METRO: Migrating to the cloud to better serve customer needs

Variety is the spice of life

Dusseldorf-based METRO is one of the largest B2B wholesalers in the world, serving some 16 million hotel, restaurant, and catering (HoReCa) industry customers distributed across 34 countries. METRO owes its size and breadth to a long period of organic growth. Like any growing business, METRO has needed agility and flexibility in order to take advantage of opportunities to connect with new customers and build new capabilities as the industry shifts and changes.

Flexibility is especially important when working with HoReCa because the needs of organizations in this industry can vary wildly between businesses—picture the differences in restaurant menus between a coastal city and a small inland town. And a single business can have dramatic differences in its seasonal demands—imagine what a catering company serves at a wedding in January and how that compares to a wedding in July.

Working with such a diverse range of businesses, METRO’s leadership knew the company couldn’t be successful with a one-size-fits-all approach to its relationship with its customers. Instead, METRO needed to build something personalized. The company needed to be able to predict its customers’ needs, not just at the industry level, not just at the regional level, but for each individual business. To do that, METRO needed data.

Scalability that supports data complexity

Getting data from HoReCa customers isn’t always easy. Dr. Werner Rath, Unit Owner of IT Operations at METRONOM, METRO’s tech subsidiary, explains, "The industries we supply tend to have fixed ways of working. Until recently, orders were still often made by telephone, or even fax. When that started to change, we decided to fully embrace digitization, with a close focus on our customers and their needs." METRO’s membership-based business model made it possible for the company to collect individualized data about what each customer purchased, when, and in what quantities. METRO also examined other data sources, like local weather patterns, that could influence the purchases made by the customers of METRO’s customers.

To make use of this data, METRO initially invested in an on-premises data center with hardware and software specialized for machine learning (ML). Unfortunately, the company ran into challenges. "As we moved to more advanced ML techniques, we quickly reached the limitations of a hardly scalable environment," says Dr. Ehler Lange, who is responsible for data science at METRONOM. To improve scalability, METRO moved the data lake to the cloud. Google Cloud, with its managed services and well-developed ML offerings, seemed like a natural fit.

"With a lift-and-shift migration, we reduced infrastructure costs by 30–50 percent," says Dr. Rath. "Instead of 10 VMs rebooting every week, we rarely have to deal with one. Outages and periods of instability are down by up to 80 percent, and there have been no major incidents whatsoever since we made the migration."

With the help of Google Cloud Partner freiheit.com technologies GmbH, the METRO team migrated its ecommerce platform to Compute Engine instances on Google Cloud, using Virtual Private Cloud to create easy integrations with the company’s backend systems. After the move to Google Cloud, it was possible to optimize in ways that weren’t possible in the on-premises environment. Dr. Rath explains, "To optimize cost and performance, we were able to switch some workloads to more cost-effective instance types, including switching between CPU processors, in a short amount of time and on a huge scale. This would not have been feasible with our own data center because of the long life cycle of the hardware."

A data-driven road map

With its customer data in the cloud, METRO could focus on making that information more readily available. This, in turn, would allow for a more data-driven customer engagement model. "We ran extensive workshops with people from all around the organization to find where AI could add the most value," says Dr. Lange. "We picked out key subjects to focus on, from improving customer analytics to optimizing pricing, reviewing our assortment, and making our supply chain more efficient. This isn’t abstract for us. We have a clear road map to make gains from machine learning."

Historically, business intelligence (BI) at METRO was mainly based on a well-accepted, heavily used, and user-friendly system for enterprise data warehouse reporting. The introduction of a large data lake and advanced analytics solutions created an opportunity to improve on the company’s existing BI. Because of the shift to the cloud, METRO was able to take on a major milestone from its road map: the democratization of data analysis. METRO hoped to collect more semistructured and detailed data from point-of-sale systems, customer databases, and marketing campaigns, and make that data available inside the company.

To make that happen, METRO built a data lake and analytics solution, mainly based on BigQuery and other Google BI services. "The main advantage of the managed services is that we can scale up and down," says Marko Schwob, Domain Owner of Analytical Platform Engineering at METRONOM. "Scaling is about more than storage space, it's about having analytical power available on demand. Calculating item recommendations for customers requires a lot of CPU. With Dataproc, we can create a cluster for the calculation, get the results, and then shut the cluster down, which is much more efficient."

Schwob continues, "In addition to the flexible scaling, which is one of the biggest advantages of cloud-based analytics, this solution gives us access to a wide variety of analytical tools for many different use cases. We can use everything from data management tools to pipelines to machine learning and AI, which we can offer to our users and stakeholders. It would be nearly impossible to provision all these tools on-premises."

Stefan Richter, founder and Head of Engineering at freiheit.com, adds, "When we built the new ecommerce platform with METRO, our software engineers integrated the collection of behavioral data from day one. We stream the data into BigQuery and we are using Datalab infrastructure to create customer insights and develop machine learning models with TensorFlow. We trained the product owners to use BigQuery SQL, so that they can explore the data themselves."

With one API for data ingestion and another for integrating with other products, the solution enables real-time reporting on data streamed from stores and applications. The data is analyzed to identify trends and predict customer needs. These insights can be used to create tangible action plans, helping customers make improvements such as reducing time spent on complex tasks and enhancing their product offerings.

"With the data lake, we can do much more for customers," says Sven Lipowski, Unit Owner of Customer Solutions at METRONOM. "By connecting data points, we can offer advice on how to comply with hygiene laws for certain foods or information on provenance. We can even integrate a local weather forecast so a store doesn't run out of ice cream on a sunny day."

Internally, the data lake supports a huge range of projects. "We see Looker Studio dashboards that pull data from our data lake built on Google Cloud popping up in every part of our organization," says Lipowski. "They're driving a new emphasis on outcome-driven product management. Instead of relying on opinions, we use KPIs on dashboards as a commonly agreed-upon framework to set out our priorities."

To help internal product teams use the data lake, METRO created more than 100 data analytics workbenches. "The workbenches are like playgrounds for data," says Marko. "Each one is tailored to a certain function and gives a team easy access to services on BigQuery, Cloud Dataproc, and other tools. People can play with machine learning applications there, too, exploring TensorFlow and other possibilities in the future."

For METRO, machine learning is already making an impact. Drawing on customer behavior data from many areas, a new app measures and predicts levels of customer satisfaction, so sales teams know which customers to reach out to, when to do it, and why. Small business owners can also use technology to predict customer demand, making it possible to make more accurate buying decisions, save money, and reduce waste.

"We expect to be able to integrate with our customers' point-of-sale and ERP systems in only the next couple of years," says Lipowski. "That will mean being able to automatically replenish stock, forecast shortages, and match a customer's seasonal needs for produce. It means taking the stress out of planning ahead."

All these capabilities were dependent on the scalability and complex analytics offered by Google Cloud. Dr. Lange explains, "Our migration to Google Cloud came at absolutely the right time, allowing us to use much more advanced TensorFlow-based models for our customer product recommendations. This would not have been possible in the previous tech stack."

Seizing the opportunity to do more

METRO was able to take a step further, facilitating the digital transformation of its customers. METRO leadership and Hospitality Digital, a subsidiary of METRO, partnered to develop the DISH platform. The platform simplifies the process for METRO’s small-business customers to get online, providing a range of digital solutions that help with day-to-day business. Restaurateurs are given their own website. They get access to a reservation tool, a track-and-trace solution for restaurants, and listings on services like Google My Business and Reserve with Google. MenuKit, another tool provided through DISH, digitizes the menu and uses ingredient costs to calculate the profit margin on a per-dish basis. With their websites already integrated with other services, METRO customers end up with more visibility online, driving more business.

Meanwhile, more elements of METRO's business are connecting to the data lake. In the ramp-up phase of the data lake at the end of 2018, its daily intake of events increased by 45 times, from 1.2 million events a day in October 2018, to 54 million in January 2019. Now the company is optimizing its analytical tools as it works toward achieving omnichannel analytics: tracing a customer's whole experience of interacting with METRO. "We're committed to that goal," says Marko. "All our future customer-facing applications will soon be measured by Google Analytics 360, and then that data, too, will be added to the data lake we built on Google Cloud."

The new ecommerce platform is now live in 18 countries. "From 2017 to 2019, we tripled the number of orders executed through the platform," says Lipowski. "We see a positive effect in overall sales as well as ordering customers."

And thanks to the team's success connecting the ecommerce platform to a legacy backend, METRO is now migrating its SAP S/4HANA finance systems to Google Cloud. Timo Salzsieder, CIO/CSO at METRO, recently announced the move: "Google Cloud, in addition to technical advantages, offers the possibility to optimize collaboration within the individual teams. Moreover, we can now adapt our system to customer demands in real time."

METRO’s commitment to facilitating digital transformation, both internally and in collaboration with its customers, serves the company well. Its customer-centered approach helps businesses stay afloat in a notoriously unstable industry. This collaborative approach to customer relationships is even more important in the context of the impact of the COVID-19 pandemic on the HoReCa industry. The implementation of a stable, scalable cloud infrastructure, and the democratization of data-driven business intelligence, has enabled the success of not only METRO, but also the restaurants, hotels, and catering companies it serves.