Tesco uses supply chain analytics to save £100m a year

Warm day in Scotland

The group also used regression testing to understand correlations between weather data and sales patterns. It found that not only do sales of certain products, such as barbecue food, rise in warm weather, but demand also varies according to location and context. A warm day in Scotland will produce different demand to the same temperature in the South. Similarly, demand on the first weekend of a hot spell will be different to one in the middle of a heat wave.

Factoring weather into Tesco’s demand forecasts has helped it to avoid having too much or too little stock, saving £6m each year.

Similar techniques have been applied to understanding demand patterns caused by special offers, cutting out-of-stock by 30% on these items, said Apthorp.

Computer models produce a coefficient which is fed into Tesco’s live order management system, adjusting the amount suppliers are expected to deliver each day. Run on an IBM System Z mainframe, the system manages stock orders from thousands of suppliers, worth around £100m every day. But the mainframe’s reliability comes at a cost, said Apthorp.

“If I want to put new software into that mainframe, there is one update every six months. And typically I have a year's lead time to get that software cued up, written by the IT guys and delivered. If I have a simple change, I can get it done more quickly. But for big substantive things, it’s a year,” he said.

As such, Apthorp’s team has to ensure that the changes to the ordering algorithms are correct. Mistakes would be costly. To avoid them, the team built a computer model of Tesco’s supply chain, which can run on a desktop, and applied new scenarios to it. “We can run 100 stores for 100 days in about half an hour. We can figure out quickly whether what we are doing is right and we can optimise that,” he said.

Computer simulations are also used to try out new ideas for best practice within store. Tesco then tests ideas successful in simulation in the real world, creating its own experimental data.

“It is very easy to put a change into some stores and not in other stores, and make careful trial verses control measurement,” he said.

Previously left to managerial judgement, algorithms built by the supply chain analytics team now produce discounts transmitted to handheld devices in-store. Taking out the guesswork has saved ££30m of wasted stock

Trials start small and are allowed to “fail fast”. Ideas which appear valuable on early trials are then tested on a larger scale in around 25 stores.  “With three month’s data you’ve got all the proof you want for all the senior stakeholders to say yes,” said Apthorp.

Successful trials also help to engage managers furthers down the chain with new ways of working. “The store director becomes the man who is saying, ‘this is fantastic’. He will then talk to his colleagues, and suddenly you have got stores phoning you up asking you to have it, almost before you’re ready to launch it,” he said.

End of shelf-life food discounting

One successful change to working practice has been discounting of food towards the end of its shelf-life. Previously left to managerial judgement, algorithms built by the supply chain analytics team now produce discounts transmitted to handheld devices in-store. Taking out the guesswork has saved £30m of wasted stock, according to Apthorp.

This series of successful projects has helped the analytics team gain the ear of the top management team, allowing it to build on its approach of measuring, modelling, testing and acting.

“We have proven track record of success that means that we have access to some very senior people,” he said.