An AI stack for a ten-person firm, with costs
A ten-person service firm captures most of AI's currently-available value with five layers of tooling costing roughly £200–£450 a month at the time of writing: a general assistant, meeting transcription, a workflow platform, an outbound data layer, and a knowledge layer over your own documents. That is under £5,500 a year for capability that did not exist at any price a few years ago — and everything more exotic can wait until these five are actually being used.
Prices below are rough, current-at-writing figures; check vendors before budgeting. The point is the shape and the order of magnitude, which have been stable even as the tools churn — the selection logic follows AI Automation for B2B: what actually works: boring, measurable, maintained.
What are the five layers?
| Layer | Typical tools | Rough monthly cost |
|---|---|---|
| General assistant | Claude or ChatGPT team plans | ~£18–20 per seat; 4–6 seats is plenty |
| Meeting transcription | Fireflies, Otter, tl;dv | ~£8–15 per seat for client-facing staff |
| Workflow platform | Make, n8n (cloud or self-hosted) | ~£15–50 flat |
| Outbound data layer | Enrichment (e.g. Apollo), verification, a sending tool | ~£60–150 |
| Knowledge layer | RAG over your documents, or built into your assistant plan | ~£0–50 |
Two buying rules keep this honest. Not everyone needs every seat — a paid assistant for the six people who write, sell, or deliver beats ten licences of which four go unused. And prefer monthly billing while the market is moving this fast; annual lock-ins on tools this young are a false economy.
What does each layer actually do all day?
The assistant is drafting, summarising, analysis, and thinking-out-loud — the per-person productivity layer. The transcription layer ends the era of nobody updating the CRM after calls. The workflow platform is the connective tissue; the honest comparison between the main options is in n8n vs Make vs Zapier. The outbound data layer finds, enriches, and verifies prospects — the machine research capability I detail in AI in lead research: what it's actually good at. The knowledge layer answers "how do we do X here?" from your own documents instead of from someone's memory — the build I walk through in An AI assistant over your own documents.
How do the layers connect into a system?
Separately they are conveniences; connected they become infrastructure. The mechanism: when a sales call ends, then the transcript flows through the workflow platform, which writes the summary and next actions to the CRM and drafts the follow-up email for approval. When an enquiry arrives, then it is enriched from the data layer, scored against your ICP, routed to the right person, and acknowledged within a minute. When outbound replies land, then they are classified, drafted against, and queued for a human send. When someone asks how the firm handles a niche situation, then the knowledge layer answers from your own past proposals and playbooks rather than starting a Slack archaeology dig.
None of these flows is more than a day or two of competent build. Together they remove most of the retyping, chasing, and remembering that consumes a small firm's margin.
What does the arithmetic say?
Take a middling configuration: five assistant seats (~£95), four transcription seats (~£45), a workflow platform (~£30), the outbound data layer (~£110), knowledge included in the assistant plan. Call it £280 a month — £3,360 a year, or about £28 per employee per month.
For calibration: a single junior BDR runs £35k+ a year before management. The stack is not a BDR replacement in every respect — but on the specific work of researching, sequencing, and never forgetting to follow up, the cost comparison is not close, and I set it out properly in What a BDR costs vs what an outbound system costs. If a £280 monthly stack recovers even two hours per person per week, it pays for itself several times over.
What should a ten-person firm not buy yet?
Autonomous agent platforms promising to run whole departments; enterprise AI suites priced per mystery; website chatbots for a site with modest traffic; and any tool whose demo you loved but whose weekly use you cannot name. The stack above is deliberately conservative — every layer has published pricing, stable APIs, and a measurable output, which is what separates the boring middle from the expensive frontier.
Roll it out in that spirit: one layer at a time, one owner per tool, a monthly half-hour reviewing what is actually used, and the discipline to cancel what is not. A modest stack fully adopted beats an impressive one abandoned by March.
Next step: the Growth System Audit — £450, seven days, credited against any build — maps where your growth system leaks and what to build first.
Total Format builds the systems UK B2B service firms grow on — AI-powered outbound, automation, and reporting — so growth stops depending on the founder's time.
Map your growth system. The £450 audit takes seven days and is credited against any build.
BOOK THE AUDIT