We have been quoted thousands for an AI project, how do I tell if it is worth it?

What should I ask before signing anything?

Five questions, and the quality of the answers tells you more than the price does.

What does this process cost us now, and how did you work that out? A good supplier will have asked. If the proposal contains no baseline, they are selling a solution to a problem they have not measured.

What does good look like, and how will we know? A specific metric, agreed before the work starts. Handling time, cycle days, error rate, backlog. If success is defined as "the system is live", that is a delivery milestone, not an outcome.

What happens if it does not work? Not a refund promise, a genuine answer about how you would find out early and what you would do. Phased work with decision points is a good sign. A single twelve-month commitment on the strength of one meeting is not.

Who is doing the work, and will I meet them? In small consultancies the person who sold it is usually the person who delivers it, which is fine. In larger ones, ask.

Do you resell any of the software you are recommending? Not disqualifying if the answer is yes, but you deserve to know before you weigh the recommendation.

What are the actual red flags?

Vagueness dressed as sophistication, and a proposal that never mentions your people.

If you cannot understand what will physically change on a Tuesday morning after this is delivered, the proposal is not finished. If the word "transformation" is doing more work than any specific process name, be careful.

The people one matters more than it sounds. Most AI projects that fail do not fail on technology, they fail because nobody uses the thing. A proposal with no plan for how the team gets involved is a proposal with a large unpriced risk in it.

And be wary of anyone who says yes to everything. A supplier who has never told you a part of your idea is a bad one is either unusually lucky or not being straight.

How do I do the payback maths myself?

Roughly, and roughly is enough to make a decision.

Take the number of people doing the task, multiply by hours a week, multiply by 46 working weeks, multiply by a loaded hourly cost. That gives you the annual cost of the task as it stands.

Then be honest about how much of it actually goes away. Almost never all of it. Half is a strong result on a well-chosen process. If half the annual cost does not comfortably exceed the quote inside about eighteen months, the case is thin.

That is deliberately crude, and the crudeness is the point. If the decision only works on a sophisticated model, it probably does not work.

When is an expensive quote the right answer?

When the process touches money, customers or regulation, and getting it wrong costs more than the project.

Cheap work on a regulated process is not a saving. If it involves customer outcomes, pricing, eligibility or anything an ombudsman might look at later, the governance and testing are a real cost and a supplier who has not priced them has not understood the job.

The honest version is that some quotes are too high, some are about right, and some are too low in a way that will cost you more later. The five questions above sort them faster than comparing numbers.

Before you spend anything

Two free things worth doing first. The Quick Check at the Quick Check page gives you an indicative readiness score in two minutes with no call. And a free discovery call at our booking page is forty-five minutes on your process, not on my services, and a fair number of them end with me saying not yet.

For the numbers, our AI consulting cost guide sets out the pricing models honestly, and the AI ROI Calculator does the arithmetic above for you.

Frequently Asked Questions

Your next step

Take the free two-minute Quick Check for an AI readiness score, see what the £995 AI Visibility Audit covers, or book a free 45-minute discovery call. Definitions for the terms used here are in the AI glossary and the Claude glossary.