Machine learning helps grocers win the fresh-food competition
Fresh produce, which can account for a third of a food retailer’s costs and 40% of its revenues, is a fiercely competitive segment at the best of times, but more and more combatants are crowding into this key battleground. Convenience-store chains, discounters, and online grocers all understand the power of fresh food to lure customers away from the supermarket giants.
But it’s not an easy activity to get right. As McKinsey & Co. notes in its 2016 report The Secret to Smarter Fresh-Food Replenishment? Machine Learning, the product can be highly perishable and the demand for it can vary greatly, while lead times are not always predictable. Moreover, retailers want to offer a broader range of goods – including rare and exotic items with an extremely short shelf life – in their search for a competitive advantage.
“To try to win the fresh-food war, leading retailers are sparking a revolution in supply-chain planning by turning to machine learning,” says Christoph Glatzel, a senior partner in McKinsey’s Cologne office and co-author of the report. “Based on algorithms that allow computers to learn from data even without rules-based programming, this is enabling them to automate processes that used to be manual.”
Most traditional supply-management systems have an approach to forecasting and replenishment that’s based on fixed rules. “This is fine for stable and predictable product categories, but fresh food is way more complicated,” says Matt Hopkins, co-author and principal at Blue Yonder, a London-based market research firm that works in partnership with McKinsey. “All the time, retailers are making difficult judgements and trade-offs when placing orders with fresh-food suppliers.”
Retailers need to know how much of a given product to order from their suppliers. It’s an obvious question, but there’s no easy solution when demand is fluctuating. If you order too much food, it goes to waste; if you order too little, you lose sales and your customers lose patience when confronted with empty shelves and “sold out” signs. As the demand for certain produce shifts daily, planners must continually enter various types of data manually into their replenishment systems. We all know that manual processing is a bore and, as a result, prone to human error. But the planning process for fresh food relies heavily on the skills and instincts of the people involved.
By applying machine learning systems, retailers can automate these cumbersome processes, leading to potentially dramatic improvements in forecasting accuracy. In contrast to standard supply-management software, machine learning systems can collect, analyse, and adjust large data sets from a wide variety of sources efficiently and at a relatively low cost. The technology looks set to become the norm in retail and other sectors.
Early adopters of machine learning for replenishment have already seen a number of benefits, according to the McKinsey report. These include:
- Reductions of up to 80% in stock-out rates.
- Reductions of more than 10% in days of inventory on hand and write-offs.
- Gross margin growth of up to 9%.
Retailers have traditionally predicted demand by looking at historical sales data and extrapolating from them. The drawback of this approach is that it tends to skew forecasts downwards because it ignores any demand that wasn’t satisfied, according to the McKinsey report.
Machine learning algorithms should overcome this problem. They base their demand forecasts on historical sales data but take account of many other influential factors, such as advertising campaigns, the weather, and key dates on the calendar – bank holidays, for instance. In fact, retailers using this form of artificial intelligence are taking account of 50-plus parameters. Machine learning can determine the effect on every stock-keeping unit (a specific item stored in a specific location) of each factor.
Taking stock
But machine learning is about more than computer wizardry. Retailers should start to think about how the technology could influence the strategies of their fresh-food operations. They could, for instance, redefine the assortment rules governing the number of stock-keeping units in a merchandise category, group, or department.
For instance, a retailer could use machine learning to define which items should always be on the shelves at particular times. Inevitably, companies that adopt machine learning will be prompted to review their procurement processes from end to end, including store-level ordering and inventory management. This could have an effect on distribution schedules. Certain items may need to be delivered twice a day, for instance, which would have an impact on planning at distribution centres.
As the planners work out how to get the most from machine learning, it will also have an effect on their roles. Both in stores and at the head office, they will need to review performance management metrics and incentives. Of course, the impact of this technological revolution will be felt all the way along the chain, with suppliers having to adjust their own forecasting and ordering processes accordingly.
By prompting such transformational change, machine learning promises to improve the lives of food retailers and the consumers they serve – a fruitful development all around.
A version of this article was published in the April issue of Financial Management magazine.