The first move
You’re looking to implement AI in the wrong place
New AI services can wait. Point AI at the systems you already run. The savings are visible there, and the lessons are yours to keep.
What you are being offered
Something new, at a price you cannot check
The proposals on your desk are for new things. An AI service here, an assistant there. Priced as novelty, or at the old day rate with AI somewhere inside. Either way, you are being asked to buy in the one place where you cannot check the price.
A new service has no history. Nobody in your building knows what it should cost, how long it should take, or what good looks like. So the premium survives scrutiny — there is nothing to scrutinise it against. That is exactly why it is what you are being sold.
Where the saving shows
Point AI at the systems you already run
Your existing stack has the one thing the new service lacks: a history. You know what the last version cost, who built it, how long it took, and what broke. When AI builds out, finishes or fixes the next version, the saving is not a slide. It is a line in your own ledger, next to the old one. That is not nostalgia for the old estate; it is the only place a measurement can start.
The costs collapse where the hours were. By our own arithmetic, work that priced in thousands lands in tens. Your people learn how AI building actually behaves, on ground they can judge. They know these systems, so they can tell good from plausible. The learning comes free with the work.
The second move
Then build AI with AI — and price it yourself
AI-native services are worth building. Build them second. By then you have watched AI rebuild things you understand. You know what a build costs now, and what the governed tail is worth. When the proposal for the shiny thing arrives, you can price it before they do.
That knowledge is the whole game. The premiums you are being quoted for AI work are dictated by the suppliers’ legacy business-model costs, not by the work itself. They are also a bet that you cannot check the number. After the first move, you can. The firms desperate to preserve those premiums will find your building a difficult room to present in.
A firm that is good at AI should already be cheaper at the work it has done for years. Make it prove that on systems you can measure, before you pay new-thing prices for work nobody can.
Start
The first move costs from £300
Eight hours of instrumented discovery finds where the hours hide in your existing stack. Which queue, which system, what the next version would cost through an agent. You keep the model and the figures either way.
Book the discoveryWhat happens in the eight hours →
PSThe backlog you already own is the cheapest AI education on the market. It has been sitting there the whole time.