Bespoke AI · United Kingdom
Most UK businesses asking for bespoke AI do not need software built. They need an existing platform configured properly around how they already work. That distinction is usually the difference between a result in six weeks and an invoice for six months.
Bespoke AI means an AI capability shaped around one organisation's specific processes, data and constraints, rather than a product everybody buys off the shelf. That is the definition. What it does not specify is how the shaping happens, and that is where most of the money gets lost.
There are two routes. You can commission custom software, where somebody writes an application you then own and maintain. Or you can configure an existing platform, where the model and the infrastructure are supplied and maintained by the vendor, and what is bespoke is everything around them: your documents, your terminology, your approval steps, your data boundaries, your definition of done.
For an organisation of 15 to 200 people, the second route is almost always the right one. It reaches a working result in weeks instead of quarters, and when it is finished you are not left owning a codebase.
Somebody writes you an application. You get exactly what you specified, assuming you specified it correctly. You also get a hosting bill, a security surface, a dependency on whoever wrote it, and a maintenance obligation that outlives the enthusiasm that started the project. Sometimes this is genuinely the right answer. It is the right answer far less often than it is sold.
The platform is Claude, or whatever fits you best. What we build is the layer that makes it yours: the context it works from, the tasks it knows how to do, the boundaries it will not cross, the people who know how to drive it. Nothing to host. Nothing to patch. When Anthropic ships a better model, you get it.
We do both. We recommend the second far more often, and we will tell you plainly which one your requirement actually needs before any money changes hands.
The model works from your documents, your house style, your product names and your sector's language, rather than a generic average of the internet. This is the single change that makes output stop sounding like it came from a chatbot.
The tasks your team does weekly become defined, repeatable workflows with the standard already built in, so the quality does not depend on who is asking or how good their prompt was that morning.
What data may be used, what must never leave, who approves what, and what evidence gets kept. Written against UK GDPR and ISO 42001 from the start, not retrofitted after an incident.
The capability stays in the building. A bespoke system nobody can operate or change is a dependency, not an asset, and it is the most common way these projects quietly die in year two.
Three situations where we will tell you not to buy anything.
An off-the-shelf tool already does it. If your requirement is met by something you can buy for a monthly fee, buy it. Bespoke work to replicate a commodity product is the most expensive way to arrive somewhere you could have got to on a card payment.
The process is not stable enough to describe. If three people in your team do the same job three different ways and nobody can say which is correct, no AI can resolve that. It will simply automate the disagreement. Fix the process first; it is cheaper and it helps whether or not you ever adopt AI.
The data is not in a fit state. Bespoke AI built on disorganised data produces confident, well-formatted nonsense. The remedy is unglamorous and it comes first.
Why most AI projects fail, and what the pattern looks like →
We start by finding out whether you need this at all. The 3-5-4 Method surfaces a first candidate use case in a 45-minute session, and it is deliberately designed to expose the cases where the honest answer is "not yet".
If there is something real, a SPARK assessment establishes the baseline: where the time actually goes, what it costs, and what a fix is worth. Then we build the bespoke layer with your team rather than for them, and train the people who will own it.
Pricing is published rather than quoted on request. See what it costs →
Bespoke AI solutions are AI capabilities shaped around one organisation's specific processes, data and constraints, rather than a product everyone buys off the shelf. In practice, for most businesses under 200 people, bespoke does not mean custom-built software. It means configuring an existing model such as Claude with your documents, your terminology, your approval steps and your data boundaries, so it behaves like it works at your company rather than at a generic one.
Yes, and the distinction matters commercially. Custom AI software development means writing and maintaining an application, which carries build cost, hosting cost and a maintenance burden that outlives the project. Bespoke configuration means the model and platform are supplied and maintained by the vendor, and what is bespoke is the setup around them. For most organisations of 15 to 200 people, configuration reaches a working result in weeks rather than months and leaves nothing to maintain.
It depends on whether you need software built or a platform configured. Configuration engagements are typically priced as an assessment followed by a build-and-train retainer, and are an order of magnitude cheaper than commissioning custom software. Custom development is usually only worth it when your requirement genuinely cannot be met by configuring an existing platform, which is rarer than most vendors will tell you. We publish our pricing openly.
When an off-the-shelf tool already does the job, when the process you want to automate is not yet stable enough to describe, or when the underlying data is too disorganised for any model to work with. Building something bespoke on top of an undefined process produces an expensive version of the same confusion. In those cases the honest answer is to fix the process or the data first, and that advice costs nothing.
For a configuration-led engagement, a first working use case is usually running within weeks, not quarters. The 3-5-4 Method identifies a first candidate use case in a 45-minute session. Custom software development runs to a different timescale entirely, typically months, which is one of the reasons it should be the exception rather than the default.
Data residency and retention depend on the platform and the deployment path chosen, and they are a design decision to be made at the start of an engagement rather than discovered afterwards. Any bespoke AI work touching personal or client data should be scoped against UK GDPR from the outset, with a DPIA where the processing warrants one, and the governance layer built alongside the capability rather than retrofitted.
AI glossary — plain-English definitions of the terms that come up in these conversations.
Claude vs Copilot — which platform fits which job, without the feature matrix.
The hidden costs of AI — what the quote does not include.
Case studies — what this looks like when it has actually run.