Machine Learning in Retail: Maximizing Sales

According to Oliver Grob, director of Digital Transformation for Retail at SAP, 2018 will be a breakthrough year for machine learning. Though digitalization has yet to really make its mark on brick-and-mortar stores, this might be about to change.

You are in charge of the go-to-market strategy for new SAP solutions for retail. Amazon Go shows that stores do not need staff. Are machines about to take over in retail?

I think we first need to see how Amazon’s model works in practice. In certainly shows what machine learning can do. The idea is simple: When a customer walks into the store they have to call up an app that identifies them and contains their bank details. Amazon Go uses image analysis to determine which products customers take off the shelves or put back, and which items they actually leave the store with. There are no checkouts. Customers pay using their smartphones.

The next step will be for stores eventually to have no staff at all. Robots will replenish the shelves according to algorithms designed to maximize sales. Even Amazon still has a long way to go. For instance, what happens if I take a tube of mustard off the shelf and put into someone else’s cart? Who pays then?

Retailers’ margins are especially narrow. Can they even afford these technologies?

“TensorFlow makes machine learning accessible to anyone, not just mathematicians and data scientists,” says Grob.

I expect this year to be a breakthrough year for machine learning in retail. Cloud platforms offered by Google, IBM, and SAP will play a big part in this. They are inexpensive, are a high-performance and rapid resource for neural networks, and make it really easy for companies to build scenarios.

You do not have to be a math expert to use TensorFlow, the open-source scripting language, to write and deploy machine learning algorithms. Users can simply pick the latest algorithms from an algorithm library. Image analysis and voice recognition are easy to deploy and do not need any major financial outlay. Google is certainly one of the major drivers of machine learning at present. After all, TensorFlow makes machine learning accessible to anyone, not just mathematicians and data scientists.

What can machine learning do for retail?

Machine learning will change, and in some cases is already changing, retail in three main ways.

  • Sales strategies: Clearance sales are a major challenge. In fashion, collections can change every two or three weeks. There are two key questions: When should a store should start reducing prices? And by how much? In the future, an algorithm, instead of gut feeling, will provide the answers. The system learns from past sales which discount strategy works best. Ideally, this is automated, making it a process that food retailers can easily emulate for fresh produce, or for the holiday season, or for goods whose sales are stagnating.
  • Image processing: Image processing has already come a long way and can be deployed easily using TensorFlow. As the Amazon example shows, cameras recognize which products the customer selects and buys, or puts back on the shelf. Retailers can use this data to determine how they place products in the future. Although RFID chips are viable in clothing retail, they are not for food retailers. Even though each one costs only a few cents, that is often more than the margin on a yogurt. Besides, RFID chips are not always reliable. Coming into contact with metal or liquids impairs their signals and makes them harder to read. Image processing is also used to monitor stock. For example, a global retail group from the United States sends a fleet of robots, each equipped with a camera, into its stores at night. The robots scan the shelves to spot goods that are in the wrong place and to identify the shelves that need replenishing. SAP uses image processing to determine how often and for how long brand logos appear on TV reports or in videos. This is a measure of brand impact, which marketing departments are keen to have.
  • Dynamic supply chains: The vision for machine learning is that, for instance, when a T-shirt leaves the factory it knows in which store it would sell best and where it would find the customer willing to pay most for it. Let us imagine that pants designed for the UK market are produced in green, the latest trend color. However, an analysis of Pinterest and Instagram posts shows that cyan blue will be the next major trend in that country. The manufacturer has two options: Either to cut the price or ship the pants to another country where green is still in. Although logistics processes can already be planned to high degree of accuracy, if demand shifts, the supply chain has to be dynamic and adapt to the changes identified by the latest reports.

How does SAP help customers deploy machine learning in their companies?

Many customers are already using machine learning for forecasting and to optimize stock replenishment, for example. Customer inquiries are directed to the right member of staff by means of optical character recognition and sentiment analysis. Companies can even use chatbots, which are becoming smarter all the time. They pull data from a database and turn it into an answer to the customer’s question. This is all done automatically.

Despite all these examples, brick-and-mortar stores are still very slow to digitalize. Implementing machine learning takes more than simply deploying SAP Cloud Platform and TensorFlow. A company needs to invest time, perhaps in a design thinking workshop, in understanding precisely how these products will help it fulfill its goals.

To help, SAP has built the SAP Leonardo digital innovation system, which contains the design thinking workshop and the technology — including machine learning tools. It also includes the SAP Leonardo Center locations – labs for holding workshops and customer demos, and for building models of the store of the future, equipped with SAP software. In the next five years, the volume of new data is forecast to increase five-fold to more than 160 zettabytes a year.

One thing is for sure: Big Data contains huge amounts of information valuable to retailers. They just need to start using it.