For too long, grocery retailers have attempted to digitize fresh operations with minimal success, usually turning back to manual methods and spreadsheets. But to future-proof grocery stores and keep them profitable, retailers need machine learning and artificial intelligence (AI) powering every order in the store.

Real-world results with a tangible outcome in every aisle

Fundamentally, the core question we’re answering at Afresh is this: For some [store, item, date], what is the correct order that both minimizes the risk of shrink (food waste) and minimizes the risk of stocking out (lost sales)? While that’s an impossible task for humans, it’s exactly what we built Afresh to do.

One of the most important parts of getting to that answer is running simulations—which are essentially the AI version of scenario planning, but for a near-infinite amount of variables. Unfortunately, most of the technology grocery stores use just can’t do such complex calculations.

Center store solutions and perpetual inventory fail in fresh departments

Most tech-enabled grocery retailers rely on ordering solutions on the IBM mainframe; those solutions use models for ordering that don't get retrained, have no feedback loop for customer complaints, and don't have online monitoring that addresses issues when they arise. This leads to massive amounts of waste on some products while others are quickly out of stock—in short, lost profit. What these inflexible, legacy systems fail to do is apply advanced machine learning techniques to forecast demand and optimize inventory replenishment in fresh.

Read our whitepaper to learn more about fresh-focused AI.

At Afresh, we account for different possible scenarios and use AI to come up with accurate order recommendations, which have a 97% adherence rate. By applying deep learning, machine learning, and sophisticated modeling techniques, our in-store app gives order writers data-backed superpowers to:

• Plan for the future with order recommendations, production planning, display management, and other decision-making tools
• See the invisible through modeling to infer unmeasured phenomena inside the store, such as shrink and inventory

Keeping up with all the dynamics of fresh is easier with AI

Tech solutions for grocery stores only work if they can handle the dynamics of the departments they serve—and Afresh is built for the job in fresh. Before launching with a new partner or deploying an update, we create a lifelike model of each store’s operating environment and run complex simulations across thousands of realistic scenarios to test and validate all those variables center store tech just can’t take into account.

We identify unknowns that human insight just can’t keep up with or see, like shelf life and perishability. This process ensures our algorithms perform as expected regardless of what changes about the product. Factors that impact ordering decisions like shelf space, shelf life, inventory accuracy, backroom space, hourly demand patterns, and missing demand are all part of the puzzle. 

During the testing phase, we can take any strategy that produces an order—say, always ordering 200 pounds of avocados—and evaluate performance over a wide range of adverse conditions. So knowing what’s important to the store and what data we have, we can test and adjust to any number of fresh-specific scenarios like:

• What's the average impact on product fullness if a truck arrives late?
• Which weekdays get the ripest product?
• What season do age distributions skew young? 
• Where do we see underperformance on promotions? 
• What ordering scenarios lead to the highest levels of shrink?
• Which price changes resulted in lower overall item sales?
• What’s the level of shrink if we only order the product once on Mondays?

Algorithms in action: solving shrink and stockouts

One of the most difficult items to keep in stock without creating excessive shrink is avocados. They’re a must-have item in every store and customers want avocados to be just right in terms of ripeness—which varies based on what they’re eating and when. Depending on what time of year it is, the shipment may arrive with rock-hard avocados or a pallet of nearly overripe alligator pears. Writing the perfect order means balancing all the factors of that fruit to answer the question: How much do I order today so I have enough varied ripeness and don’t stock out before Wednesday? 

Now imagine trying to make that calculation on pen and paper. Right as the delivery truck arrives. After you’ve found out a team member called in sick. And today’s order is due in 30 minutes.

Our Fresh Operating System simultaneously solves two prediction problems that humans just can’t do as quickly or effectively as technology, especially when they're in a time crunch: 

• How much demand will there be for an item over the next several days?
• How should I order based on that demand?

At Afresh we know the perimeter is the center of every shopper’s satisfaction score, so we go to great lengths to make sure our models perform better than your current solution and that it improves with system updates. Testing scenarios ensures the solution will work as it should in the real world, minimizing bumps in the road during launch and giving store teams a streamlined transition to new tech. 

Read about our impact on businesses and the planet.

We all love the experience of walking into a grocery store stocked with fresh-looking, delicious foods. And most customers expect to have a variety of options in stock. Despite changing tastes and shifting product availability, fresh department employees have managed to do a remarkable job keeping displays full of the foods their customers want. But in the end, the efforts of employees can’t meet the demands of such dynamic products. And that’s where Afresh comes in. 

Stores that use Afresh have collectively prevented more than 6.7 million pounds of food waste, reduced shrink by 394 BPS, and driven $4.9 million in bottom-line savings. Give customers, employees, and your business a better experience with better tech for your teams. Set up a discovery call.