AI is buying you hours, not sales. Price it that way.
The research is consistent and slightly awkward for anybody selling this: AI saves time far more reliably than it makes money. That should change how you buy it, what you measure, and how you are charged.
Hours, not sales
Most AI proposals are written as growth documents. More leads, faster sales cycles, better conversion. It is an easy story to tell, and for a small number of businesses it turns out to be true.
The trouble is that the evidence points somewhere else, and it points there consistently.
What the numbers actually say
Among UK businesses adopting AI, 75 per cent report improved productivity. Only 12 per cent report higher revenue.
DSIT AI Adoption Research, 3,500 UK businesses, published January 2026.
UK small and medium businesses using AI say they save more than half a day a week, 5.2 hours.
OpenAI research by Opinium, 1,000 UK SME decision-makers, February to March 2026, published by Enterprise Nation, 25 March 2026.
Those two findings agree with each other and with almost everything else worth reading. The gain shows up as time. It shows up as sales far less often, and when it does, it is usually because the time was reinvested in something that sells, which is a second decision that somebody had to make deliberately.
We publish these numbers on our pricing page as well, including the 12 per cent, which is not a flattering figure for a company that sells AI agents. We would rather you saw it from us than from somebody else in month four.
Why this changes what you should buy
If the return is time, then the shape of a good project changes.
You stop looking for the most exciting process and start looking for the most expensive one: the task that quietly eats four hours of a well paid person's week, every week, and that nobody has ever costed because it has always just been part of the job.
Those tasks are usually unglamorous. Chasing a purchase order. Rekeying the same figures into a second system. Reading a shared inbox and deciding who owns each message. Reconciling two reports that should agree and do not. None of them make a good slide. All of them make a good first agent, because the work is repetitive, the rules are knowable, and the saving is measurable the week it goes live.
Agree the one number before you start
Here is the discipline that matters more than any technology choice. Before anything is built, write down a single number that has to move, how it is measured today, and who signs off that it moved.
Hours spent on the task in an average week. Days from enquiry to quote. Invoices touched by a human. Percentage of messages that reach the right person first time. One number, not a scorecard.
The reason to insist on one is that everybody can find a metric that improved. A single number agreed in advance, measured the same way before and after, is the only version of this that cannot be dressed up afterwards. It also protects you from us: if the number does not move, we have nothing to hide behind, and you know within one quarter rather than one year.
The report that comes back is one page and looks like this. The third column is the one that matters:
hours a week before, timed across four weeks with nothing changed.
hours a week in the first quarter after, measured exactly the same way.
change in the second quarter. It did not move, and the report said so on the front page.
An example of the report format, not a customer’s result. Any supplier will send you the first two columns. The third one is the reason to believe the first two.
What this means for how you are charged
If the value is hours, then pricing tied to promised revenue is selling you a thing the research says will probably not happen. Pricing tied to hours and running cost is selling you the thing that will.
Practically, that means three things you can ask any supplier for. A fixed price for the build, so the scope conversation happens before the money does. A published monthly price for running it, so it can go into a budget as a line rather than an estimate. And a report against the one agreed number, delivered on a schedule, including the periods where it did not move.
The last one is the test. Any supplier will send you a report when the figures are good.
The awkward corollary
If an agent saves your team six hours a week and nobody decides what those six hours are now for, you have not saved anything. You have created slack, and slack gets absorbed silently.
So the saving needs a destination, agreed with the person whose hours they are, on the day it goes live. More time on the work that only a human can do, a role that stops needing backfilling, a service level you could not previously offer. Something specific.
This is the part that no software can do for you, and it is the part that decides whether the project reads as a success in twelve months.
Where to start
Find the task nobody wants to admit how long it takes. Cost it honestly at the salary of the person doing it. Then ask whether an agent could take the repetitive nine tenths of it and leave the judgement with the person.
If the answer is yes, that is a first agent. If the answer is that the task needs judgement all the way through, it is not, and a supplier who tells you otherwise is selling you the demo rather than the job.
Tell us the job that never gets done on time.
We will tell you, in plain English, whether an agent can take it, what it would take to build, and what it would cost to run.