects know it is automated? No, because messages are generated in your voice, reference your actual conversation history, and are approved by you before sending. They read as personal because they are — the agent drafts, you approve, your name is on it.
What if I do not use a CRM? Build the pipeline list in Airtable first. One table, one row per opportunity, with six fields: contact name, email, deal stage, proposal sent date, last contact date, and follow-up count. The agent reads from and writes to this table. You can migrate to a full CRM later without changing the agent's core architecture.
Can I start with just the outbound workflow? Yes. Build the outbound follow-up component first, run it for 30 days, then add the inbound reply classification workflow once you have confidence in the core system. The AI delegation matrix is useful here — start with the highest-repeatability, lowest-risk component and expand from there.
If you want us to build this for your service business — or want a walkthrough of what the architecture looks like for your specific CRM and email setup — get in touch. We design and build AI operating systems for UK service businesses, and this follow-up agent is one of the fastest-returning builds we deploy.