Why so many Leicester software enquiries now start with artificial intelligence
Enquiries reaching software firms in this market have shifted noticeably. Where a brief once described a portal or a workflow tool, it now frequently opens with a request for something intelligent: automated document handling, demand forecasting, a chat interface over internal knowledge, quality inspection from camera feeds. The underlying business problem is usually the same one that would have been solved with a database and a good interface a few years ago, but the framing has changed, and so has the range of what suppliers are willing to promise.
That creates a specific buying problem. Almost every firm you contact will say it does this work. Very few have shipped a model into production and lived with it afterwards. Telling the two apart is the single highest value thing you can do before signing anything, and it has almost nothing to do with the technology names in the proposal.
Separating genuine capability from a thin wrapper
There is nothing wrong with a supplier building on a hosted model rather than training one. Most sensible projects should. The problem is a proposal that hides which of those it is, because the cost profile, the risk profile and the ongoing obligations are completely different.
Ask for a straight answer to each of these, and write the answers down:
- Is this a call to a hosted model, a fine tuned model, a retrieval layer over our documents, or a classical statistical approach? All four are legitimate. Only one of them is usually being implied.
- Whose account pays for the inference, and what happens to that cost as usage grows?
- What is the fallback when the model is wrong, unavailable or changed by its provider without notice?
- How will we measure whether it is working, and who reviews that measure once the project team has moved on?
- Has a member of the proposed team supported a live system of this kind after handover? Ask for the story, not the logo.
The last question filters more effectively than any technical interview. Building a demonstration is easy. Operating something that quietly degrades when the input data shifts is a different trade, and people who have done it describe the experience in a way that cannot be improvised.
Data readiness decides the outcome long before the build
Ambitious analytics and automation projects fail on inputs far more often than on algorithms. Before you commission anything, establish where the data lives, who owns each source, how consistently it has been recorded, and whether the historical record is long enough and clean enough to be worth learning from. A software supplier that agrees a delivery date without inspecting a real extract is quoting on hope.
Insist on a short, separately priced assessment: pull a representative sample, profile it, document the gaps, and produce an honest statement of what is achievable with the data as it stands versus what would require remediation first. That assessment often reframes the project entirely, sometimes downwards into something cheaper and more useful.
Manufacturing, food production, textiles, distribution and research organisations around Leicester tend to hold decades of operational records in systems nobody planned to connect. That history is an asset, but only once someone has established what the fields actually meant to the people entering them over the years.
Ownership questions that are specific to this kind of work
Standard software contracts assign ownership of code and designs. They frequently say nothing useful about the things that matter most in a data driven engagement, and suppliers do not always raise the gap unprompted.
Settle the following in writing. Who owns the trained model weights and any derived artefacts. Whether your data, or anything derived from it, may be used to improve a supplier's other products or shared offerings. Whether prompts, evaluation sets, labelled examples and the labelling guidelines themselves are yours, since these represent real accumulated investment and are trivially portable to another supplier if you hold them. Where processing happens and under which jurisdiction, because a hosted model may move your records further than your privacy notice promised. What the retention policy is at the provider behind the supplier, not just at the supplier.
Add a clause requiring disclosure of the components in use and a notice period before any of them is substituted. Model providers deprecate versions on their own schedule, and you want to hear about it from your supplier rather than from your users.
The unglamorous work that will consume the budget
In almost every engagement of this type, the visible feature is a small fraction of the effort. The rest goes into integration with the systems that hold the records, into building an interface that operational staff will actually use during a shift, into permissions and audit trails, into handling the cases the model gets wrong, and into training the people whose jobs change as a result.
Ask every software supplier to break the estimate into those buckets explicitly. A proposal in which the intelligent component dominates the cost has either misunderstood your environment or is quoting a prototype. A proposal that puts most of the money into integration, interface and operational change is describing reality, and deserves credit for saying so rather than suspicion for being less exciting.
How to structure the commercial side
Stage the commitment. A short assessment, then a narrow pilot against a single real process with a defined success measure agreed in advance, then a decision point with the right to stop, then production hardening and rollout. Each stage should be priced on its own and should produce something you could hand to a different supplier.
Resist a single fixed price covering all of it. The uncertainty in this work is genuine, so a supplier offering total certainty has either padded heavily or intends to renegotiate later. Equally, resist an open ended arrangement with no checkpoint. The middle path is a series of small, firm commitments with an explicit exit at each boundary.
For the pilot, agree in advance what result would cause you to stop. Teams that define the failure condition before they start are the ones that avoid spending two years on something nobody uses.
Building a shortlist worth your time in Leicester
Look for domain proximity over technology fashion. A software firm that has delivered stock control across a distribution operation, or instrumentation reporting for a research group, understands the constraints of your shift patterns and your compliance obligations better than a generalist with a more impressive stack. The presence of universities, a space research cluster and a dense manufacturing base means specialist experience exists locally, but it is unevenly distributed and rarely visible on a homepage.
Keep adjacent disciplines in their own briefs. Explaining a new internal system to the people who must adopt it is a communications task, and comparing communications specialists or explainer and training video studios separately produces better results than asking a software firm to absorb it. Use the software development company listings to widen the field when the specialism you need is scarce, and put the same brief in front of several suppliers so that the answers differ because the thinking differs, not because the question did.