AI and ML powering the next generation of personalisation for on-demand apps: Madan Thangavelu, Senior Director, Engineering, Uber

In a time when the modern smartphone user is becoming more discerning and demanding, the truest measure of excellence for any on-demand service company is the ability to solve real problems while being as assistive as possible to the user as they interact with the platform.

The on-demand service industry primarily relies on mobile apps, which have limited user interface (UI) space due to the small screen sizes of phones. This contrasts with desktop websites, which offer more extensive interaction surfaces. As the range of on-demand services expands, user experience (UX) faces challenges in presenting all available options in an easily accessible manner. Therefore, personalising the UX is crucial to keep the app relevant for each user and to minimise navigation effort.

At Uber, we utilise machine learning in crucial parts of the booking process, such as product selection. The machine learning system identifies the appropriate list of products to display to a user based on their usage history, current market conditions, and personal preferences. For instance, if a rider frequently uses Uber Auto, the system will prioritise showing Uber Auto in various app sections.


For a frequent Uber Auto user, alternatives like Uber Moto and Uber Go may also be suitable. In these instances, the system recommends these alternatives on different screens to inform the rider about the range of Uber’s offerings.

Tailoring the product recommendations to match rider preferences is vital. For example, if a rider uses Uber during office hours and values quick service, the app should highlight options that offer faster rides, even if they are more expensive. Conversely, for a casual trip on a Sunday evening, the rider might prefer to wait for a cheaper option. Presenting the appropriate ride options based on the rider’s context significantly enhances their experience with the Uber app. Personalisation is integral to this process.

The effectiveness of this personalisation system can be measured both directly and indirectly. Direct metrics like click-through rates on personalised components provide immediate feedback on the system's usefulness. Long-term metrics, such as the number of trips over several months, also indicate the system's impact.

Uber offers various products. Monitoring riders’ awareness of these products and their usage patterns helps evaluate our personalisation technology.

The machine learning algorithm must account for geographical preferences and differences. This involves training the models with datasets that include regional attributes. The model then adapts to weigh different features according to regional behaviours.

Regional customisations benefit users. For example, in Asia-Pacific, riders might prefer low-cost, high-capacity vehicles, whereas in the US, the preferred low-cost option might be ride-sharing with UberXShare. Considering cultural and regional preferences is crucial for recommending the right product to every rider.

Personalised systems tend to favour experiences a user already enjoys. However, advanced techniques are used to explore other potential interests. The app’s information architecture allows riders to view all available options beyond the recommended ones, ensuring they can access any product they choose.

An effective information system ensures that riders can easily find all services offered. This balance allows personalisation to tailor the app to individual preferences without hindering overall usability. When personalisation is precise, the experience becomes more enjoyable for every rider.

Contributed by Madan Thangavelu, Senior Director, Engineering, Uber