Tesco is starting to incorporate machine learning algorithms across the business, from internal applications such as driver routing to customer facing apps like integration, with Google's home assistant device.
First, Tesco had to lay the groundwork for machine learning techniques to be brought into the organisation, and this meant getting its data lake in order so that near to real-time data could be used by the developers and data scientists within the company.
Speaking at the AI Summit in London today, Tesco group CTO Edmond Mesrobian spoke about the importance of creating a data loop where "everything knowable is captured and then we can reason about it and build models. Take those models and reflect them back into the business, wether that's a colleague or a customer, to make better decisions." He says that Tesco has been working on this for a year and a half and is now starting to see the benefits.
While Mesrobian recognises that online retailers like Amazon have been doing this since day one, with recommendation engines and warehouse optimisation, retailers like Tesco are still reliant on their physical stores, so have to try and blend data from the physical and online world. "We want to represent all of Tesco, be that through fulfilment, delivery, retail, online. So it needs to be connected intelligence," Mesrobian said.
This desire to bring AI-powered products to market quicker at Tesco was the driving force behind an open source project called Mewbase, which was announced in February.
Developed by Tim Fox, principal software engineer at Tesco, Mewbase is an open source system which brings together messaging, events and database to allow Tesco's developers to "manage their events and data, eliminating the need for them to communicate with other databases and event stores...so our teams can generate a new working service from scratch in seconds from metadata", according to Tesco Labs.
By giving developers access to streaming data from across shopping basket events, IoT sensors and supply chain stock in real-time they can start to apply AI to give more effective suggested actions, like adding a missing item for a customer or optimising stock processes internally.
Mesrobian calls Mewbase a "toolkit to allow anyone to build a real time streaming engine that emits and consumes events and to do analytics at the edge".
Examples: Google Home, stock availability and van routing
Mesrobian says that the aim of any AI application at Tesco is for "delighting customers or providing efficiency benefits across our enterprise, both are equally meaningful".
For customers, Tesco's Labs division has been working with IFTTT to open up its APIs and create 'recipes' so that online shoppers can start to personalise their shopping, get automatic price drop alerts for certain items and order groceries though AI-powered home assistants like Google Home.
Tesco now integrates with Google Home, allowing customers to add an item to their basket at any time using voice commands. Then, if they have a delivery slot already booked it will amend and checkout that order, or if they don't, then items will be added to the basket for next time.
The idea is that Tesco can start to better personalise its services for customers, such as the AI that spots missing items from a user's basket which then automatically suggests that it gets added. This is similar to the work being done at Ocado with TensorFlow to personalise and optimise the shopping basket when shopping for groceries online.
One example of an employee-facing machine learning project at Tesco includes better in-store routing algorithms to reduce the walking distance Tesco Online personal shoppers take when picking items in stores.
These staff members pick 1.5 billion items a year and Mesrobian says that by optimising their routes using machine learning algorithms the retailer has been able to reduce the average walking time of these staff by 20 percent, which most importantly allows them to complete more orders.
A similar use case is around van routing and scheduling for better efficiency for drivers. Mesrobian calls this a "deep computation problem" but the goal is to have vans making more efficient delivery routes to reduce their impact on the environment.
Tesco has also been using computer vision algorithms through its static in-store cameras to tackle item availability so that store staff can better react to empty shelves to get them restocked quicker, cutting down on customer disappointment.