Strategy
How should a small or medium business buy AI consulting?
Diagnostics, operating reviews and ongoing partnerships explained, with what an SME should expect each engagement to include.
The short answer
For a business of 10 to 250 people, AI consulting usually lands in one of three shapes: a focused diagnostic for a single function, a full operating review over two to three weeks, or an ongoing partnership with someone accountable for the systems in production. The commercial model matters less than a defined outcome, a clear boundary and named ownership.
Key takeaways
- Buy a deliverable, not a day rate. Scope and output decide value, not seniority billing.
- A diagnostic should be cheap enough to say no after, and specific enough to act on.
- A full operating review for an SME is two to three weeks of work, not two to three months.
- Anything priced per seat or per licence is a software sale wearing a consulting jacket.
- Ask what happens after the roadmap. Advice with nobody accountable for delivery is the expensive kind.
AI consulting proposals are often difficult to compare. Here is how the common engagement shapes work, what each should contain, and the questions that expose a weak proposal.
The three engagement shapes
| Shape | Best used for | What you should get |
|---|---|---|
| Fixed-fee diagnostic | One function or decision | One function or up to five workflows mapped, ranked and costed. A short written verdict, not a deck. |
| Operating review | A whole-business view | Whole-business workflow map, ranked use cases with payback, a costed roadmap and a list of what to leave alone. |
| Ongoing partnership | Systems already in production | Someone accountable for what is live: monitoring, fixes, model changes and a monthly report against a baseline. |
A day rate can be a reasonable unit of billing and a poor unit of buying, because it prices attendance rather than outcome. Ask for a defined deliverable, acceptance criteria and the decisions you will be able to make when the work ends.
What should be inside an operating review
- Interviews with the people doing the work, not only the function leads.
- Volumes and annual hours attached to each recurring task, so ranking is arithmetic rather than opinion.
- Build cost, running cost and human review load estimated per use case.
- A written list of what not to automate, with the reasoning.
- Data, tooling and governance gaps that must close before anything ships.
If the proposal cannot say what artefacts you will hold at the end, the price is unknowable by definition. Our own version of this is described on the AI consulting for SMEs page, including the two to three week shape.
Where budgets get wasted
- Paying for a strategy document nobody is accountable for delivering.
- Running five pilots at once so none of them gets a proper evaluation set.
- Buying licences before building skills, which reliably produces low seat usage and a disappointed board.
- Signing a per-seat platform deal when the real cost driver is usage volume.
- Treating governance as a later phase, then discovering a data restriction after the build.
How the tiers should connect
The tiers work best when each one credits into the next. A diagnostic should be deductible from a full review, and a review should feed a build scope directly rather than being re-priced from scratch. If a supplier charges you again to re-learn your business, that is a commercial choice, not a technical necessity.
The sequence we run is diagnosis, then capability, then build, then operation: AI consulting, AI training for employees, AI agents for business and, where the work is telephone-shaped, AI voice agents.
Questions to ask before you sign
- What exactly do I hold at the end, in files rather than adjectives?
- Who owns the code, prompts, evaluation sets and accounts?
- What is the monthly running cost at my expected volume, and what is the ceiling?
- Is the diagnostic fee credited if I proceed?
- Who is accountable when a live workflow breaks at 4pm on a Friday?
The cheapest AI consulting is the engagement that talks you out of four of your five ideas.
Frequently asked questions
- What should AI consulting include for a 50-person company?
- A focused diagnostic should cover one function and end with a clear written recommendation. A full operating review should include interviews, workflow mapping, a ranked shortlist and a costed roadmap over two to three weeks. The proposal should name every deliverable before work starts.
- Should I pay a day rate or a fixed fee?
- Fixed fee, tied to a defined deliverable. Day rates transfer scope risk to you and give the supplier no incentive to finish. Use day rates only for genuinely open-ended advisory time where you control the agenda.
- Is AI consulting worth it if we have technical people in house?
- Often yes, but for a narrower brief. In-house engineers can build; what they usually lack is comparative view across dozens of similar businesses and the time to map workflows outside their own function. Buy the diagnosis and keep the build internal if that suits you.
- How long before we see a return?
- Where scope is narrow and a single workflow is instrumented from day one, payback in the region of six months is a realistic planning assumption. Broad multi-workflow rollouts take longer and frequently fail to produce a measurable baseline at all.
Work through this with us
brep runs AI operating reviews, workforce training and agent deployment for businesses of 10 to 250 people in the UK and US. One diagnosis, one workflow, someone accountable for keeping it working.