Tchibo: Optimizing demand forecasts with AI to match customer needs
About Tchibo
One of the leading retailers in Europe, Tchibo has its roots in roasting and selling coffee and is one of the largest roasters in the world. Its business model has expanded to incorporate coffee-making equipment and non-food ranges and services, including communications. In addition to its online operations, it has 900 stores and over 11,300 employees worldwide, with a presence in more than 24,300 retail depots.
Tell us your challenge. We're here to help.
Contact usLeading German retailer Tchibo built a forecasting model with Vertex AI and BigQuery, generating millions of predictions each day, to manage customer demand in a timely and cost-effective way.
Google Cloud results
- Fully automated forecast service now generates over six million predictions a day
- Online demand estimated up to 84 days in advance
- Significant time and cost savings by reduced overstock and handling efforts
- Anticipated significant sales increase by reduced stock-outs
Automated predictions provide availability to Tchibo
Tchibo launched as a family business in Hamburg, Germany, in 1949—revolutionizing the coffee market by sending bags of beans out by mail. In the years since, it has evolved to become one of the biggest e-commerce businesses in Europe. The company offers a wide and frequently updated range of non-food items and services, such as communications. This creates challenges in managing supply and demand for the many non-food items—from clothing to homeware—in its primary warehouse.
The business model of Tchibo, with fast-changing weekly sales phases and multi-channel distribution online, in retail stores, and in Tchibo stores in multiple countries depends on a wide logistics distribution network. It brings in around 3,000 new products every year, with some campaigns planned a year in advance, so optimizing the company's logistics processes was critical for cost savings and the successful fulfilment of customer service expectations.
"Our data scientists and engineers really saw a big advantage in what Google Cloud was offering. It gives us a lot of flexibility."
—Sören Götze, Principal Data Scientist, Tchibo"If we supply a branch with too many goods, we have to pick them up again and then bring them somewhere else, which means logistics costs with every process. But if the goods that people want are out of stock and not available in the online shop, we'll lose business," says Marcel Knust, Head of Data Science and AI, Tchibo.
Digitalization has made the right technology vital for Tchibo, not only for customer journeys, but also for the digital data services that support its operations. In 2019, having evaluated the capabilities of a number of suppliers in performance, security, cost and usability, it decided to build a central analytics platform for big data processing and storage on Google Cloud.
"Our data scientists and engineers really saw a big advantage in what Google Cloud was offering," said Sören Götze, Principal Data Scientist, Tchibo, who led the demand forecast project. "It gives us a lot of flexibility."
Managing supplies, streamlining operations
Tchibo collects a lot of sales data, but found that its previous data analytics solution was manually maintained and lacked the forecast quality that Tchibo needed. It sought a state-of-the art platform for the development of data solutions with machine learning and advanced analytics.
Tchibo had worked with Google Cloud since 2019, when it moved to a cloud-based infrastructure in order to modernize and strengthen its operations. Encouraged by the positive outcomes from that integration, Tchibo turned again to Google Cloud in 2022 to help build an online demand forecasting service—known as DEMON—which is now accurately predicting demand for the online sales channel. This helps Tchibo manage its warehouse supplies and reduce the time spent by its operations teams on logistics. It also gives the company more of a steer on what products are landing well with customers and should be brought back again.
Tchibo created a forecast covering 84 days for each item, using a temporal fusion transformer model (TFT) on Google Cloud for time series forecasting. The TFT relies on an event-driven microservice architecture, where new incoming data triggers the next steps in a workflow. Model training and production are separated and the models, versions, and experiments are centrally managed and easily deployable.
DEMON is based on a microservice architecture in which different tasks, like data gathering from the Tchibo data vault, feature-store creation for training and inference, prediction, delivery of the results, reporting to stakeholders, and technical monitoring are organized. The whole process is scheduled and orchestrated by Google Workflows. Forecast results are then stored and handed over to subsequent IT systems, like ERP and allocation, where logistical operations by Tchibo fully automatically replenish the warehouse.
Generating six million predictions a day
The platform uses a constantly enhanced learning model that takes into account a large number of growing sales and campaign data, as well as additional features, now underpins data-driven decisions and market-leading position of Tchibo. The forecast service generates over six million optimized predictions a day, helping Tchibo to ensure it has the right level of products to meet customer demand in a timely and cost-effective way, without overflow sitting in the warehouse.
"We use BigQuery as a massive engine for all our data gathering, joining data from many different sources and creating a feature store where all the information that we need for our forecasts is collected for use as a training source and as the basis of all our predictions."
—Sören Götze, Principal Data Scientist, TchiboTchibo makes extensive use of native Google Cloud services during development and production – often beginning exploratory data analysis, experimentation, and initial development in Vertex AI platform notebooks, while complete data gathering and transformation pipeline relies on BigQuery. Historical data going back more than three years is fed into the forecast model, including product data, online shop planning and sales data, marketing data, and logistics data from six countries.
"We use BigQuery as a massive engine for all our data gathering," says Götze, "joining data from many different sources and creating a feature store where all the information that we need for our forecasts is collected for use as a training source and as the basis of all our predictions."
Continuing to grow with automation
Thanks to the success of its demand forecast service, Tchibo is anticipating a significant sales increase and cost savings. The platform is constantly undergoing optimization based on stakeholder feedback, and the company is in the process of transferring its learnings from the service set up to other data services, with an aim to try and increase the maturity of its machine learning operations pipeline.
"We have been using our developments on Google Cloud to deliver reliable and rapidly scalable data and AI services at enterprise level for some time now. We are growing together with Google Cloud, and building new services for different use cases, as well as optimizing existing services."
—Marcel Knust, Head of Data Science and AI, Tchibo"We have been using our developments with Google Cloud to deliver reliable and rapidly scalable data and AI services at enterprise level for some time now," observes Knust. "We are growing together with Google Cloud, and building new services for different use cases, as well as optimizing existing services."
Knust says that the Tchibo team is excited about testing further rapid AI and tech developments, which have opened up new fields of action and opportunities, especially in recent months with the use of LLMs and vector databases. It anticipates its platform becoming even better and more automated with integrations of the rapidly growing Google model zoo, Vertex AI pipelines and automated content creation with generative AI tools.
Tell us your challenge. We're here to help.
Contact us
About Tchibo
One of the leading retailers in Europe, Tchibo has its roots in roasting and selling coffee and is one of the largest roasters in the world. Its business model has expanded to incorporate coffee-making equipment and non-food ranges and services, including communications. In addition to its online operations, it has 900 stores and over 11,300 employees worldwide, with a presence in more than 24,300 retail depots.