Can you add AI to our existing product?

We can, and we’ve done it for products where getting it wrong wasn’t an option. But the first thing we’ll check is whether you need to build anything at all. Many teams get most of the value from a licence, and we’ll say so.

That goes for founders and for operators running an established business. The question is the same either way: what job should the AI do, and what has to be true for it to do it well?

Buy, configure or build

There are three ways to put AI to work, and most teams should start at the top.

• Buy a licence. ChatGPT Business or Microsoft 365 Copilot covers drafting, summarising and research for a monthly fee per person.
• Configure an assistant. Set one up over your own documents when people keep asking the same questions of the same files.
• Build custom. Write software that connects to your systems, takes actions and keeps a record of what it did.

Our rule of thumb is blunt. If an off-the-shelf tool does the whole job, buy it and don’t look back. If it does 80% of the job, look hard at the other 20%, because that’s where the pain lives. Our guide to off-the-shelf AI versus custom sets out the full decision.

Where custom AI pays

A build earns its place when at least two of these are true:

• Volume. Hundreds or thousands of items a week, every week.
• Integration. The AI has to read from and write to your own systems.
• Control. You need to log every decision, or keep data in a specific place.
• Conditions. It has to work where a chat window can’t, like offline or inside your app.
• Ownership. The AI is part of what you sell, so you need to own the code.

If none of those fit, keep the licence and spend the money elsewhere. Our AI readiness scorecard helps you rank the jobs worth looking at.

Two products we added AI to

Everyone’s Invited was receiving over 100 submissions a day. Each one had to be stripped of names, dates and locations before anyone reviewed it. Pasting survivors’ stories into a chat window was never going to be acceptable. The system we built has handled more than 50,000 submissions and saved around 12,500 hours of manual review.

Pristine Condition needed AI that checks someone’s lifting technique through a phone camera, in real time, with no signal. A spot check might happen in an aircraft cargo hold. We built it with TensorFlow, running on the phone itself. Usage has grown by around 9,000%, and around 150,000 warehouse staff now use it.

Neither project started with the model. Both started with the job, the people doing it and the data they had.

Is your data ready?

Most AI projects that go wrong don’t fail on the model. They fail because nobody checked the data first.

Here’s a quick test. Could a new member of staff do the job using only the data you’d give the system? If they’d need to ask a colleague or guess which version is current, the AI will struggle too.

Pick fifty real examples and try it. Our guide to getting your data ready for AI has the full checklist. If your product handles personal data, is our data safe if you use AI? covers that side.

Built so the model can change

Models change every few months, so we treat the model as a part you can swap. The instructions, the test set and the scoring sit outside it. When a better or cheaper model arrives, we run the tests against it and switch if it wins.

The target isn’t 100% accuracy. It’s doing better than the person doing the job today, measured on the same examples.

What it costs

Less than you might fear to run, and more than the model bill to build. The model fees are often the smallest line in the budget. Our AI chatbot cost guide works through the numbers.

If AI is part of a new product or a major rebuild, it fits LaunchPad. Growth is £60k over three months and includes complex data and AI features, excluding VAT. For an internal tool, it starts with a free chat, then a paid discovery that ends in a scoped plan and a price.

Either way, you own the code and IP, as do I own the code and IP? explains. If someone has already half-built your AI feature, our free architectural review tells you what we’d keep and what we’d change. Our AI + data service shows how we approach the rest.