“AI consultant” is not a protected term, and it currently describes two very different jobs. One is a genuine piece of work: understanding how your business runs, finding where automation would pay, and putting it in with proper controls. The other is a presentation about the future of AI followed by a proof of concept nobody uses.
Both cost real money. Telling them apart before you commit is the useful skill.
What the job actually involves
A good engagement spends most of its time on things that are not AI at all.
Finding where the time and margin go
Mapping how work actually flows from enquiry to invoice, talking to the people doing it, and quantifying where hours and errors accumulate. This is ordinary operational analysis. It is also where the value is decided, because automating the wrong process efficiently achieves nothing.
Deciding what not to automate
Plenty of processes should be simplified, standardised or scrapped rather than automated. A consultant who never recommends the boring fix is selling you a tool rather than solving a problem.
Choosing and configuring
Picking between off-the-shelf tools, features in software you already pay for, and anything custom. Most established businesses need far less custom work than they expect.
Governance and guardrails
What data goes in, who checks the output, what happens when it is wrong, and whether any of it creates obligations under UK GDPR. This is the part most often skipped and the part most likely to cause a problem later.
Making it stick
Training, documentation and a handover so the thing still works in six months when the consultant has gone.
Signs of a good one
They ask about your business before talking about AI. The first conversation should be mostly questions about how you operate, not a capability tour.
They quantify before they build. A baseline – hours spent, error rates, turnaround times – so there is something to measure against afterwards.
They will tell you not to. The most useful answer is sometimes “this would not pay back”. A consultant who finds a use case in every business is not assessing, they are selling.
They are specific about data. Clear answers on where your data goes, what the provider does with it, and what your client contracts allow.
They plan their own exit. Documentation and training, not permanent dependency.
Warning signs
A fixed solution before a diagnosis. If the answer is a chatbot before anyone has looked at your processes, the answer was decided before the question.
Savings estimates with no basis. Percentage claims that do not reference your actual volumes are decoration.
Vagueness about who checks the output. Every automation needs a named human in the loop somewhere. If that is not in the proposal, it is not a finished design.
No mention of failure modes. What happens when the tool is unavailable or confidently wrong? A serious proposal answers this unprompted.
Pressure about falling behind. Urgency is a sales technique, not an analysis.
What it should cost, roughly
Most engagements with established businesses start with a short discovery phase – days rather than weeks – producing a prioritised list with estimated effort and payback. That should be a modest fixed fee, and it should be useful even if you stop there.
Implementation is priced per process. Be wary of large upfront programmes covering a dozen processes at once; the businesses that get value do one properly, measure it, then do the next. Our examples of AI workflow automation give a sense of what a single well-chosen process looks like.
Do you need a consultant at all?
Sometimes not. If you have a technically confident person internally with the time to own it, and a clearly defined process to start with, you may be better served by giving them the remit and a budget for tools. Consultants earn their fee when the problem is genuinely unclear, when governance matters, or when nobody internal has the capacity.
Frequently asked questions
Should we hire a consultant or an agency?
A consultant advises, an agency usually advises and implements. If you have people who can build, advice may be enough. If you do not, buying advice alone often produces a document nobody acts on.
How do we know if their recommendations are sound?
Ask them to show the workings – the volumes, the time per task, the assumptions. A sound recommendation survives you asking where the numbers came from.
What should we have ready before the first meeting?
A rough sense of where your team loses time, your main software, and any client or regulatory restrictions on data. You do not need an AI strategy; that is what you are buying.
Is AI consultancy regulated?
No. The title carries no qualification requirement, which is exactly why references and evidence of delivered work matter more than credentials here.
Start with the process, not the technology
We work the way described above — understand the business first, automate only what pays back, and leave you able to run it. See our process and profit optimisation service, or get in touch.
