Three high-impact areas and easy wins in ML for grocery retailers

Being around for decades, machine learning (ML) technologies have proved to unlock new opportunities for all types of businesses from internet giants and manufacturing to banking and agriculture. Grocery retail is no exception: industry leaders reap increased operational efficiency and decreased costs as a result of intelligent automation. However, not everyone succeeds. According to the recent MIT Sloan research report in collaboration with BCG, seven out of 10 companies surveyed report minimal or no impact from ML, and 40% of organizations making significant investments in ML do not report business gains from ML. More often than not, digital transformation is seen by retailers as a full-scale (and costly) infrastructure project rather than means to achieve immediate and long-term business goals. In the current article, we will highlight the three high-impact areas and easy wins of implementing ML that allow retailers to gain measurable effects and return on their investments.

In a highly competitive battle for customer loyalty, grocery retailers have their bets on fresh food. Fresh food drives sales, basket size, and traffic to the store. High-quality fresh food can even be equal to price as the most important criteria for choosing a grocery store. This being said, retailers do want to make sure that fresh products are always in stock.

However, fresh food is perishable, and with a shelf life of 1 to 7 days, it requires daily forecasting and daily replenishment posing a challenge to every retailer's planning department. And results are often far from satisfactory: demand forecasting errors cost retailers billions of dollars annually, with fresh food comprising 70% of food wasted. As it turns out, while popular forecasting systems perform well for basic stock items, most of them are just not suited for fresh food.

This poorly optimized area of demand forecasting gives retailers a great opportunity to get tangible results with every improvement made. ML technologies, and Bayesian methods of machine learning, in particular, can give retailers a fresh perspective on tackling this challenge. Instead of point estimations, Bayesian methods of machine-learning utilize interval estimations that allow computing quantile grids and approximating probability distributions for different levels of future demand. With fresh and ultra-fresh products losing their market value every day, interval estimations can help drive optimal inventory allocation decisions.

This approach requires fewer amount – months rather than years – of data and allows getting business gains within the first month after starting the project. The key to such rapid outcome is resistance of Bayesian methods to overfitting and high turnover typical for fresh food. Retailers can track results of implementing fresh food forecasting solutions by using business metrics such as the number of write-offs and out-of-stocks. According to DSLab experience of implementing ML-based fresh food forecasting, retailers can attain write-offs reduction of 10% while out-of-stocks tend to decrease by 5%. For example, food delivery service Samokat with over 120 warehouses in Moscow and Saint Petersburg obtained a 14% decrease in write-offs and a 7.5% reduction in stockouts.

To learn more about fresh food forecasting, read the article Demand forecasting: challenge of fresh and ultra-fresh assortments.

Safety stocks are essential to the planning process: they are used to calculate reorder points, smoothen unexpected demand fluctuations and mitigate the risk of stockouts. The latter often puts inventory managers under immense pressure, especially when holidays or sales promotions are around the corner. The fear of being left with an empty shelf and losing customer loyalty drives an understandable motive to order a "little extra" inventory. At its peak, a "little extra" can be as high as 20% of the forecasted demand.

Implementing ML-based safety stock forecasting can eliminate the emotional factor of inventory management, reduce the level of manual fine-tuning and optimize safety stock levels for seasonality, upcoming holidays and sales promotions. Incorporated into the forecasting systems, optimized safety stocks will lead to better demand forecasting and balanced inventory policies. Depending on the policy in effect, potential decrease in safety stocks can reach up to 20% within 2-3 months for fast-moving goods and within one year for slow-movers.

Another gap that can bring retailers an easy win is intelligent pricing. There are a few well-established pricing methods that grocery retailers currently utilize: rule-based pricing management, competition-based market alignment, or algorithmic pricing management. However, price optimization software tends to be expensive and doesn't account for such influencing parameters as item cross-effects or sales promotions; while algorithmic pricing is usually based on simple machine-learning tools that offer derivative-free modeling without gradient optimization. Focusing on the maximization of immediate revenues and margins, these methods don't guarantee long-run effects on retailers' business.

It is hardly a surprise that retail executives name intelligent pricing among the AI automation activities their companies plan to engage in by 2021. With electronic labeling readily available, intelligent pricing is already expanding from online shops to brick-and-mortar grocery stores. Among potential gains are the elimination of costs associated with printing and revising price tags, instant price changes to thousands of items in hundreds of stores, dynamic prices optimized for both internal and external parameters such as time of day or weather. In addition, ML-based repricing process can be fully automated and integrated into the planning routine: from optimizing price tags to placing automated order in the retailer's replenishment system. But most importantly, intelligent pricing drives more traffic to the store in the long run while maximizing immediate revenues as well. Potential business effect includes 10-15% of margin increase within 3-6 months after project launch.

The table below summarizes required data, estimated time expenditures and investments as well as potential gains of the described high-impact areas of ML implementation in grocery retail:

Machine learning conceals wide application potential for grocery retailers: from demand forecasting and inventory planning to pricing optimization and sales promotions management. Applied correctly, it can help retailers reach new heights of operational efficiency, deepen customer loyalty and win a competitive advantage over industry rivals. To achieve it, retailers need to embrace practical approach and pursue ML transformation not as yet another infrastructure project with dim prospects, but rather the means to achieve immediate and long-term business goals with measurable results and defined success criteria.