TL;DR: AI in mobile apps has moved from visible chatbots to an invisible layer: personalization engines, on-device computer vision, biometric login, predictive interfaces and semantic search. We have shipped this kind of AI in production, from Shiri's recommendation engine to Savyon's on-device test reader. Most production apps combine on-device models for speed and privacy with cloud models for heavier work.
The AI features users actually use
A decade ago, adding AI to an app meant integrating a basic chatbot. Today, AI capabilities are woven into every layer of the app experience — and users don't just tolerate them; they expect them.
The shift happened because AI moved from a visible feature ("chat with our bot") to an invisible infrastructure layer that makes the entire app smarter.
Where AI adds real value in mobile apps
Personalization engines
The most impactful AI feature in modern apps is invisible. Users don't see "AI" anywhere — they just notice the app seems to know what they want. Product recommendations, content ordering, and adaptive learning paths all run on ML models that observe behavior and adjust.
We built this kind of engine for Shiri, Israel's national music streaming app. Its recommendation engine uses collaborative filtering trained on Israeli listening habits, so local artists stay visible. The app reached 600,000 users and #1 in both app stores on launch day. In IBI Smart, machine learning personalizes flows and content by trading behavior: new investors get guided onboarding, and experienced traders get advanced data. Users with AI-personalized dashboards traded 3x more frequently than users with static interfaces.
On-device computer vision
The phone's camera is now a sensor that AI can read. For Savyon Diagnostics, we built a computer vision SDK that reads COVID-19, pregnancy and ovulation tests inside major Israeli hospital and HMO apps. The model runs on the phone, so results appear in seconds, work without internet, and no data leaves the device during scanning. The same approach powers our AI traffic-management system, which detects vehicles up to 700 meters from an intersection with 99% accuracy using existing city cameras.
Biometric authentication
Face recognition, fingerprint scanning, and voice verification have replaced passwords. The technology runs locally on-device, making it both fast and private. For enterprise apps handling sensitive data, biometric authentication is a baseline expectation.
Predictive interfaces
Smart apps anticipate user needs. A navigation app pre-loads your commute route. A banking app surfaces your most-used payment recipient. These features reduce friction by eliminating repetitive actions.
Intelligent search
Users type natural language queries and get results that match their meaning, beyond exact keywords. We've implemented semantic search in enterprise apps where natural language search reduced average search time from 4 minutes to 30 seconds.
