Here’s something you do often in Dubai: Open a ride-hailing app, enter where you want to go and watch a handful of cars moving around the map. One looks particularly close, so naturally you assume that is the driver you’re going to get. Then the app assigns someone farther away, and for a moment it feels slightly ridiculous.
The thing is, the car closest to you may not actually be the closest in any useful sense. At a recent Yango Digital City session in Dubai, I got a look at some of the technology sitting behind interactions like this, and what appears on our phones as a very simple transaction is actually the final step in a much larger system constantly looking at roads, traffic, historical behaviour, supply, demand and, increasingly, what it thinks is likely to happen next.
That last part is what stayed with me after the session. The smartest digital services in a city are no longer just responding to what we’re doing. They are trying to get there first.
Why doesn’t the app always send the closest driver?
One example from Yango neatly explained something I’ve wondered about plenty of times. Imagine three drivers are available nearby: one is only 250 metres away but sits on the opposite side of a divided road, requiring a U-turn and a traffic-light cycle before reaching you. Another is 650 metres away, finishing a trip and already moving toward your location. A third is farther away again.
Pure distance would select the first driver, but the algorithm might select the second. According to Yango, its matching system considers direction of travel, road layout, traffic conditions, whether a driver is finishing another trip, and the actual time required to reach the passenger. That sounds obvious once someone explains it, but it also illustrates how much complexity gets hidden underneath an interface intentionally designed to feel simple.
The blue dot isn’t the difficult part. Working out which moving car can reach another moving or stationary person fastest, while thousands of other people are simultaneously asking the same question, is. And that is before the system starts trying to predict demand rather than simply reacting to it.
Can an algorithm predict when an entire city will want a taxi?
Cities develop patterns. People leave residential areas in the morning, business districts become busier during working hours, entertainment areas change character later at night, weekends behave differently to weekdays, and major events can briefly rewrite the normal rules entirely. Yango described this as understanding the rhythm of a city.
By combining historical demand with real-time information, its systems can identify where ride requests typically occur at specific times and on specific days. The company showed examples from cities where demand repeatedly shifted geographically through the day, with recognisable patterns appearing most of the time. That information can then be used to move supply ahead of the requests.
New Year’s Eve is an easy example. A city may normally be relatively quiet between 1am and 6am, but historical data tells the system that New Year’s morning behaves very differently. Instead of waiting for a flood of ride requests and then discovering there aren’t enough cars nearby, the platform can try to position more drivers around anticipated hotspots. Planned events add another layer because a concert, football match, or exhibition can create demand in a place that would ordinarily be quiet at that hour.
What we’re really talking about, then, isn’t finding a taxi. It is predicting where hundreds of people are likely to need one before many of them have even opened the app.
Why does surge pricing suddenly appear?
This also leads directly to one of ride-hailing’s least loved features: surge pricing. Yango’s explanation of dynamic pricing was refreshingly straightforward. If 1,000 people want a ride and only 500 drivers are available, no software trick can instantly create another 500 cars, so the system has a capacity problem.
Increasing the fare is partly intended to reduce that demand. Someone who absolutely needs to reach the airport might accept the higher price, while someone whose journey can wait may look at the fare and decide to try again in 10 or 15 minutes. That reduces the number of simultaneous requests while potentially encouraging additional drivers to move into areas with high demand.
It doesn’t make paying twice the normal fare any more delightful, but it does explain why dynamic pricing exists beyond the wonderfully convenient explanation that a company simply enjoys charging us more. Forecasting tries to prevent the imbalance before it happens, driver positioning tries to move capacity toward demand, and pricing becomes one of the controls available when those two things still aren’t enough.
Is AI actually doing all of this?
AI was inevitably mentioned throughout the session, although one distinction was worth making. Much of what Yango described wasn’t generative AI in the ChatGPT sense. It was machine learning, forecasting, routing and recommendation systems that have been quietly operating inside digital services for years.
That is arguably what makes this more interesting. The current AI conversation tends to focus heavily on chatbots answering questions or writing text, while another form of AI is already making millions of small decisions around us without asking us to type a prompt: which road is fastest, which driver should receive this passenger, where demand will appear in half an hour, how long a bus will take to reach its next stop, or which restaurant should appear at the top of someone’s screen.
None of those interactions feels particularly futuristic anymore, and that’s precisely the point. The technology disappears into the experience.
How much does an app learn about your routine?
This is where the conversation moved beyond mobility and became considerably more thought-provoking. Yango operates across ride-hailing, delivery, food, entertainment, and other digital services, and its presentation described how interactions across these services can serve as signals for personalisation.
Some examples are familiar. If you regularly order vegetarian food, the platform can prioritise vegetarian options, and if people with similar behaviour tend to prefer particular restaurants or dishes, those patterns can help inform recommendations before the system knows very much about a new user. Routines provide an even stronger signal: if someone repeatedly opens a mobility app at the same time and travels to the same destination, the system can begin to predict that intention.
The same logic can extend elsewhere. Someone who regularly orders desserts late in the evening may see those recommendations move higher at that time, while repeated interactions with an assistant can become shortcuts or routines. None of that requires an algorithm to somehow understand you as a person; it just needs enough repeated signals.
That is perhaps the more important shift happening in everyday AI. We tend to imagine personalisation as an app learning what we like, but increasingly it is learning what we are likely to do next.
Why is Dubai particularly interesting for this technology?
Another point from the session that resonated is that digital cities can’t simply run identical software everywhere, and Dubai is a particularly good example. A 700-metre walk may be an entirely reasonable part of a journey in a European city; make the same assumption outdoors during a Dubai summer and your beautifully optimised route has developed a fairly obvious flaw.
Expectations differ too. Yango contrasted markets where a 40- or 50-minute delivery may be considered perfectly normal, but in the UAE, consumers have become accustomed to services promising groceries in minutes. Once people get used to that level of speed, their definition of an acceptable digital service changes with it.
That means algorithms can’t just understand roads. They need to understand how people actually use a particular city, including climate, public transport, road design and local expectations. Dubai’s rapid development makes that challenge even more pronounced as new residential areas emerge, business districts expand, road networks evolve, and transport demand shifts with them. The digital representation of the city has to keep catching up with the physical one.
What happens when this invisible layer stops working?
One of the best observations from the session came right at the beginning: nobody spends much time thinking about the electricity grid when the lights turn on — we notice it when they don’t. Digital infrastructure is beginning to work the same way.
We expect our location to appear instantly on a map, an ETA to be reasonably accurate, a payment to go through, a driver to find us, food to arrive and a navigation app to know that Sheikh Zayed Road has turned red again. When all of that works, we barely think about it. When GPS stops behaving properly, a map sends us somewhere absurd or a ride-hailing app can’t find a driver, the enormous amount of infrastructure hiding beneath those tiny icons suddenly becomes very visible.
Perhaps that’s the real definition of a digital city. It isn’t a collection of giant screens, autonomous robots and futuristic buildings; those are merely the bits that photograph well on influencer feeds. It is thousands of small systems quietly absorbing complexity so that the person using them doesn’t have to.
