AI Consulting: What to Expect, What It Costs, How to Choose
AI consulting ranges from a $5k opportunity assessment to a $500k multi-year implementation. How to scope, evaluate vendors, and avoid common mistakes.
A good AI consulting project starts with a list of things not worth automating. If nobody tells you that, you've found a salesperson, not a consultant.
What a proper project looks like
Week 1–2: discovery. Someone looks at how you actually work. Not what the procedures say — what people do. Timing things, asking, noting where the hours go.
The output is a list of processes, each with time consumed and frequency. Numbers, not impressions.
Week 3: prioritisation. Each process gets a cost estimate and a gain estimate. Some drop off the list here, because they don't pay back. Those are the most useful part of the whole report.
Week 4+: build. One thing, the clearest one. Not three at once.
Then: measurement. The same timing exercise as at the start, on the automated process. The difference is the result, not what someone says they feel.
What you get, concretely
Not a deck. A workflow that runs, in your accounts, with documentation on what it does and how to change it. Plus the list of processes left alone, with reasons.
What it costs and why it varies so much
The difference between £4,000 and £40,000 isn't technical complexity. It's how many exceptions your process has.
A booking flow with 5 services and one calendar is simple. The same flow with 40 services, three locations, different rules by weekday and three systems that must talk to each other is a different project.
Always ask: what makes the cost go up in my case? A good consultant already knows the answer.
Red flags
Promising a percentage before looking at your processes. They have no way of knowing.
Selling the technology, not the problem. If model names come up more often than process names from your company, it's a tools sale.
Not saying what can't be done. Every AI system has cases where it gets things wrong. A project with no plan for those isn't finished.
Running on their accounts. Then you're not buying a system, you're buying a subscription to your own data.
What you have to bring
Access to the people doing the work, not just the ones describing it. Answers in days, not weeks. And acceptance that the first month produces no visible results — it produces understanding, which is the precondition for results.
How many hours a week do you lose to repetitive work?
In 20 minutes on Zoom we work out which of them can run on their own. You leave with the answer, whether we work together or not.
Free call, no obligations. You leave with a concrete list of problems and automation opportunities, whether or not you choose to work with us.