M&S Food Put Its Entire Range on Machine Learning

Running a fresh food supply chain leaves little room for error.

For M&S Food, the challenge is amplified by a range dominated by own-label products, thousands of lines, short shelf lives, constant innovation and demand that can change rapidly around weather, events and products going viral.

M&S describes itself as a UK retailer with over 140 years of retail heritage and one of the most beloved food brands by customers in the UK. Its food strategy, as Massicot summarised at NRF Europe, has been consistent for five years: "protect the magic of M&S and modernise the rest."

For chilled and fresh food, there is an additional complication that makes modernising the supply chain particularly demanding.

For chilled and fresh food, M&S operates what Susan Massicot, Head of Supply Chain at M&S Food, described as a fully stockless supply chain. Products arrive at depots and flow directly out to stores rather than sitting in distribution-centre inventory.

That means forecasting mistakes cannot simply be absorbed by a warehouse full of safety stock.

At NRF Europe in Paris in September, Massicot explained how M&S has spent the past four years rebuilding the technology and operating model behind that system, moving its entire food operation onto RELEX for forecasting, replenishment and allocation.

The lesson from the transformation is more interesting than simply putting machine learning into a supply chain.

M&S discovered that the technology was only one part of the job.

Fresh Food Leaves Very Little Margin for Forecasting Error

M&S Food carries around 7,000 products in an average store, according to Massicot, with much of its sales coming from own-label and short-life products. Some products have only five or six days of life, while sandwiches can have as little as one day of shelf life in store.

That creates a difficult equation.

Order too little and availability suffers. Order too much and waste increases, while customers may also encounter products with less remaining shelf life.

M&S must solve that equation across an estate ranging from large out-of-town stores to small railway-station locations, each with different customer missions and demand patterns.

The retailer also has pronounced seasonal peaks around Christmas, Easter and summer, while its product innovation creates another forecasting problem: history does not always exist.

M&S launched nearly 1,500 new food products during its 2025/26 financial year, according to the company's annual reporting. Massicot pointed to the strawberry sando, launched in partnership with Wimbledon, which sold one million units in four weeks after going viral.

That is commercially valuable innovation. For a forecasting system, it is uncertainty.

What Happens When a Product Has No History?

Traditional forecasting becomes particularly difficult when a retailer launches something customers have never bought before.

Massicot highlighted the challenge during the NRF Europe session when discussing products that unexpectedly go viral.

There is no historical demand curve for a completely new product. And social media can accelerate demand much faster than conventional planning cycles.

Her answer was revealing because it showed the limits of technology.

"I would love a crystal ball to get that right every time."

For products with viral potential, Massicot said the answer is not simply a better algorithm. M&S works closely with suppliers to understand how quickly they can respond, where production bottlenecks could emerge and what needs to be prepared upstream before demand materialises.

That distinction matters.

Machine learning can improve the interpretation of demand signals. It cannot manufacture additional production capacity when a product suddenly explodes on TikTok.

In those moments, supplier relationships become part of the forecasting strategy.

M&S Decided It Did Not Want to Build the Technology Itself

The transformation began in 2021.

M&S wanted forecasting, ordering and allocation to become more data-driven while reducing the amount of manual intervention required from its supply chain teams.

It also made an early strategic decision not to build the underlying planning platform itself. As Massicot put it at NRF Europe, M&S is a food retailer, not a specialist supply chain software company.

The retailer selected RELEX, with the original implementation beginning in March 2021 across 1,050 stores and 13 distribution centres in the UK and Ireland. The initial ambition included improving forecasting, fresh optimisation, ordering and allocation while reducing manual processes around areas such as weather and promotional planning.

The choice was partly technological. Machine learning could incorporate variables such as weather, seasonality, promotions, events and store-level differences into forecasts.

But Massicot said another consideration was configurability. M&S knew many of its problems were specific to the characteristics of its business, particularly short shelf lives, high levels of innovation and the requirements of its own-label model. It therefore wanted a platform that could be adapted rather than forcing every category into a standard planning process.

The Transformation Took Four Years

The original ambition was broad.

M&S wanted everything flowing through its food supply chain to eventually operate through the same planning system. That included chilled products, ambient grocery, frozen food, loose produce, internal bakery, café ingredients and other categories.

But those categories behave very differently. A sandwich with a one-day store life cannot be planned in the same way as a bottle of wine with a long supplier lead time.

That forced M&S to implement the system in stages.

The retailer began with the technology and its data foundations, including tagging historical data around events, local activity and weather so the forecasting engine could interpret the information correctly. It then piloted the system before progressively adapting it to different categories.

Chilled, the largest and one of the most complex areas of the food business, was rolled out by September 2023. Ambient required different ordering and replenishment logic because M&S holds stock for those categories in national distribution centres.

Frozen, bakery, cafés and loose produce followed.

By March 2025, Massicot said the full M&S Food operation was running on RELEX.

It had taken around four years. Longer than M&S initially expected.

The Biggest Lesson Wasn't About Machine Learning

That delay produced perhaps the most useful lesson from the project.

Technology implementation was not the transformation. Changing how people worked was.

Massicot said M&S initially underestimated the degree of organisational change required around the new platform. Experienced supply chain employees who understood the previous processes deeply suddenly had to learn a new system, new ways of working and, in some cases, different roles.

M&S therefore had to build an operating model around the technology rather than simply installing software and expecting the benefits to follow.

"It really never is about just the technology."

Massicot said attempting to introduce the new platform while allowing teams to continue working exactly as they had before would have delivered none of the intended benefit.

That is an important distinction for retailers investing in AI. The value does not come from possessing the algorithm. It comes from redesigning decisions around it.

Automation Gave Planners Something More Valuable Than Efficiency

Reducing manual work created another benefit: time.

Seasonality provides a good example.

Before the machine learning implementation, Massicot said seasonal forecasting required considerable manual intervention. The new system can use historical information and tagged seasonal data to generate a stronger initial forecast automatically.

But Christmas ranges do not simply repeat themselves. Many seasonal products are new each year.

That means planners still need to decide which historical products provide the best reference points for new launches. Massicot said automation has freed supply chain teams to spend more time having those conversations with product and innovation teams well before products reach stores.

This is one of the more significant implications of retail automation. The objective is not necessarily to remove human judgement. It is to stop using human judgement on work that machines can perform repeatedly, so people can concentrate on decisions where context still matters.

Waste and Availability Are Two Sides of the Same Problem

Fresh food makes the consequences of poor forecasting unusually visible.

Too little inventory creates empty shelves. Too much creates waste.

But Massicot added another dimension: freshness.

If M&S is consistently generating waste, it can also indicate that too much short-life stock is reaching the shelf, potentially leaving customers with less remaining life on the products they buy.

M&S therefore treated waste as a key KPI during implementation and worked through different strategies depending on the product and its shelf life.

The aim was not simply to minimise inventory. It was to find the right balance between availability, freshness and waste.

RELEX said in March 2026 that M&S was using its platform to improve availability and limit waste, while increasing automation and reducing manual intervention. The vendor did not publish specific M&S percentage improvements in that announcement.

That absence is worth noting. The transformation story is therefore better understood through the operational changes M&S has disclosed than through unsupported claims of a particular percentage improvement.

A Stockless Supply Chain Makes Store Execution Part of the Algorithm

M&S's chilled and fresh model introduces another important constraint.

Because those categories operate without depot stock, the system depends heavily on accurate end-to-end execution. During the NRF Europe Q&A, Massicot explained that stores cannot simply make independent ordering decisions when additional inventory is not sitting upstream waiting to be requested.

M&S largely operates a push model for these products. That makes store execution critical.

Displays need to be implemented as planned because the data used for forecasting and allocation depends on the physical reality in the store matching the assumptions inside the system. Promotional displays, shelf requirements, product life and replenishment practicality all feed into the decision.

In other words, sophisticated forecasting does not remove the importance of basic retail execution. It makes consistency more important.

M&S Has Now Moved From Implementation to Optimisation

Since completing the full rollout in March 2025, M&S has entered what Massicot called the optimisation phase.

That is where the retailer believes more of the benefits can now be found. With teams trained and the platform stable, M&S can identify remaining manual interventions, adjust forecasting and replenishment processes, and continue working on the balance between availability, waste and freshness.

The scope is also expanding. M&S is now working with RELEX on space, range and display planning, with Massicot saying at NRF Europe that the capability is expected to go live next year.

The attraction is integration.

Forecasting what customers will buy is one problem. Deciding what range should sit in a store, how much space it receives and how that physical merchandising decision influences demand is another. Bringing those decisions closer together could give M&S a more connected view of how commercial choices translate into supply chain requirements.

The Next Step Is AI Across Systems

M&S is also exploring a newer AI layer.

Massicot said the retailer is looking at RELEX Cortex, which she described as a way of coordinating information and activity across multiple systems rather than limiting intelligence to data already sitting within the core RELEX applications.

The potential becomes clearer when viewed against the broader M&S technology estate. Forecasting, allocation and replenishment contain one set of information. Space, range and display contain another. Commercial and operational systems contain more. The opportunity is to connect those environments so that decisions are not optimised in isolation.

M&S's own FY2026 reporting points in the same direction. The retailer says AI is being used selectively where it can reduce costs or improve decisions, including pricing, waste reduction and personalised customer offers, while supply chain modernisation remains one of its core investment programmes.

M&S Is Modernising the Physical Supply Chain Too

The software transformation is only one part of a much larger investment in food logistics.

In May 2026, M&S began construction of a £340 million automated food distribution centre in Northamptonshire, its largest-ever supply chain investment. The 1.3 million sq ft facility is scheduled to open in 2029 and forms part of the retailer's effort to modernise and increase capacity across its food network.

Taken together, the investments show how M&S is approaching supply chain transformation on two levels. One is physical: capacity, automation and distribution infrastructure. The other is computational: forecasting, allocation, replenishment and increasingly AI-assisted decision-making.

Neither works particularly well without the other.

The Real Transformation Is Moving People Away From the Forecast

There is an easy way to tell this story: M&S implemented machine learning to forecast fresh food.

But that misses what actually changed.

The retailer has been gradually moving human effort away from manually producing forecasts and towards managing the exceptions, relationships and commercial decisions that forecasting cannot resolve on its own.

Machine learning can recognise seasonality. It cannot decide how ambitious M&S should be with an entirely new Christmas product. It can process weather and demand patterns. It cannot create extra capacity at a supplier when a sandwich unexpectedly goes viral. It can calculate replenishment. It cannot fix a store that has failed to execute the display the forecast assumes is there.

That is why the four-year implementation matters.

M&S did not simply automate a forecast. It redesigned the operating model around it.

And now that the foundation is in place, the next stage is less about asking whether AI belongs in the supply chain and more about deciding how many decisions it can connect.

For fresh retail, where a forecasting mistake can become an empty shelf or food waste within days, that distinction matters.

The future of AI in grocery may not be about removing people from supply chain decisions. It may be about making sure people spend their time on the decisions the machine still cannot make.

Sources and Editorial Verification

Primary source: Susan Massicot, Head of Supply Chain at M&S Food, "Fresh at Scale: How M&S Food Transformed Its Supply Chain", NRF Europe, Paris, September 2026. RetailNews.ai source transcript supplied from the session.

Additional verification: RELEX Solutions, "M&S Food Selects RELEX Solutions to Provide Forecasting, Ordering, and Allocation Management", 21 April 2021; RELEX Solutions, "RELEX Partners with M&S Food", 4 March 2026; Marks and Spencer Group, FY2026 preliminary results, 20 May 2026; Marks and Spencer, 2026 annual reporting; Marks and Spencer announcement on its £340 million automated food distribution centre, 13 May 2026.