What a proof of concept should cost, and what you should get for it

Nobody publishes what a proof of concept costs, which is odd, because it is the first thing anyone wants to know. You can find a hundred articles on how to run one and almost nothing on what a fair price looks like.

So here is ours, with the reasoning attached.

The market rate

For a fixed-price AI proof of concept in the UK, the going rate is roughly £15,000 to £40,000. One firm publishes £18,000 to £45,000 for four to eight weeks. Analysis of 2026 consulting rates puts a scoped POC — one use case, one data source, an evaluation harness — at £15,000 to £40,000, with two to four weeks of senior effort as the right cap.

Freelance day rates for this work benchmark at £600 to £900. That matters, because it tells you what the floor is and what it means. Twenty days at £700 is £14,000. If someone quotes you £6,000 for a four-week proof of concept, they are either not spending four weeks on it or they are not senior.

We charge £18,000, or £10,000 if you are happy to be named as a reference. For smaller businesses where that is out of proportion to the decision, we sell a single day at £1,250 instead.

Why the comparison is not other suppliers

The instinct is to price a POC against other POCs. It is the wrong frame.

A proof of concept is not a deliverable you want for its own sake. Nobody has ever wanted a prototype. What you are buying is a decision you can defend — permission to commit a much larger budget, or permission not to.

So the number to compare against is the decision. If you are weighing a £500,000 programme, £18,000 is 3.6% of it. If the answer comes back negative, you have bought the best possible outcome for the price. If it comes back positive, you have a working thing and a design document instead of a slide.

Priced that way, the question is not "is £18,000 a lot for a prototype?" It is "would I pay 4% to know?"

What you should actually get

This is where published advice gets vague, so here is a concrete list. If a proposal is missing these, ask why.

A prototype that is walkable end to end. Not a happy path. Most prototypes only work while the person who built them is holding the mouse — click the wrong thing and it falls over. Yours needs to survive your champion demoing it to a steering group without the supplier in the room.

A real deployment. Running on a real URL, in an environment somebody could actually inspect. Not a laptop, not a design file.

At least one genuine integration. This is the one people cut, and it is the one that matters — 63% of enterprises name integration complexity as the reason their pilots never reached production. A prototype that touches none of your systems has not tested the thing most likely to kill the project.

A written solution design. Architecture, data model, the flows that matter, the interfaces touched, what is stored and where it goes. This is what lets somebody else build it properly, and it is what your architects will ask for first.

Success criteria agreed before you start. Every serious guide says the same thing: define success upfront, not after the results arrive. "At least 50% reduction in handling time" is a criterion. "See what's possible" is not.

An explicit statement of what was not covered. Scope always narrows during delivery. If the document does not say what the proof did not establish, someone will later assume it did.

A costed route to production, and a verdict. What building it properly involves, roughly what it costs, where the risk sits — and whether you should.

On timescales

Most guidance lands on four to twelve weeks. We fix ours at four.

Not because faster is inherently better, but because a fixed price on a fixed timescale moves the risk to the right side of the table. If it overruns, that is our problem and your cost does not move. Time-and-materials on an exploratory piece of work puts the incentive in exactly the wrong place.

Two weeks would be achievable on a narrow scope, but it invites the obvious question — what are you cutting? Four reads as an estimate rather than a sales pitch, and it is still a fraction of the three to four months a large consultancy will quote.

The question nobody asks, and should

Can it come back negative?

Ask any supplier that directly and watch what happens. A proof of concept that can only conclude "yes, and here is our quote for the build" is marketing with a working demo attached.

Ours can conclude that you should not build the thing. We are not bidding for the implementation, so there is no contract riding on the answer — which is the only reason a positive verdict is worth anything.

If that outcome arrives, the engagement did its job. Finding out for £18,000 beats finding out for half a million and a year.