Many of our clients have a successful app that's a few years old. They don't need a rebuild, but they're being asked to "add AI". The good news is that some of the most valuable AI features can be added to an existing app in weeks.
Here are six we see pay off most often, roughly ordered from easiest to most involved.
1. Search that understands what people mean
Keyword search fails when users don't use your exact wording. Semantic search matches by meaning, so "something for a sore throat" finds the right products and "change my plan" finds the right page. It usually runs as a back-end change, with no app release needed.
2. An in-app assistant grounded in your content
An assistant that answers from your FAQs, policies and product information, and hands off to a person when it should. It works best when it can also look up the signed-in user's own information, such as their booking, bill or order status, within their permissions. Our guide to AI agents covers how to scope this.
3. Personalised recommendations
Reorder the home screen, products or content based on each user's behaviour. Even simple models trained on your own usage data beat a one-size-fits-all layout, and you can A/B test them safely.
4. Smarter forms with document and image capture
Let users photograph a document, such as a MyKad, receipt, invoice or meter reading, and fill in the form for them. On-device text recognition and validation cut drop-off in onboarding and claims flows. For identity checks, add face match and liveness, which is what our NexID platform does.
5. Summaries and drafts
Summarise long histories (a patient's past bookings, a customer's tickets) for staff-facing apps, or help users draft messages and reviews. These features are cheap to build, and users notice the time they save.
6. Proactive nudges
Predict what a user is likely to need next, such as a refill, a renewal or a booking, and send a well-timed reminder. This needs good event data and careful tuning. Done badly it's spam. Done well it brings in real revenue.
On-device or in the cloud?
| Approach | Best for | Trade-offs |
|---|---|---|
| On-device models (Core ML, TensorFlow Lite, platform AI APIs) | Camera features, text recognition, privacy-sensitive data, offline use | Smaller models; app size and older devices to consider |
| Cloud models via your back end | Assistants, search, summaries, recommendations | Latency and running costs; data leaves the device, so plan for PDPA |
| Hybrid | Most production apps | More moving parts, but the best user experience |
One rule we always follow: never call an AI provider directly from the app with an embedded API key. Route calls through your own back end, so you control authentication, rate limits, logging and cost.
Shipping it safely
- Release behind a feature flag to a small percentage of users first.
- Measure against a clear metric, such as search success, support deflection or conversion.
- Design for AI being wrong: show sources, make corrections easy and keep a human fallback.
- Watch costs per active user from day one.
Have an app that could do more?
We've built and maintained mobile apps since 2006, and now add AI to them. Send us your app and we'll suggest the two or three AI features with the best return. Request an AI app review →
