AI internal tools: what they cost to build in the UK

A quick reality check on those numbers. Bespoke software in the UK spans roughly £15,000 to £500,000 and up, and the bulk of B2B projects land between £40,000 and £150,000. Internal tools sit toward the friendlier end of that range because the audience is small and known, the design can be plain, and you are not fighting for conversions. The AI adds cost in two places only: the engineering to wire a model in safely, and the extra testing needed when software makes judgements rather than just storing data.

Integrations are usually the biggest single line. A tool that reads from one system and writes to another is cheap; one that has to reconcile three CRMs, a finance package and a shared drive, each with its own quirks and permissions, is not. Every connection is a small project of its own, and legacy systems with no proper API can double the effort on their own. If you are weighing that up, our note on AI data engineering covers the groundwork that makes everything downstream cheaper.

Then there is the question of how much the AI is trusted to do. A tool that drafts a reply for a human to approve is far simpler, and far cheaper, than one allowed to act on its own. The moment software takes an action without a person in the loop, you need logging, guardrails, a way to review what it did and a way to turn it off — the territory we describe under agentic AI. That is not a reason to avoid it, but it is a reason the same feature can cost two or three times more depending on where the human sits.

Data is the third driver. If your records are clean and in one place, a tool can be built on top of them quickly. If they are scattered, inconsistent or full of free text that means different things in different teams, someone has to sort that out first, and that work rarely shows up in an optimistic quote. Reporting and dashboard features add a little on top again; the patterns behind those live in our AI-driven reporting work.

Before you commission anything custom, it is worth asking whether a low-code internal-tools platform would do. Tools in the Retool mould let a developer assemble screens over your data quickly, and for a straightforward admin panel they can be the right answer. The catch is the pricing model: most charge per seat, so the bill grows as more staff use the tool. Per-seat licensing can run from around £4,500 a year for ten users to £45,000 a year once a hundred people are on it, and heavy enterprise use climbs higher still. Open-source and self-hosted options exist that charge for builders rather than users, which changes that maths entirely if adoption is going to be wide.

The rough rule we use: if the tool is simple, the users are few, and the data lives somewhere the platform can reach, buy low-code and move on. If the workflow is genuinely yours, the AI behaviour is specific, the integrations are awkward, or the tool is going to be used by a lot of people for a long time, a custom build usually wins on both control and total cost — you own it outright instead of renting it per head, year after year. A short readiness assessment is often the cheapest way to settle the question before money is committed.

London and the South East run roughly 10 to 15 per cent above the rest of the UK across the board. It is also worth remembering IR35 when a quote leans on contractors: to take home the same money inside IR35, a contractor charging £500 a day outside it needs closer to £640, and that gap tends to find its way into the price one way or another.

The single most useful thing you can do before asking for a price is describe the workflow as it really runs, including the exceptions. Not “we process invoices” but “most invoices match a purchase order, but about one in six don’t, and then Priya emails the supplier and waits”. Those exceptions are where internal tools live or die, and a quote written without them is a quote that will change.

Get those four things down on a page and most of the uncertainty in a quote disappears. The teams that overspend are almost always the ones that left the scope vague and discovered the real requirements halfway through, when changing direction is expensive. If you want a sense of how we approach the build itself, our AI-driven development page lays it out, and there are more cost breakdowns across our insights.

Be wary of a fixed price offered before anyone has seen your systems; it is either padded to cover the unknowns or it will grow once reality bites. Be equally wary of a quote with no line for testing or handover — an AI tool that nobody has stress-tested on real, awkward data is a liability, not an asset. And treat any promise that the AI will be “fully autonomous” from day one with caution. The internal tools that work keep a person in the loop until the thing has earned trust, and the good suppliers will tell you that themselves rather than sell you a fantasy.