You may have forgotten, but just ten years ago, to order a taxi, you had to either catch a car with checkers on the street, or dial the number of the control room, and then wait for an unknown amount and pay the same amount after a trip with a mysterious stranger at the wheel. Of course, there was a certain charm in this, but in terms of safety, reliability and price, analog reality is categorically inferior to digital aggregators.
Now the whole call process takes no more than five minutes, and you know in advance the make of the car, its color, numbers, the driver’s name and his rating. Yes, it happens that applications get stupid, taxi drivers come across different ones, and sometimes it takes a little longer to travel than planned, but fast movement in space in an unknown way has ceased to be a luxury and has become a daily routine. We owe the sudden comfort to machine learning and algorithms, and of course, to the people behind them. Now we’ll tell you how everything works.
Hexagons instead of districts
After you order a car, the aggregator starts looking for a driver inside your hexagon. If an ordinary resident sees a city as a combination of districts, administrative districts and other territorial units, aggregators look at the world with a grid of regular geometric shapes that cover the map. Why a hexagon? Because, unlike, for example, a square, in it the distance from the center to each neighboring cell is the same. Aggregators consider each hexagon as a separate information unit. Inside it, they set their own prices and look for drivers for the user who ordered a taxi in this cell. In the analytics of geographic units, aggregators are assisted by the H3 open source library developed by Uber.
Special points on the map are considered separately – for example, airports. It makes no sense to embed them in the structure of hexagons. Arriving with a passenger at the airport, the driver is free or for a certain amount – here it depends on the aggregator, he queues up to order.
Prices
Everyone is annoyed by the increase in prices, especially when it is raining and getting home without a taxi is very problematic. But as practice and calculations show, the absence of such regulation would lead to a collapse of transport in large cities.
It is important to understand that aggregators are not only dispatching services, but also full-fledged market regulators. If you do not influence prices, then with an increase in demand, a situation may arise called wild goose chase (chasing a wild goose, an English idiom meaning a useless and long search with an unjustified purpose). Drivers run out in the high demand zone – at this moment, orders are distributed among drivers from neighboring or more distant zones. The time that drivers spend on the way to the passenger is not paid. Therefore, in a state of wild goose chase, taxi drivers can travel the whole city for several orders, and passengers simply cannot wait for their trip. If the situation repeats, this will alienate both drivers and users from aggregators. In particular, drivers, since refusal of orders entails a decrease in the rating in the system. Therefore, one of the tasks that all aggregators seek to solve is the dynamic alignment of supply and demand. This is done with the help of high demand coefficients.
Example
Let k be the demand coefficient and the base price of a trip along the route – 200 rubles.
If k = 1 (normal demand), the price for a trip will be k x 200 = 200 rubles. At this price, conventionally, eight users will search for a ride, but there will be only three available cars.
With k = 1.5 (increased demand), the price for a trip will be k x 200 = 300 rubles. This price will scare three passengers away, but attract two drivers from other zones. Thus, supply and demand will equalize.
The price of a trip is calculated by multiplying a certain base price by the coefficient of high demand. This ratio is calculated both using machine learning, the task of which is to approximate supply and demand to each other, and with the help of a team of analysts. These specialists are well aware of the market, the impact of holidays and other factors on taxi operations.
Uber, as a pioneer among aggregators, instead of multiplying the base price, uses the added amount, which allows for more flexibility in changing the price depending on the conditions.
Example
Let s be the additional amount, and the base price of a trip along the route is 200 rubles.
If s = 0, the price for a trip will be 200 + s = 200 rubles. At this price, conventionally, eight users will search for a ride, but there will be only three available cars.
With s = 150 (increased demand), the price per trip will be 200 + s = 350 rubles. This price will scare three passengers away, but attract two drivers from other zones. Thus, supply and demand will equalize.
To equalize prices depending on the weather, Yandex.Taxi uses the services of its ecosystem, in particular Yandex.Pogod. Citymobil uses machine learning based on data obtained in the corresponding hexagons under various weather conditions.
How long to go?
As for traffic jams, Yandex.Taxi uses its own maps to monitor the current state of traffic. Competitors have to improve their machine learning techniques to predict and create a rough picture of road load along the entire journey.
Aggregators call the switchback method to test new algorithms. Since each hexagon is unique, the usual division into control and treatment groups cannot be used. The essence of the switchback method is to randomly transfer hexagons from the control group to the experimental group and back at fixed intervals. The data obtained in one hexagon for the entire duration of the experiment is divided into data using the tested algorithm and data without it. This allows you to make a decision on the need to implement a new algorithm in a specific hexagon.
Future
Nowadays, Uber allows you to book not just rides, but helicopter flights. Other companies are actively following this, and Yandex.Taxi is even preparing to introduce something similar.
It seems that in the future aggregators are planning to enter our life more and more: multimodal combinations with other transport (one price per train + taxi), displaying advertisements in the showroom in exchange for discounts, integrating services into retail, mobile communications and banking. When we finally see unmanned vehicles, no one can say for sure, but development is underway in this area in the most active way.
There are also more pressing problems that still have to be solved at the level of algorithms by modern aggregators. For example, services are not yet able to predict and take into account the presence of their competitors’ cars in the hexagons in order to adapt to this data, to better distribute drivers and set prices. This means that the field for future developments is still vast, and the number of vacancies for data analysts will only increase.
Comments