We’ll remind you to book a ride in advance whenever we predict low ride availability

To do that, our data science team built something called the Fulfillment Rate Forecasting system. It quietly works in the background, predicting where passengers might have trouble getting a ride in the next few hours. And if the system spots a potential pinch point, for example, very few drivers near your drop-off location, it will send you a notification suggesting an advance booking. Think of it as a heads-up from Grab, so you’re not left waiting around.

Looking into the near future

At the heart of the system is a simple idea: understanding how reliably we can match passengers and drivers. We measure this through something called the fulfillment rate: basically, how many requests successfully get a driver and complete their ride. If we think the rate in a certain area might dip soon, that’s a sign we should warn you early.

The forecasting models look at what’s happening in real time and what has happened recently. They study patterns in how many people are requesting rides, how many drivers are around, what time of day it is, what day of the week it is, and how each city behaves differently. All of this lets Grab “peek” up to six hours ahead, updating the view every twenty minutes for specific areas across a city.

Making it work at Grab scale

Building this wasn’t straightforward. We needed to make sure it could run reliably across Southeast Asia, where traffic patterns shift quickly and every city has its own rhythm. As we expanded the system to more places, we initially overloaded some internal tools with too many requests. The teams solved this by redesigning how the forecasts are handled behind the scenes, making them faster and more efficient.

We also learned something surprising along the way: giving the model less historical data sometimes makes it smarter. In a fast-moving marketplace like ride-hailing, yesterday often matters much more than last week. When we shortened the model’s “memory” from seven days to just one, predictions actually became more accurate, improving overall performance and helping more users take action when a forecasted shortage appeared.

What’s best: Our learnings from solving the fulfillment rate forecasting use case actually triggered a wave of innovation here at Grab.

We realised that many teams within Grab could benefit from applying advanced forecasting models. With the help of Grab’s platform engineers, we packaged our forecasting process into a toolbox called Spyce. Other teams can deploy this to significantly speed up their forecasting capabilities.

[Also read: We’re making the power of forecasting available to more Grabbers]

What’s coming next

The fulfillment rate forecasting system is still evolving. Soon, it will get better at understanding the impact of major holidays and big local events, which can dramatically change how many people need rides. We’re also working on making the forecasts more fine-grained, so they reflect what’s happening on the ground at a highly localised level, not just across larger zones. And in the future, we plan to connect city-level and neighborhood-level predictions for even greater accuracy.

The goal is simple: to help you get where you need to go, without the stress of wondering whether a ride will be available when it’s time to head home. That little notification you sometimes get—the one suggesting an advance booking? That’s the forecasting system watching your back.