iFood Implements AI Area to Enhance Client and Restaurants Experience Using AWS

Over a year ago, iFood, a leading player in the Latin American food delivery market, decided that it was time to invest in artificial intelligence (AI) in order to be more assertive in restaurants and consumers service. To support the new structure, as well as its demand for processing high data volumes, the company relies on the solutions provided by Amazon Web Services, which offer the flexibility and scalability required to provide information in real time on a number of the company's operations.

Founded in 2011, iFood, a foodtech leader in Latin America, operates in Mexico and Colombia. In its early years, iFood improved its operating model, following the evolution in the market and growing larger year by year.

Initially, iFood operated as an integrated network for fulfilling online orders from its member restaurants. As time went on, the company started developing its own technology. For example, 18 months ago iFood developed its own fleet by having delivery drivers register with the platform; today, those drivers account for a significant part of the deliveries made by iFood.

Nowadays, more than 20 percent of orders placed use the entire iFood sales platform, from its marketplace to its CRM to its delivery drivers. “There were already AI initiatives in the company, but they were not centralized, and were used basically to answer to questions on business areas,” recalls Sandor Caetano, chief data scientist at iFood, pointing out that the company already relied on a data structure.

Over a year ago, iFood decided to create its Artificial Intelligence Academy, focusing on the research and development of machine learning, deep learning, logistical efficiency and other areas related to the company’s data structure. The initiative is the result of an investment of 20 million USD, as part of an contribution made by shareholders in 2019.

“We created this area to meet the need for growth, to rise and meet our competition,” explains Caetano, recalling that the goal was to count on a data structure that would take into consideration the existing AI team and stand to serve the decisions made by the company board. “That would change the way machine learning was handled, since it would start to be used for the automation of decisions, providing plain answers where an excess of data was on hand,” he said.

In order to perform the developments, make the necessary tests, and put them to work, iFood would need a robust IT infrastructure that could scale to meet peak demand. iFood determined it best to search for a company that would offer such infrastructure in the cloud, enabling a pay-per-use model and providing the necessary flexibility and scalability to support its work.

iFood already used Amazon Web Services (AWS) solutions like storage and managed databases, among others. As a result, expanding its scope of work with AWS was a natural choice. “The structure offered by AWS allows us to train several machine learning models the way we need and, more than that, to put these models into production in a simple way, without needing to start from scratch. Moreover, we rely on tools that make the data scientists’ lives much easier, making them as productive as possible,” says Caetano.

These are some of the main solutions used: Amazon SageMaker, which holds all the model testing infrastructure; Amazon Kinesis, which makes streaming data collection, processing and analysis easier in real time; and Amazon Elastic Kubernetes Service (Amazon EKS) to keep the models running.

The AWS infrastructure enables iFood's Artificial Intelligence Academy to expand its use of algorithms. One of the first business areas to benefit from this was Logistics. “We need to tell the client how long it will take for his meal to be delivered,” says Cateano. To this end, iFood currently relies on a route simulator, on which it analyzes different operation parameters according to week days and times.

Simulations are carried out in an AWS testing environment and, once proved, they are put into production. “The use of such algorithms sped up the process of discovering new parameters and, thus, our Logistics business area is far more efficient today. We deliver more today than two years ago,” notes Caetano.

iFood takes the same approach for developing recommendation lists for customers. The iFood app recommends restaurants and dishes according to a user’s taste by means of ML models processed on AWS. “All this goes through these models. Today we are able to control what is going to be shown in the app, including strategies and promotions, thanks to these algorithms. Such a level of service customization is possible thanks to Amazon SageMaker, which processes all of our models,” says Caetano.

Today, all models tested by iFood's Artificial Intelligence Academy go into production with a benchmark already in place. As a result, the company has been seen demonstrable gains in terms of productivity and improvements in its level of service. According to Caetano, since the algorithms started to be used, performance on the delivery SLA increased from 80 percent to 95 percent.

“When it comes to recommendations, there was also a significant improvement in conversions,” says Caetano, in that today it is possible for iFood to optimize lists so that they suggest restaurants near customers’ houses, which in turn optimizes the deliveries as well. In the Logistics business area, the distance traveled by delivery drivers was reduced by 12 percent thanks to route optimization. Similarly, delivery driver idle time was reduced by 50 percent.

iFood has grown to reach more than 1,000 cities across Latin America, with a platform comprised of more than 220,000 restaurants and 170,000 registered delivery drivers. On average, iFood delivers more than 39 million orders per month.  

Already in operation for more than a year, iFood’s Artificial Intelligence Academy will continue to expand its scope. One of those applications uses AI in order to improve the images of dishes that the restaurants display in the application. In another application, AI will be used to aggregate data to the description of such images, enabling customers to identify the ingredients of the dishes, for example.

Get more information on AWS artificial intelligence solutions.