Supr Daily Improves Partner and Customer Experiences Using Machine Learning on AWS

Based in Bangalore, Supr Daily is a grocery ordering and delivery service that makes it convenient for customers to order fresh groceries to their homes. More than 200,000 customers use Supr Daily to order milk, eggs, fresh fruits and vegetables, and other groceries for delivery by 7:00 a.m. each day. But to support its network of thousands of delivery partners across six cities and manage its inventory of thousands of products, Supr Daily needed to scale its order delivery verification and inventory planning systems. At the same time, the company’s growth accelerated as restrictions due to the COVID-19 pandemic limited in-person shopping, contributing to a 70 percent surge in new users.

To support this growth and make its inventory and delivery systems scalable, Supr Daily built new solutions on Amazon Web Services (AWS). Using a suite of AWS services, including Amazon Rekognition, which automates image analysis with machine learning (ML), Supr Daily made the delivery verification process faster, inventory management simpler, and its overall architecture more scalable to support continued growth.

Supr Daily is a grocery delivery startup that was founded in 2015. A subsidiary of online food ordering and delivery company Swiggy, it started as a milk delivery company but has since expanded to feature thousands of items that it delivers in the morning to customers in six cities. With every Supr Daily order, the delivery partner uploads a photo as proof that the delivery was made correctly. These images must be high quality for customer support needs and to build customer confidence. Sometimes, though, delivery partners submit photos that are blurry or taken from a poor angle. This could delay the processing of refunds and lead to unnecessary or fraudulent refund claims. The company estimates that as many as 25 percent of refunds were issued incorrectly, primarily due to poor quality or missing delivery photos. However, verifying images was too resource-intensive to do manually. “We could only examine 5–10 percent of images, and doing even that was difficult and time consuming,” says Praveen Kumar, director of products for supply products and customer experience at Supr Daily. Further, most deliveries occur within about 3 hours, from 4:00 a.m. to 7:00 a.m., making it difficult to follow up on the photos in real time.

Supr Daily wanted to automate the photo verification process so that it could provide instant feedback about whether a delivery partner needed to take a better photo. The company decided to use ML to reduce manual work and increase the speed and accuracy of the system. After realizing that building the technology in house would be too costly and complex, the company began working on multiple proofs of concept using Amazon Rekognition in 2020. “By building on AWS services, we no longer had to manage the infrastructure,” says Siddardha Garimella, senior engineering manager at Supr Daily. “Using Amazon Rekognition also simplifies the solution, so we could have someone with little ML experience onboard and begin building models quickly.”

Each time a delivery partner submits a photo of the delivered goods, Supr Daily automatically sends that photo to Amazon Rekognition, which uses ML to check the photo quality and verify that it is valid as proof of delivery. This system facilitates verification in near real time that deliveries have been fulfilled and that confirmation images are high quality. Using Amazon Rekognition Custom Labels—a feature that businesses can use to identify objects and scenes in images that are specific to their business needs—the process takes only 350 ms per image. “We were able to develop a solution that not only works better but also works faster,” says Praveen. “We could stop depending on people to make judgments and instead rely on quantitative data to understand what is actually happening.” Supr Daily uses the images it stores on Amazon Simple Storage Service (Amazon S3), an object storage service that offers industry-leading scalability, data availability, security, and performance, to build custom Amazon Rekognition models that can recognize items in the images with 95 percent accuracy. “When someone on the team wants to update or improve a feature, they can go straight to Amazon Rekognition and build a new model,” says Siddardha. “They can update the system, and within a few seconds, it will start using the latest model.” By using Amazon Rekognition, the company has also reduced the costs of image recognition. 

Since 2020, Supr Daily has also used Amazon Forecast, which can forecast business outcomes easily and accurately using ML, to analyze customer behavior data and to make sure it had the right inventory to meet demand. The demand forecasting workflow used to be manual, but Supr Daily has automated it on AWS, performing the forecasting on data stored in Amazon S3 and receiving results rapidly. Then, the company uses Amazon Simple Queue Service (Amazon SQS), which provides fully managed message queuing for microservices, distributed systems, and serverless applications, to send a notification containing the results to the procurement teams, who can place orders and make sure that items are in stock. “Our business users can simply go to the dashboard, upload their data, and see the results,” says Siddardha. “The process is incredibly fast.” On AWS, Supr Daily has improved its mean absolute percentage error by 25 percent.

Because AWS manages the infrastructure, Supr Daily can iterate more quickly and accelerate time to market for new features. Meanwhile, the app’s backend is hosted on AWS Elastic Beanstalk, a simple-to-use service for deploying and scaling web applications and services, making the solution scalable enough to support millions of customers in multiple cities. Implementing these services was simple. “One of the best parts of using AWS services is that they’re ready-made for someone who might not be extremely tech savvy to start building solutions to business challenges,” says Siddardha. Even without dedicated data scientists, Supr Daily built a solution that improves its app for partners and customers and can scale to include multiple cities. “AWS makes it simpler to experiment, faster to deploy, and more convenient to access our data,” says Siddardha.

Supr Daily is planning for continued growth and looking for ways to improve its delivery app for shoppers using AWS. One goal is to offer suggestions tailored to users’ interests as they browse the app using Amazon Personalize, which developers can use to create real-time personalized user experiences faster at scale.

“From infrastructure provisioning to building image recognition and inventory planning systems, AWS handles everything we need. And it’s simple to scale up to support our use cases,” says Siddardha.