Where AI fits in a product

A product can gain a lot from AI in one place and very little in another. The people who run it have to decide where it belongs. The UK government has written guidance for its own teams on making this choice. This page draws on that guidance, and on rules that apply to products sold in the EU.

Start with the problem

The government's AI Playbook says the choice of where to use AI must be led by business and user needs. It should start from pain points and inefficiencies, and from what people need, before what the technology can do.

The GOV.UK Service Manual makes a similar point about any new service. Teams are often handed a ready-made solution at the start. The manual asks them to reframe it as a problem to be solved. Its example is a team told to build an interactive map of contact centres. The real problem is how to help people find their nearest centre when they need a face-to-face appointment.

Tasks that suit AI

The Playbook suggests focusing on problems that only AI can solve, or where it has a clear advantage over existing methods. It gives some signs. A task may involve more volume, complexity or speed than a traditional system can handle. It may involve spotting patterns in large data sets, or tailoring something to each person. The Playbook suggests a cost-benefit analysis to weigh the likely gains.

For generative AI, the Playbook adds one more question. Could a solution without AI, or a simpler machine learning model trained for one purpose, work just as well?

Decisions that need a person

The Playbook asks teams to be cautious about fully automated decisions that are significant for people, such as those about health or safety. It says AI should not be used on its own in high-risk areas. Teams that use AI to support decisions about people can turn to guidance from the ICO and the Alan Turing Institute. It explains how to tell the people affected how a decision was reached.

Can the team build it and run it?

The Playbook asks whether a team has the skills and the infrastructure to build and maintain an AI feature. If not, it may need to train staff, hire or work with a partner. An AI product also has a full life cycle. A team needs to be able to support it day to day, update it and close it down securely. The Playbook also suggests a fallback, so the service keeps working if the AI part has to be switched off.

Selling into the EU

Products sold in the European Union fall under the EU AI Act. The European Commission says the Act became applicable on 2 August 2026, with some exceptions. It sorts AI by risk. Most AI in use in the EU, such as spam filters, falls into the minimal-risk group, which has no new rules. Chatbots carry a transparency duty, so people know they are dealing with a machine. Rules for high-risk uses such as employment and education apply from 2 December 2027.

Useful links