For M&S, forecasting food demand means preparing for predictable peaks like Christmas and Easter, while also dealing with something far less predictable, especially for a 140-year-old retailer. Like a new product suddenly taking off on TikTok.
At NRF Europe in Paris, Susan Massicot, Head of Food Supply Chain at Marks & Spencer, shared how the retailer is using machine learning to tackle both as part of a four-year transformation of its food supply chain with RELEX.
M&S has more than 1,000 stores across the UK, with around 7,000 products on average in its stores. Most of its sales come from short-life food with around just five or six days of shelf life. For chilled and fresh food, M&S operates a stockless supply chain, with products arriving at depots and going straight back out to stores. Stores cannot influence their own orders, making it a “fully push model”, which Massicot said relies on stores executing the plan accurately. “We need to have a very responsive and very precise supply chain,” she said.
Teaching machine learning about Christmas
Seasonality is one area where machine learning has changed the way M&S forecasts demand. Christmas and Easter create obvious peaks, while even summer is an event for M&S Food, with customers stocking up on the classic “picky bits”.
Previously, forecasting these peaks required significant manual intervention. M&S can now tag historical data around events, allowing RELEX’s machine learning to use previous patterns to create an initial forecast. But historical data can only go so far. Around half of M&S’s seasonal range is new every year, Massicot said, “because we need to keep the excitement”.
Its food strategy is to “protect the magic of M&S and modernise the rest”, and innovation is a big part of that magic. “We launch products every single week, and a lot of these products are completely brand new to the market,” Massicot said.
That creates an obvious problem for forecasting as there isn’t always historical sales data to work from. For new launches, M&S maps products against similar existing lines to create a reference forecast. Massicot said automating more of the day-to-day supply chain execution has also given teams more time to work with product and innovation colleagues earlier in the process. But then there are the products that suddenly take off.
When a product goes viral
Asked how M&S forecasts demand for viral products, Massicot was clear about the limits. “I would love a crystal ball to get that right every time,” she smiled.
She used M&S’s Wimbledon inspired sweet sandwich as an example, but said that when the retailer launches something completely new to the UK market, there is little history to indicate exactly how customers will respond. At that point, the challenge becomes less about predicting demand and more about reacting to it quickly. Massicot said close supplier relationships are critical. M&S works with suppliers to understand how quickly they could respond to a jump in demand, where bottlenecks might be and what needs to be put in place to increase supply.
“We don’t want to disappoint customers when we’ve got a viral product, because the customers will tell us on TikTok that they’re not happy that they couldn’t find the product,” she said.
It highlights an interesting limit to what machine learning can currently do for retailers. Christmas, Easter, weather and differences between individual stores all leave patterns that can be learned. Predicting how shoppers will respond to something genuinely new is much harder.
Four years of supply chain transformation
The technology forms part of a wider transformation of M&S Food’s supply chain that began in 2021. M&S chose RELEX rather than building its own system, with the aim of making forecasting, ordering and allocation more data-driven and reducing manual intervention.
The rollout took around four years, longer than originally expected, partly because M&S wanted every food category managed through the same system. That meant accommodating everything from sandwiches with only one day’s life in store to chilled, ambient and frozen food, loose produce, in-store bakery and ingredients used in M&S cafés.
Chilled came first, with the remaining categories following. Since March 2025, the whole M&S Food operation has been running on RELEX. Massicot said M&S is now in the optimisation phase, looking at what else can be automated and where it can further improve availability, freshness and waste. Waste was a key KPI throughout the implementation, with M&S adjusting the system by product type to balance availability against reducing waste and improving freshness for customers.
M&S is also tackling food waste beyond its forecasting operations, having announced a partnership with Too Good To Go this week to tackle bakery waste across 354 stores.
More than a technology project
One of the biggest lessons from the four-year programme was that changing the technology wasn’t enough. Massicot said M&S had to change its operating model alongside the implementation and support teams as they adapted to new roles and ways of working.
“If we tried to implement the new system with the team still working the way they used to work, we would have delivered 0% of the benefit that we had intended to deliver,” she said.
M&S is now extending its use of RELEX into space, range and display, which is expected to go live next year. Three RELEX engineers are also embedded within M&S teams as part of a newer AI-enabled development approach designed to speed up changes, while the retailer is exploring RELEX Cortex, an AI layer designed to work across multiple systems.
But perhaps the more interesting takeaway from Paris was where the technology stops. Machine learning can help M&S understand what Christmas, summer or a change in the weather might do to demand. But knowing which new food product shoppers will decide to make famous on TikTok is another matter.
