Start with the problem you can describe most clearly. If you can name one recurring task, its owner and the result you need, a scoped automation pilot is usually the most direct start. If you have several ideas and no clear priority, start with an assessment. If people are already using AI inconsistently, start with training. If AI work exists but nobody owns the decisions, you need ongoing leadership rather than another experiment.
The four routes below are the engagements I offer. Their published starting fees are included so you can compare like with like. They are starting points for a defined scope, not market benchmarks, and the free introductory call exists to check which one fits before any paid work begins.
How do the four starting points compare?
| Starting point | Best first step when | What you receive | Published starting fee |
|---|---|---|---|
| AI training | Your team needs practical working rules for real tasks | Tailored sessions, working rules and an adoption plan | From £1,250 per half-day session |
| Readiness assessment | You have several ideas and need to compare value, data and risk | A workflow map, an opportunity and risk register and a prioritised 90-day roadmap | From £2,500, about two weeks |
| Automation pilot | One recurring workflow has a named owner and usable inputs | A working workflow, acceptance tests and documentation | First pilots from £1,750, four to eight weeks |
| Fractional AI leadership | AI work exists but nobody owns the priorities and decisions | Defined decision ownership, a weekly operating cadence and a monthly evidence review | From £3,000 a month, one to three days a week |
When should training come first?
Training is the right first step when the work already exists and the main risk is how people use AI on it. Typical signs: colleagues paste client information into tools without agreed rules, outputs are used without a check, or each person has invented their own approach. A build at this stage would automate a process nobody has agreed yet.
My AI training for businesses is built around your organisation's own tasks and risks. It produces working rules your team can follow: what information to use, what to check and when a person remains responsible. It does not include an automation build. If the sessions reveal a workflow worth building, that becomes a separately scoped piece of work.
When is an assessment the better first step?
An assessment fits when you have more ideas than evidence. You may suspect that reporting, enquiry handling and document preparation could all improve, but you do not yet know which has usable data, a clear owner and a tolerable cost of failure. Building the wrong one first is the expensive mistake.
The AI readiness assessment compares candidates on potential value, data access, failure cost and governance needs, then sets out an ordered 90-day plan. Implementation remains a separate engagement, so commissioning the assessment does not commit you to a build. If you would rather work through the comparison yourself first, the first-process scorecard covers the same questions in a simpler form.
When can you go straight to a pilot?
Go straight to a pilot when you can answer five questions without guessing: what starts the task, which information it uses, what the finished output looks like, who checks it and who owns the process afterwards. If those answers exist, an assessment would mostly confirm what you already know.
An AI automation pilot turns one of those workflows into a working system with acceptance tests and documentation. Anything that would be sent or published still waits for a person to approve it. The pilot cost guide explains what the fee covers and which running costs to plan for.
When do you need ongoing AI leadership instead?
Some businesses have already run several experiments. The problem is not a missing workflow or missing skills. It is that nobody owns the priorities, the technical choices and the decision to stop or continue. Another pilot added to that situation usually becomes another unowned experiment.
A fractional chief AI officer takes defined ownership of the roadmap, important technical decisions, governance and delivery oversight for one to three days a week, without a full-time hire. Progress is reviewed monthly against agreed criteria and the available evidence.
What are the signs you picked the wrong starting point?
- Training that cannot be applied. The sessions were well received, but nobody changed how a specific task is done because no task was chosen.
- A pilot without an owner. The build works in testing, then stalls because no one is responsible for running it or approving its output.
- An assessment for an obvious case. The process, data and owner were already clear, so the roadmap only repeated the decision.
- Leadership without a decision to make. A retained role with no agreed remit drifts into general advice.
Each of these is avoidable by naming the problem first. That is the purpose of the first conversation.
Can you combine them?
Yes, but in sequence rather than all at once. Illustrative sequence, not a client result: an assessment identifies weekly reporting as the strongest candidate, a pilot builds and tests that workflow, and a half-day training session then shows the people who will operate it how to review its output and handle exceptions. Each step is scoped separately, and you can stop after any of them.
A fractional engagement can sit around the same sequence when the business needs someone accountable for the whole programme rather than a single piece of work.
How do you decide in one conversation?
Bring a short description of the work that is causing friction, who does it today, what information it depends on and what has already been tried. With that, the free 30-minute introductory call is usually enough to say which of the four routes fits, or whether none of them does. Detailed diagnosis and scoping are paid work, so the call stays focused on fit.
If you are hiring for a role rather than commissioning a project, my experience page sets out the contract work I am open to.