Yulu Uses a Data Lake on AWS to Pedal a Change
Yulu improved service efficiency by 30–35% using its prediction model and AWS data lake.
Traffic congestion and air pollution are serious issues in India, particularly in megalopolises such as Bengaluru. Yulu’s mission is to address such challenges by providing sustainable and hassle-free micro-mobility solutions for commuters travelling short distances. Launched in December 2017, Yulu provides a network of over 10,000 shared vehicles, which include Yulu Move (smart bicycles) and Yulu Miracle (smart light-weight electric scooters), in Bengaluru, Pune, Mumbai, and Bhubaneswar. These vehicles can be easily rented with a user-friendly mobile app on a pay-per-use basis. But smart bicycles and electric scooters are just the beginning—various form factors catering to multiple use cases and infrastructure across India are on the horizon for Yulu’s ambitious management team.
Platform of Choice
Yulu is a data-driven organization, streaming data from users’ mobile phones and using bikes as the Internet of Things (IoT) devices. A team of over 50 operators manage the fleet to ensure bikes are well-positioned in high-demand areas—a key activity Yulu calls “Rebalancing.” Profits are derived from a high utilization ratio, meaning how many times a bike is ridden in a day and how many total bikes are in service that day.
The startup selected Amazon Web Services (AWS) to launch its business for several reasons. As with most startups, speed-to-market and low upfront costs were top priorities. When recruiting staff, CTO and Cofounder Naveen Dachuri found that in the Indian market more engineers had experience with AWS than any other cloud platform, so the learning curve would be less steep. This would also allow him to launch the application quickly without heavy investments in training time or cost. Yulu did consider Google Cloud but found that AWS’ 10GBps network with dedicated fiber was much better than Google’s offer at the time. The existence of an AWS data center in Mumbai was another compelling factor, as data residency requirements in India were becoming more stringent.
“The biggest benefit of being on AWS is that my team completely focuses on application development and spends more time coming up with new features.”
Says Naveen Dachuri CTO and Cofounder, Yulu Bikes
Data Pool to Data Lake
Dachuri has more than 18 years’ experience managing databases and was careful to select the right product to fit his business needs. He chose Amazon Relational Database Service (Amazon RDS) with Amazon Aurora for its speed and demand-based scalability, particularly for real-time instances. Dachuri’s team relies on such instances to feed real-time data to its proprietary prediction model, so operators can rebalance the location of its bikes when demand spikes in certain areas or at certain times. “With Amazon RDS, it was simple to move from smaller to larger instances based on the type of data we were getting,” Dachuri says. Yulu uses Amazon EC2 M5 instances, which Dachuri says are well suited to analytics and IoT environments because of their storage capacity and flexibility.
Yulu spent the first six months of operations collecting data to understand usage patterns. It then began constructing its prediction model using Amazon EMR for deeper analysis. “Amazon EMR gives us a seamless integration to move our data from our transaction system to Yulu Cloud – our data lake, which runs on Amazon Simple Storage Service (Amazon S3),” Dachuri says. “We can now proactively manage our vehicles, so they are always in great condition and act quickly on vehicles that move outside our operational zone to bring them back to high demand areas.”
In the four months since launching its prediction model, Yulu has seen excellent results. “The accuracy is increasing day by day,” Dachuri explains. “The interesting factor is that even on day one when we launched the model, our rebalancing efficiency increased by close to 30 or 35 percent, which is a very significant number.”
The model is used in all four cities of operations, and Yulu is fine tuning it to work in new cities as the company expands across India. The model is at the heart of Yulu’s business and will help the company not only grow its bike market, but also add customers when it introduces electric scooters and other vehicles. The company started with just 500 bicycles and is now approaching 7,000 bicycles and electric scooters across the four cities.
Lean Teams Need Support
Yulu can maintain a lean operation despite a rapidly growing user base. Though the IT team has increased from 2 to 15 people, they have no one dedicated to infrastructure, DevOps, or database management—and do not plan to change. Only 2–3 percent of its engineers’ time is spent on databases, a task that used to absorb at least one dedicated employee in Dachuri’s past work environments.
Monitoring tools such as Amazon CloudWatch and 24/7 access to AWS Business Support enable this business model. Support tickets are usually resolved in 1 to 2 hours, Dachuri confirms. He comments that while most questions can be resolved by referencing online documentation, the ease and speed of calling someone at AWS on the phone are invaluable in fast-paced environments.
The team frequently refers to live tutorials and similar documentation when using a new AWS service such as Amazon Cognito. Yulu uses this service to authenticate and authorize users, securely store customer data, and generate verification codes. One-time passwords are then sent to riders via SMS using Amazon Simple Notification Service (Amazon SNS).
Time to Innovate
The Yulu app has improved greatly over time, as the IT team can easily deploy updates. They test and push new features in one to two weeks. “Without the AWS Cloud, that would take at least six to seven weeks,” Dachuri explains. “The biggest benefit of being on AWS is that my team completely focuses on application development and spends more time coming up with new features,” he adds.
In addition to development, Yulu’s engineers use their time to read up on and explore new features. “My analytics team wanted to use [Apache] Kafka but figured out there’s already a service called Amazon Kinesis, which they have started exploring. If this meets our requirements, we will start using it right away,” Dachuri says. “With AWS overall, we get ease of use, fast time-to- market, reduced risk, stability, and, of course, cost savings because you pay for what you use.”