How Grab uses AI to generate more precise delivery instructions

Grab Geo Team: Ameya Mannikar, Echo Zhangjie, Gwen Dang, Liu Ye Feng, Shi Qiu, Shirley Yang

. January 29, 2024

If you frequently use Grab, you’ve probably done this yourself: Type additional instructions in the GrabFood app to help your delivery-partner find the drop-off point faster. Your note might include crucial information, such as preferred entrances, specific parking spots, nearby landmarks, or unit number.

In fact, we saw that 84 per cent of deliveries were accompanied by notes from consumers—a sign that our delivery-partners often require more information to navigate to the destination.

So we thought about how we could make use of these notes to improve our in-app navigation system so that delivery-partners can get to their drop-off points more efficiently.

Improving Grab’s delivery time

By combining point-of-interest (POI) data and consumer notes, we were able to provide more precise delivery instructions to drivers when they’re about to reach the drop-off point.

Now, when a delivery-partner is approaching a destination, an expanded pin within the Grab Navigation app displays additional information that will help them locate their drop-off point. This is accompanied by images from our crowdsourced map data.

Delivery-partners can expect to see a combination of information including nearby landmarks, a specific meeting point or the unit number of an apartment.

The feature is currently available in selected cities across the region. We plan to roll this out to more cities this year.

AI to sift through consumer-generated notes

Previously, our POI data only went as far as to cover specific locations and sites, leaving out an additional layer of information that would be helpful to delivery-partners during their commute to a drop-off point. For example, which entrance should they use? What are some notable landmarks in the vicinity?

The notes left by Grab users, on the other hand, were rich in nuanced details—information we wouldn’t have been able to capture without being there in-person. These notes provide additional contextual information for delivery-partners.

But processing the notes manually is tedious given the sheer volume and variety of data. Consumers may also communicate the same information in different ways. To process the data, we used an AI model that could distil wayfinding instructions from consumer-generated notes.

For example, the model is able to extract information such as building names and the colour of landmarks from a consumer’s note.

Thanks to these notes, we were able to provide delivery-partners with more precise instructions for them to complete their deliveries efficiently. This could potentially translate to higher overall earnings.

Consumers, too, can benefit from such productivity gains. Apart from getting their food on time, we can better optimise our existing pool of drivers, and make Grab’s services more available and affordable.

Deliveries and rides