Business challenge

To maximize usage and user satisfaction, the operator of one the world’s largest bike-sharing system wanted a smarter way to distribute 13,000 bicycles across more than 800 stations and enable tens of thousands of journeys per day.


DecisionBrain used IBM® Decision Optimization to calculate the optimal number of bikes for each station at any given time, and plan efficient routes to help maintenance teams redistribute bikes accordingly.



optimal bike inventory, distribution and maintenance problems within seconds


operational costs and improves performance on service level metrics


number of journeys per month since the launch of the bike-sharing system

Business challenge story

Back in the saddle

Bicycle-sharing systems have become a common convenience in many major cities across the world, giving tourists and commuters a quick, convenient, affordable and eco-friendly mode of transport for shorter journeys. Riders can take a vacant bicycle from one of the hundreds of docking stations distributed around the city, and dock it back in at another station once it is no longer needed.

Have you ever wondered how cycle sharing organizers make sure there’s a bike waiting for you where you need to take it, and a free dock to deposit it at the end of your journey? It’s much easier said than done. If a docking station runs out of bikes, users may find themselves stranded. Or if a docking station is completely full, it leaves no spaces for the bikes to be returned, so users must cycle around until they can find another nearby station with vacant docks. In both cases, the inconvenience may make them less likely to use the service again.

Filippo Focacci, CEO of DecisionBrain, explains: “A common use-case is when customers take a bike from outside a train station to cycle the last leg of their commute to work. They deposit the bike at a docking station close to their office, and then cycle back to the train station at the end of the day. But these usage patterns are not always consistent, because there are many other factors in play.

“For example, if it rains in the morning, many commuters will opt for the bus, so most bikes won’t move from the train station. But if it’s dry in the evening and they want to cycle home, they will still expect a bike to be waiting for them in the dock by the office. To complicate matters further, other users such as tourists and off-peak travelers also move bikes around the city at different times, which makes it difficult to predict usage patterns precisely.”

Circulating cycles

In many cases, the contract for a bike sharing system will assign penalties to the operator if too many stations are left full or empty. Cycle hire companies therefore rely on small teams of field-workers to travel around the city in trucks throughout the day, redistributing bikes between docking stations. But it is often difficult to predict which stations are most likely to be full or empty, or where and when someone will need a bike next.

To make optimal use of a limited number of bikes, docking station spaces, field-workers and trucks, operators in a control room are constantly trying to work out where bikes are most likely to be needed, the optimum number of bikes to keep in each station, and the most efficient way for distribution teams to move these bikes around.

Most companies can only do this reactively, making rough estimates based on current demand, and resolving problems as they happen. Without the ability to predict future needs, this can be a real uphill struggle.

This was the case for the operator of one of the world’s largest bike-sharing schemes, which uses only 15 distribution trucks to administer 815 docking stations and 13,000 bikes. To help it create a sophisticated, predictive, data-powered solution to solve these complex inventory and redistribution problems, the operator turned to decision support solutions provider DecisionBrain.

Decision Optimization provides the core engine for our optimization tools – as the market leader for this type of solution, IBM was the obvious choice.

Filippo Focacci, CEO, DecisionBrain

Transformation story

Finding the perfect route

The DecisionBrain team began to design a solution based on machine learning and optimization techniques, using a combination of SparkMLlib and IBM® Decision Optimization software.

Focacci adds: “IBM ILOG® CPLEX® Optimization Studio provides the core engine for our optimization tools – as the market leader for this type of solution, IBM was the obvious choice. Working with IBM products is always a pleasure, and we have a long and beneficial partnership with the IBM team, both as a partner and a provider.”

To solve the key problems, DecisionBrain created several predictive models with SparkMLlib, using techniques such as simple and exponential moving averages and random forests. The predictive models forecast the number of bike pick-ups and drop-offs of each station throughout the day. The model considers circumstantial data, such as performance data, historical bike demand, historical weather data and current forecasts, to make a unique prediction for every day.

The IBM Decision Optimization solution is then used to run several optimization models to find optimal inventory and distribution strategies. The first optimization model uses a mixed integer programming (MIP) technique to compute the optimal inventory of bikes for each station at any given time.

The second model uses a combination of MIP and heuristics to solve the redistribution problem—deciding which bikes should be moved from which stations to achieve the desired inventory at each station, and calculating the most efficient way for distribution vehicles to transport the bikes across the city to these optimal locations, moving as few bikes as possible, with minimal travel time, taking peak traffic into account.

After training the solution on two years of historical data, the tool was ready to provide real-time predictions for where bikes should be circulated, and how to get them there, every day. The solution runs online and is available 24/7.

With IBM Decision Optimization, we can find efficient solutions almost instantaneously.

Filippo Focacci, CEO, DecisionBrain

Results story

An easy ride

Since deploying the machine learning and optimization solution from DecisionBrain, the cycle hire scheme operator enjoys rapid, sophisticated predictions and recommendations, which helps its distribution teams increase the availability of bikes and docking spaces at each station.

“With the IBM Decision Optimization solution, we can find efficient solutions almost instantaneously,” Focacci says. “We can solve the inventory problem for the entire 800 station network in 12 seconds, find the best way to redistribute bikes in 70 seconds. This empowers the operator with the right information and analysis to reduce the incidence of full and empty stations, keeping bikes available at all times.”

The increased efficiency enabled by the solution has allowed the distribution company to meet its targets around cycle availability, as well as improving service and increasing end-user satisfaction.

“After the implementation, our customer was able to significantly reduce the number of full and empty docking stations,” Focacci remarks. “After we introduced the solution, usage of the cycle hire scheme reached a record high for two months in a row.”

After the initial launch of the solution, DecisionBrain also helped the operator build a third model to calculate the optimal time for each bike to be taken in for maintenance and repair, and the best route for repair trucks to collect them. Building this new dimension into the wider optimization model is helping the operations teams manage their workload even more efficiently, and ensuring that the city’s bikes are in better condition than ever.

The DecisionBrain solution is the first application of its kind that uses both optimization and machine learning to solve cycle hire inventory, distribution and maintenance problems, and could easily be re-deployed for other cycle sharing systems around the world. As an official IBM Business Partner, DecisionBrain will continue to utilize IBM products to help clients solve complex optimization problems.

Focacci concludes: “Over the years, we have built a strong relationship with IBM, and I know I can trust IBM products to help us bring the best decision-making abilities to our customers.”


Headquartered in Paris, DecisionBrain helps companies improve their supply chain and business operations with innovative optimization solutions, as well as consulting services and training.

Solution components

Take the next step

IBM Analytics offers one of the world's deepest and broadest analytics platform, domain and industry solutions that deliver new value to businesses, governments and individuals. For more information about how IBM Analytics helps to transform industries and professions with data, visit Follow us on Twitter at @IBMAnalytics, on our blog at and join the conversation #IBMAnalytics.

To learn more about IBM Decision Optimization solutions, please contact your IBM representative or IBM Business Partner, or visit

View more client stories or learn more about IBM Analytics