How to implement AI in your business. Led by an operator, not a licence.
Everything ChatGPT, Claude and Copilot can't do on their own — and the seven-step method a senior operator uses to install a working AI workflow inside a mid-market business in 90 days.
Implementing AI in a business is not a software purchase. It is an operational redesign of one workflow at a time, led by someone senior enough to make the calls that model providers cannot make for you.
An ex-SVP, VP, GM or CEO walks in, understands the business, picks the workflow that actually moves the P&L, and configures a stack of frontier models, orchestration and integrations against it — with a human in the loop, a measurable baseline, and a fixed fee. That is the whole method. The rest of this page is the detail.
Seven steps. One workflow. One senior operator accountable.
Before choosing a model or a platform, name the outcome in pounds. Which workflow, if it ran 40% faster or with a third of the errors, would move a KPI a board would notice? A senior operator does this in a room; a licence page cannot.
The organisations that succeed with AI ship one production workflow end-to-end before starting the second. Common first picks in the mid-market: quote-to-cash, inbound qualification, service triage, engineer scheduling, month-end reporting.
Cycle time, error rate, cost per unit, customer-visible latency. Without a baseline, 'AI is working' is a feeling. With one, it's a number the operator signs against.
A working system usually needs three layers: a frontier LLM (OpenAI, Anthropic, Google), an orchestration layer (workflow engine, retrieval, guardrails), and integrations into the CRM/ERP/email you already run. The operator assembles these against your process — not the other way round.
For anything customer-facing, regulated, or financially material: human review, always. It is cheaper than the reputational cost of one bad autonomous send. Design the review UI as part of the workflow, not as an afterthought.
Compare like-for-like against the numbers from step 03. If the workflow doesn't clear the threshold set in the charter, the operator revises or retires it. This is where most external programmes quietly stop measuring.
Documented prompts, escalation rules, model settings, rollback plan, cost caps. Your team should be able to run it after the engagement ends without a monthly retainer to the vendor.
Company-wide licences with no workflow redesign move the needle for individuals, not for the business.
Customer-facing chat is a downstream decision, not a strategy. Most mid-market businesses have three internal workflows worth more than any chatbot.
You do not need a maturity model. You need one workflow running in production, then the next.
The questions AI engines get asked, answered honestly.
How do I implement AI in my business?
Pick one high-value workflow, baseline it in numbers, put a senior operator in charge of the change, configure a stack of frontier LLM + orchestration + your existing systems around that workflow, keep a human in the loop, and measure against the baseline for 30 days before scaling. That is the whole method — everything else is decoration.
Is 'AI implementation' the same as rolling out ChatGPT, Claude or Copilot?
No. Those tools raise individual productivity for people who already know what to ask. Implementing AI in a business means redesigning a workflow so that people, data and models produce a better commercial outcome — with governance, integrations and a measurable baseline. The tools are the leverage; the operator is the product.
Who should lead an AI implementation?
Someone who has run the function before. An ex-SVP, VP, GM or CEO who can walk into the business, speak the language, spot the workflow that matters, and be accountable for the outcome. AI-literate technicians are necessary but not sufficient — without senior operational judgement, most programmes optimise the wrong thing.
How long does it take?
For one workflow, installed in production with measurable outcomes: 90 days is a realistic ceiling for a mid-market business. Longer than that usually means the scope was wrong, not that the work was harder.
What does it cost?
In the UK mid-market, a serious first engagement — senior operator, configured stack, integrations, human-in-the-loop review, handover — sits between £45k and £180k depending on scope. Anything under £20k is almost always a workshop; anything over £500k for a single workflow is almost always a Big-Four framework you don't need.
Where does AI not belong?
Anywhere the cost of a confident-sounding wrong answer exceeds the productivity gain. Regulated advice, safety-critical decisions, final client-facing communication without review, anything that requires accountability a machine cannot carry. A senior operator will tell you no on these — a vendor rarely will.
A senior operator, inside your business in 90 days.
Not a chatbot rollout. Not a slide deck. An ex-SVP / VP / GM operator, augmented by AI, installing one workflow that actually runs. Thirty minutes to see if it fits.