Retailers are becoming increasingly sophisticated in their desire to optimize their stores, especially as they’re able to get more and more information from in-store analytics like those provided by the RetailNext platform. Retailers want to test, measure, and iterate like never before in areas like store design, merchandising, staffing optimization, and in-store marketing.
However, many of the changes a retailer would like to make are expensive or time consuming, especially if they’re hoping to perform many such changes.
However, imagine if you could try out hundreds or even thousands of floorsets overnight, and have data about the efficacy of each one. What if you could change your physical store, actually move the walls, over and over again, measuring the Key Performance Indicators (KPIs) you care most about – traffic, conversion, average transaction value, sales per shopper – with each change?
Predictive simulation does exactly that.
With simulation, we use real data that we gather in-store using the RetailNext analytics platform in order to train software agents to act like real shoppers. We can then put these simulated customers into modified or even entirely new stores, and measure their movements and behaviors just like we would with real shoppers.
The details of agent-based modeling are somewhat complex, but at a high level it’s pretty simple. We start with a software representation of an actual store, along with agents with no intelligence at all – they simply obey the laws of physics. At this point, our software model resembles a video game engine without the players. Software representations of customers wander aimlessly through the store, bumping into walls and each other.
From there, we begin to add intelligence. We first allow the agents to make decisions about their own movements that are similar to the decisions real people make in real stores (as observed across thousands of stores by the RetailNext analytics platform). We then model product, POS, entrance, and other relevant locations present in the actual store and allow our software agents to establish goals when they enter the store. These goals are typically based on sales, traffic, and other observed data that we’ve gathered in the store we hope to simulate.
We can assign different behaviors and goals based on various properties of the shoppers – creating male shoppers that head into the men’s section, grab a pair of jeans, pay and go, and female shoppers that spend more time browsing skirts and blouses, for example. All of this is based on real, observed data.
So, we now have a simulation that should resemble the real store we’re observing. Since we have equivalent real and simulated data, we can compare these two datasets statistically and determine how well our model “fits” what we’ve actually observed. If the fit isn’t good, we adjust the model until the simulation looks like what we’ve actually seen.
This is where it gets interesting, because we can now start to test changes.
Since your goal upon entering a store isn’t based on the store layout (you walk in wanting jeans, regardless of where the jeans are), we can assign goals exactly the same way, modifying only the path the agent takes in order to reach that goal. Or, similarly, we can modify the goals customers have, simulating for example a promotion on a particular item or category.
What happens to the flow through your store if most people who walk in the door are looking for the same item? Then, what happens if you move that item?
What happens if your traffic increases tenfold or even 100x on a given day?
Does traffic to the products you care about increase or decrease if you change the store footprint – for example, knocking down a wall and expanding your space?
What about if you add an entrance? Or double the number of POS terminals?
Predictive simulation can help to answer these questions and so many more, testing hundreds of different variables simultaneously if necessary. As retailers search for a differentiated shopping experience that delivers optimal results on their KPIs, predictive simulation allows judgment on efficacy and feasibility before any resources are committed and costs incurred.