Getir Halves Time to Insight and Boosts Forecast Accuracy 10% Using Amazon Redshift and Amazon Forecast

Founded in 2015 in Turkey, Getir has expanded to three continents. Customers use a mobile app to access Getir services, which include (along with on-demand delivery services for grocery items) GetirWater, GetirFood, GetirLocals, GetirBiTaksi, GetirJobs, and GetirDrive, To address the technical variety and scalability required to fuel its business expansion, Getir made the strategic decision to migrate to AWS.

Getir built its data warehouse on Amazon Redshift, which uses SQL to analyze structured and semistructured data across data warehouses, operational databases, and data lakes, using AWS-designed hardware and ML to deliver optimal price performance at virtually any scale. “Our use of technology over the years has helped us to optimize processes in both delivery and operations at scale,” says Nafi Ahmet Turgut, head of data science and analytics at Getir. “It is our main competitive advantage. We use huge amounts of data for operational efficiencies and growth opportunities.”

At first, every team operated within a single Amazon Redshift cluster. Spurred by a rapid increase in customer demand during the COVID-19 pandemic, Getir realized it needed a more agile infrastructure to meet growth objectives and simplify processes for its engineers. “The backbone of our operations is our data warehouse, not just for the supply chain but also for customer engagements,” says Yağız Yanıkoğlu, head of data engineering and platform at Getir. “All our campaigns, any kind of personalization features, any kind of customer notifications—everything is built on top of our data warehouse. It’s very important for the entire business.”

Getir created a hub-and-spoke architecture using Amazon Redshift RA3 instances with managed storage, which let companies scale compute and storage independently for fast query performance and lower costs. The decentralized architecture spreads queries across multiple nodes, cutting the time it takes to gain critical insights.

For example, Getir generates insights regarding customer retention periods twice as fast as it did with its previous architecture. The company also cut failed queries by 50 percent. “Our overall cluster health is much better now,” says Yanıkoğlu. “When we answer most queries successfully, we help teams speed up their processes as they analyze the data. We gain more time to engage with our customers.”

The company isolates the workloads that access shared data using Amazon Redshift Data Sharing, which facilitates virtually instant, granular, and fast data access across Amazon Redshift clusters without the need to copy or move it. “It made a huge impact for us,” says Yanıkoğlu. “We share data among clusters, which helps improve availability. And because we separate the workloads, we isolate the access of teams. That brings us more security than before because we can control who is accessing which data.”

Getir accesses many different types of data that help to optimize its operations. For example, the company incorporates location data such as GPS coordinates, routes, or addresses through the Amazon Redshift support for the spatial GEOMETRY data type. The company also uses third-party data sources that engineers manually and seamlessly add into Amazon Redshift. “Getir is a data- and tech-driven company, and we do a lot of analysis before we decide to expand,” says Turgut. “Our analysis is heavily dependent on our use of Amazon Redshift. We use it to explore potential areas for growth as well.”

For demand and supply predictions, Getir uses Amazon Forecast, a time-series forecasting service based on ML and built for business metrics analysis. Rather than develop and train new models for each country, Getir automatically uploads 2 years of historical data and related calculations from Amazon Redshift into Amazon Forecast. Amazon Forecast automatically inspects the data, identifies key attributes, selects the appropriate algorithms, and displays insights in a visual dashboard.

For supply chain planning, Getir provides product forecasts as well as distribution and replenishment information to warehouses and other points along the supply chain. “We improved our forecast accuracy by 10 percent using Amazon Forecast,” says Turgut. “In terms of business outcomes, we significantly enhanced our product availability and reduced waste.” In addition to performance improvements, Getir also cut model training time for more than 10,000 stock-keeping units—unique identifiers for specific products—by 90 percent.

Getir’s data scientists seamlessly use data from Amazon Redshift as a source for projects in Amazon SageMaker, which lets organizations build, train, and deploy ML models for virtually any use case with fully managed infrastructure, tools, and workflows.

Getir’s data team delivers more than 500 customer features every week. Using Amazon Redshift, the company has achieved 98 percent accuracy in its analysis of customer cohort behaviors—targeted insights into the behavior of customers based on their shared characteristics. Getir also attained a Forecast Value Added score of 100.6, an indication of extremely high accuracy in Getir’s ML modeling.

“We built 99 percent of our infrastructure on AWS, and we are heavily using its services,” says Yanıkoğlu. “We are planning a lot of different use cases for the future. We love using AWS.”