Feeding Brazil and delivering a brighter future

Data analysis bottleneck prevents timely optimizations to delivery routes

iFood isn’t just another food delivery service. The company strives to make the world a better place by partnering with local restaurants for mutual success, compensating and treating its drivers fairly and minimizing plastic consumption throughout its operations. To work toward these ambitious goals, iFood innovates constantly — which means its employees need easy access to business data.

“We’re committed to building a data-driven organization,” said Marcia Freitas, Group Product Manager at iFood. “We want all our business users to be able to run their own reports and generate their own insights to contribute to the success of our business. That means putting all our data in one place and giving them tools that make it easy to run analysis.”

iFood thrives on creating simulations and running scenarios that help its decision-makers validate their business hypotheses. For example, the company’s fleet optimization team moves the levers constantly to balance cost of delivery with quality of service. And the company will often test service policies in one geographic region before applying them to other regions. But until recently, visualizing the future meant asking the data team for assistance and waiting perhaps weeks to be able to run a scenario.

“There were so many requests for data analysis projects and we had only a small data team available to help,” recalled Leandro Braga, Senior Staff Data Scientist at iFood. “So everyone was used to submitting a ticket and waiting their turn, which could take weeks. This meant that any business plan took longer to put into place because we had to wait for validation before we could take action.”

Simulator powers smarter decisions by testing hypotheses in minutes

Seeking to eliminate bottlenecks in its analytical processes and drive faster decision-making, iFood decided to launch a simulator that would dramatically accelerate the process of running business scenarios. iFood chose Databricks Data Intelligence Platform and Tableau as the core technology of this platform.

When iFood first launched Databricks and Tableau, relatively few employees had access to the new systems to create data tables. After a few months, the company opened up access to the data lake and allowed many more business users to create their own data pipelines. iFood then ran a massive training program on Databricks and Tableau. Today, more than 2,000 employees use the solutions to generate their own data.

“The simulator we’ve launched with Databricks and Tableau is a breakthrough for our logistics team,” said Freitas. “We can test many hypotheses at once to get faster answers. We now have all the tools we need to run simulations and get immediate results.”

iFood’s fleet optimizer team works to assign the right delivery drivers to the right orders and clients. They used to test hypotheses by designing experiments, selecting regions for implementation, letting the experiments run for several weeks and then gathering data directly from the production server. Finally, the company would identify and implement a solution. Today, the simulator allows them to balance drivers, orders and routes to come up with solutions that satisfy customers while optimizing delivery costs and minimizing emissions.

“The simulator lets us run highly specific simulations in minutes,” said Braga. “By tapping into our Databricks data lake, we can get real data on the exact scenarios we want to test without impacting our production server. For example, we can use a week’s worth of real data from all our deliveries in Sao Paulo to simulate how those numbers would change with a few adjustments to our delivery routes.”

iFood’s logistics team publishes all its data within Databricks so that other teams can analyze it down to the level of an individual order or driver simulation. Because the company built a simple interface for the simulation program, stakeholders across different business units can analyze and publish data without needing sophisticated technical skills.

“Here in Logistics, we use Databricks for almost everything,” Freitas confirmed. “My product team keeps notebooks to help people audit data that our services have published. For example, we have a freight structure that determines how we pay our drivers. All the data updating for that function happens in Databricks.”

Meanwhile, iFood uses Tableau to send daily reports to stakeholders automatically through Slack or email. By eliminating barriers to data, iFood keeps everyone’s focus on achieving objectives. The company has about 1,000 direct users of Tableau and many more who receive reports.

“We have company-wide goals that we define for specific time periods, and we create dashboards so everyone can track progress,” explained Freitas. “Thanks to Tableau, all our decisions across iFood are now based on data. One of the first things we teach new team members is how to run queries and use notebooks on Databricks and how to build visualizations and dashboards on Tableau.”

Thousands of monthly simulations show the path to outstanding service

iFood now uses its simulator to run at least 3,000 simulations per month, with an all-time high of 10,000. The company’s decision-makers will typically form hundreds of hypotheses on how to address a particular business challenge before narrowing the list to about 10 — one for each of the company’s geographic regions. iFood continually tunes the parameters of the simulator for each region to generate the most realistic results.

“The simulations we run on Databricks are enabling us to find the ideal balance between improving our service levels and controlling our cost of delivery,” said Freitas. “But it’s not just about finding the answer — it’s about finding it quickly, because the more time we spend testing, the longer we have to wait to put the solution in place. With Databricks, if we come up with a hypothesis that could save us even a small amount across many millions of orders, we can test that change and put it in place quickly to maximize our savings. In the past, it would have taken us at least a month just to test a hypothesis.”

iFood continues to use the simulator to find new ways to optimize its business. “I’ve launched all kinds of projects in Databricks and Tableau,” Braga remarked. “Last year, we set out to reduce delivery delays of more than five minutes. We determined that it would cost us a little more to meet that goal, but we had all the insights we needed to achieve the right balance. Right now, we’re looking at ways to increase the hourly wages of our drivers to keep them happy. With all our data in one place on Databricks, we finally have the perspective we need to make the right decisions.”

As iFood continues to grow, the company plans to expand and refine its use of the simulator. “We’re getting bigger every year, and as we expand, our risks get bigger too,” Freitas concluded. “Every change we make to our processes will have ramifications for many millions of orders in terms of cost, customer satisfaction and driver satisfaction. We’ll keep innovating on Databricks and Tableau to find the solutions that enable us to continue to lead our market.”