Grocery’s crystal ball: AI-driven forecasting
Accurate demand forecasting in grocery has never been more critical.
Even before the COVID-19 pandemic struck, grocery retail was a difficult business. This is due to the fact that maintaining loyalty of the modern shopper isn’t easy; especially with new food trends and products continuously popping up and increasing online and delivery options changing the food shopping experience altogether.
Legacy purchasing and forecasting systems are at the heart of the problem. For starters, these rigid systems don’t account for local events, weather patterns and dozens of other variables. Often, the promotional calendar ends up a manual override rather than part of the plan.
They also lack complete visibility into the many factors that actually drive consumer purchasing decisions and make it difficult for retailers to respond to the day-to-day needs of their customers. This results in out-of-stocks, lost sales, excess inventory, and unnecessary waste.
Forecasting and replenishment solutions driven by artificial intelligence (AI) can help mitigate these issues. Not only can these solutions help grocery retailers gain a better understanding of customer behavior by evaluating multiple costs across multiple outcomes, but they also enable informed decisions about pricing, replenishment, and assortment at a highly granular level.
So, if you’re a grocery retailer looking to implement AI-powered forecasting and replenishment solutions, here are three essential things to keep in mind to ensure success:
1. Expand beyond historic data forecasts
AI forecasting models should be able to self-learn from other items and stores to create forecasts even when there is limited data, such as when you are running a promotion or introducing a new product. These algorithms should draw on detailed information from similar promotions or items to predict how the new product will respond to different prices, weather and events.
Looking beyond historic data enables forecasts to respond quickly to demand fluctuations. This also allows retailers to evaluate changes across multiple stores and products to understand whether these fluctuations represent an isolated incident or an emerging trend.
2. Explain AI in simple terms
One of the biggest obstacles to AI adoption is trusting a “black box.” While an AI-driven system may make decisions using calculations beyond human cognition, its recommendations should not defy human understanding. The right forecasting and replenishment system must present deductions in an explainable form.
Explainability starts with ensuring planners can easily view, monitor, and understand the influences on demand to improve performance. For instance, if the algorithm suggests a sale on T-bone steaks, it could be because said products are reaching their expiration date. In this case, the user interface could highlight key decision-making factors in the machine learning (ML) models to reveal what is behind the optimization decisions.
3. Automate forecasting by item, store and day at scale
Products behave differently in individual locations and at different points of the year. Treating your products as part of a product group in a cluster of stores fails to recognize the local dynamics at work. The best ML models evaluate and learn impacts at a granular level to identify local and regional trends. They make hundreds of millions of calculations every day, and include not only the influence of many factors on demand, but also the influence of factors on one another.
For example, when planning stock levels for an event such as Thanksgiving, you should not expect all variants of a water to sell in equal volume (sparkling water may have more prestige). These effects may be more profound at some locations than others, as well as be inter-connected to other demand influencing factors such as how many days are left until the holiday.
Ultimately, ML should be able to evaluate and learn impacts at a granular level to identify local and regional trends. This allows retailers to make hundreds of millions of calculations every day, allowing complexity to be automated while improving KPIs.
Driving significantly higher levels of accuracy
Without AI, it is impossible to understand the relationships between influencing factors at the speed and scale the modern grocery retailer’s needs. Intelligent solutions look beyond the isolated influences of each factor. They seek to understand how each factor impacts the others. They determine how promotions, events, weather and other variables impact individual locations over time, and they even compute day-level variations.
Smart ML algorithms can discern these impacts to deliver new levels of forecast precision at the item-location-day level. And by making these forecasts explainable, expanding their analysis beyond historic trends and automating them by item, store and day at scale, retailers can drive significantly higher levels of inventory accuracy.
Jim Hull is senior industry strategies director at Blue Yonder.