®AI is now easy to add to almost any product, which makes it tempting to add it everywhere. The best AI features start with a specific job that people already find slow or difficult.
Good places to start
- Customer support. An assistant that answers common questions from your own help content, and hands over to a person when it is unsure.
- Search. Let users describe what they want in plain words instead of using exact filters.
- Documents. Summarise long reports, pull details out of invoices or draft replies for your team to check.
- Content. Suggest product descriptions, titles or translations that a person approves.
- Automation. Sort incoming emails, tag leads or route tickets to the right team.
How AI features are built today
Most products do not train their own model. They connect to a hosted model, such as those from OpenAI, Anthropic or Google, through an API. Your own data is supplied when needed, often through a technique called retrieval, so the model can answer using your content.
For narrow tasks such as image recognition on a phone, smaller models can run on the device itself, which helps with privacy and offline use.
Plan for the hard parts
- Accuracy. AI can be confidently wrong. Show sources, keep a person in the loop for important decisions and test with real examples.
- Privacy. Decide what data may be sent to an AI provider, and check their data terms.
- Cost. Usage-based pricing grows with your users, so estimate it early and cache results where possible.
- Measurement. Define what success looks like, such as fewer support tickets or faster processing, before you build.
Start small
Pick one workflow, build a focused prototype, and test it with real users for a few weeks. You will learn more from that than from a long list of AI ideas.
Our AI and automation team can help you find the right first use case and build it properly.