Machine Learning in Retail: Know Your Customers’ Customer

This guest post is authored by Stephen Kinns, CEO and Co-Founder at CatsAi, who will be presenting this use case on November 3, 2021 in our upcoming webinar about Machine Learning in Retail .

Are you in food and beverage retail (FnB)? Do you want to reduce waste? Here’s a common situation in FnB that you may recognize.

Your business delivers to clients or sells at more than 10 sites. Maybe even up to the 1000’s.  You might be geographically spread or locally dense. Your business may sell fresh produce,  artisan, or prepacked. The product line maybe a handful or hundreds.

You have long term plans for strategy, perhaps medium term for steering resources, and short term operational forecasts to line up the products and people ready to deliver.

You’ve been using the same planning and forecasting process for a few years and it works absolutely fine. You’re broadly content with the performance. There’s likely room to improve, but not really seeing a driving need to change anything.  

On the other hand, your retail sites throw away food on a daily basis due to slight over supply.  Your customers, your investors, and your employees want to know why. Perhaps they’re financially challenging the situation, or more likely in these times of climate change awareness they’re looking to you to make a positive contribution.

So you’ve got some waste that worries you. Then you’ve also somehow got empty shelves. Customers are demanding you fix the problem with your deliveries. Their own customers are going elsewhere. And you know an empty shelf is a missed opportunity – another sale gone.

So how do these line up?

You’ve got the baseline forecasting but you’re not getting it right on the day at the location. The likely reason is it’s a naive forecast. That’s not to be rude; that’s the technical term. If you’re forecasting on a basic statistical level, or applying a “last week+10%” or “average of last 6 weeks” then that’s normally called a naive forecast.

So what?

A naive forecast is unable to respond quickly enough to real-life.

At every location, for every product, in the next week things will deviate from what was originally planned.  Your customers may respond to adverts, the weather may change, local traffic or events may change access times to locations. Each factor, at each location for each product has a different outcome to sales. Millions of possibilities.  Too much for humans, or even most advanced software.

And the answer?

Hyper-Focused Predictions. This is a new type of forecasting and it’s very short term. It’s only thinking about the critical next few days. It’s hyper-local, and hyper-agile. It can take thousands of influencing factors and adapt to the changing environment. Every day without fail, learning and adapting.

It’s a tool that sits alongside your existing forecasting helping you adapt.  

There’s only one way available to do this – and that’s Machine Learning. Unfortunately, that’s normally a very expensive proposition if you choose to build it from scratch as talent, software, and hardware costs can run very high. Fortunately, there is one company that has encapsulated all of that into a solution with a timeline to positive ROI counting in days/weeks not years for any size of firm.

We will be revealing more about how this alternative forecasting process works with a peek into the associated machine learning workflows in our upcoming webinar: Machine Learning in Retail on November 3, 2021. So be sure to reserve your spot today.