Adding AI to vertical software: where to start

Start with a costly, repeated task

The best first AI projects take something people do often, that follows patterns, and that costs real time. Ask your support, sales and product teams where the hours go, and look for questions that get answered the same way every week.

Avoid starting with "we need a chatbot". Start with "our support team answers the same 40 questions every week" or "customers get stuck setting up their first job".

Good first projects

• A support assistant trained on your manuals and help docs, that answers customers' questions and shows the source it used, and hands over to a person when it isn't sure.
• AI inside your product, at the point where customers get stuck: drafting a job description, suggesting the next step, summarising a long history.
• Internal tools for your own team: summarising calls, drafting follow-ups, triaging tickets, finding the right answer in years of documentation.

Build features that help your customers do their own work better. That's what makes your product harder to leave.

Prototype on real data, quickly

A working prototype on your real data, within weeks, tells you more than months of planning. Collect a set of real cases (real questions, real jobs) and test the prototype against them. Keep that set: it becomes your test for every future change.

Measure it like any feature

Decide up front how you'll know it works: tickets resolved without a person, setup completed, time saved, customers using it again. Watch for wrong answers as closely as right ones, and make it easy for users to flag a bad one.

Data protection from day one

If customer data is involved, UK GDPR applies. Work out what data the feature needs, where it's processed and stored, who can see it, and how long it's kept. Use providers that don't train their models on your data, and write it into your privacy notice and customer contracts.

What it costs and how long it takes

A focused first AI feature, scoped properly, typically takes weeks, not months, to prototype, then longer to harden, test and connect to your systems. Running costs depend on usage and can be estimated from the prototype. Our AI Products offer sets out how we scope, build and hand over.