January 21, 2020 - It’s almost a cliché to say that retail is always evolving, but it’s never been more true than it is today.
As retail increasingly migrates online (Forrester predicts that ecommerce will represent 17 percent of US sales by 2022, up from 12 percent in 2017), competition among brick and mortar retailers to attract and retain customers will get more intense. Machine learning and the Internet of Things will be essential for retailers – both online and off – in the coming years.
Here are four practical use cases for including machine learning in your retail operations:
1. It can help predict what the customer will buy next. Recommendation engines, popularized by online retailers like Amazon, are powerful machine learning-powered tools to increase sales and foster customer loyalty. They allow retailers to increase a customer’s lifetime value by extending retail sessions and up-selling them.
There are a number of ways recommendation engines are currently implemented. A technique called content-based filtering, for example, analyzes a customer’s shopping history (along with other data, like demographic information and reviews and ratings the customer has given) to predict other products he or she might want to buy.
Similarly, collaborative filtering relies on machine learning to compare a customer to similar demographic cohorts to make future purchase recommendations. Often, these purchase recommendations are described to shoppers as “inspired by your browsing history…” or “other customers also bought…” but make no mistake: these recommendations aren’t simply culled directly from recent purchase or search data.
Thanks to the insights of a machine learning algorithm that synthesizes multiple data sources, they’re often uncannily good predictors of customer behavior.
2. It can help optimize inventory. The practice of inventory optimization is a critical part of the retail workflow, and one that is also ripe for a machine learning transformation.
There are a lot of reasons why retailers need to adopt inventory optimization. For starters, for a lot of businesses, the very nature of inventory has changed in recent years. Long tail demand is increasingly a part of the landscape for retailers, which makes forecasting and supplying a challenge. And traditional ABC Classification systems aren’t robust enough for modern retailers.
Machine learning, though, can analyze vast quantities of sales and inventory data to find patterns that elude traditional systems and human inventory planners. This enables retailers to rely on probability forecasting, which can more easily accommodate volatile customer demand.
3. It can help with employee utilization. The most expensive resource in almost any organization is human resources and making the best use of employees on the floor is an eternal struggle.
There are a lot of strategies for optimizing employee time, but one increasingly popular approach is to leverage wearable tracking devices. By combining these trackers with machine learning systems that combine logged data about where they spent their time, smart systems can predict where employees will be needed on the retail floor, when peak demands will occur, and how to prepare to staff the store for peak and minimum loads.
That’s not all – in a large retail environment, you can enhance the customer experience by offering “call buttons” for customers – either via in-store kiosks or in mobile apps – and let smart software request employees who are closest to the customer to assist, not entirely unlike the way some healthcare facilities have begun using the IoT to geo-track medical staff and direct the closest ones as needed to assist with medical emergencies.
After you have success with employee utilization, the same tracking-and-analysis methodology can be applied to other in-store assets like shopping carts.
4. It can help prevent fraud. Another area filled with opportunity for retailers to make a substantial difference using machine learning is fraud prevention. It’s certainly ripe for improvement: The National Retail Federation estimates that fraud and shrinkage cost the industry $50.6 billion in 2018.
Traditional methods for flagging suspicious activity are rules-based, which is both inflexible – unable to detect novel patterns – and overwhelms investigators with false positives. But machine learning algorithms can be trained not just with rules but also with real-world fraud data, allowing the system to classify suspicious fraud cases far more accurately.