The most expensive part of an AI project can be the meeting where everyone agrees it looks promising and nobody agrees what happens next.
For a UK business buying AI consultancy, the useful question is simple: what will your team be able to do when the engagement ends that it cannot do now? That answer should survive without the consultant standing beside the screen.
The short answer: commission a specific workflow improvement, with a baseline, an accountable owner and an acceptance test. Match the delivery arrangement to your team’s availability and data requirements. Senior advice earns its keep when it changes a decision or a working process.
Choose the kind of help you need
An advisory engagement helps you decide where to invest and what to avoid. An implementation engagement changes systems and working practices. Ongoing technical leadership makes decisions across a portfolio of work. These can overlap, but buying one while expecting another creates a predictable argument.
If the problem is “we do not know where AI would help”, start with discovery. If you already have a chosen use case and a process owner, ask for a scoped implementation. If several suppliers are making incompatible decisions, you may need leadership and governance before another pilot.
Use the AI readiness checklist before commissioning a build. Missing process ownership and inaccessible data are not details a model will politely sort out for you.
Make remote delivery explicit
I’m based in Dubai and work with UK businesses remotely; any on-site requirement needs to be agreed. That is a delivery arrangement to plan around, not something to bury in a footer.
Specify the overlapping working hours, the people who attend decisions, how blocked work is escalated and where decisions are recorded. Check the actual time difference for the dates involved, including UK clock changes. Avoid making a programme dependent on everyone remembering which week the clocks moved.
A useful weekly rhythm is a short decision meeting supported by a written update: what changed, what evidence was collected, what is blocked, and what needs approval. Demonstrate working increments in the client’s environment when appropriate. A running commentary in chat is not an adequate project record.
Ask for this evidence before accepting delivery
| Area | What the client should inspect | Who accepts it |
|---|---|---|
| Business result | Baseline and pilot results, including checking time | Process owner |
| Access | Named accounts and a demonstration of restrictions | System or data owner |
| Quality | Representative and deliberately difficult examples | Operational reviewers |
| Cost | Usage assumptions, monitoring and agreed limits | Budget holder |
| Support | Failure route, recovery steps and a named responder | Service owner |
| Exit | Exportable records, source where agreed and account ownership | Contract owner |
The table is my commissioning checklist. It is not a substitute for legal or security review. The NIST framework provides a broader voluntary risk-management reference.
A UK service-business example
Imagine a professional-services team drafting client proposals from approved reference material. Start with one proposal type and a controlled source library. Exclude confidential material from other clients. Separate drafting from the authority to set prices or make contractual commitments.
Test whether reviewers can verify every substantive claim against a source. Measure preparation time and correction time together. Include an obsolete price sheet, missing evidence and a request that should be declined. Those examples tell you more about operational readiness than another beautifully written paragraph.
The output of a good first engagement might be a working drafting assistant, a tested source set and a review procedure. It need not be an elaborate agent system. The smallest useful system is often easier to own.
Keep commercial comparisons honest
Compare proposals on the same scope and responsibilities. One supplier may include data preparation and training; another may assume your team does both. Confirm licences, model usage, hosting, support and any travel separately. Record the pricing currency and treatment of additional work.
Avoid assigning a return on investment to every minute a tool claims to save. Establish whether released time becomes capacity, faster service or a genuine cost reduction. My AI ROI calculator helps make those assumptions visible.
What should the first conversation cover?
Bring the workflow, its volume, a sample of the difficulty and the person who owns the result. We can establish whether you need advice, a workshop or a bounded implementation, and whether the proposed delivery arrangement fits your team.
The right starting point is a decision you need to make, not a platform someone needs to sell. Agree the evidence for that decision, then commission the work that produces it.
Sources and scope
Reviewed 1 November 2026. The examples and worksheets are my suggested working methods, not measured client results or official standards.
