How machine learning and AI can transform retail as we know it
By Suresh Acharya, Chief Scientist, JDA Labs
We’re living in amazing times for artificial intelligence and machine learning. The potential for new innovations coming out of research in those fields is capturing people’s imaginations. For example, some predict that self-driving cars will be so much a part of everyday life by the time today’s babies are teens that the kids born after 2016 won’t need driver’s licences.
For scientists like my colleagues and myself who work with these emerging technologies, it’s an exciting challenge to reimagine the world’s supply chains and explore ways we can transform the ecosystem to be more intelligent and efficient.
In retail, this means understanding how machine learning can change the way we learn about customers – and then, lay the AI foundations for a truly self-learning supply chain. That’s an exciting prospect in a world where customer expectations are changing fast, forcing retailers to change even faster.
What does AI mean for retailers?
A lot of people use “artificial intelligence” and “machine learning” interchangeably, but they’re actually different. Machine learning is just one part of a much larger AI ecosystem that overlaps with other areas of data science. And it includes not just predictive analytics that can give us insights about future results, but prescriptive analytics that can tell us the decisions we need to make today so that we’re more likely to see the results we want.
These new systems are only as good as the data they interpret. As a retailer, you’re already collecting a great deal of data every hour – from online and in-store customer interactions, purchases and returns, to product attributes and inventory information, and so on. You may also be exploring new sources of data that can add new dimensions to your knowledge, such as sensors, connected internet of things devices and social networks.
But all this data will be worth nothing in the future unless you’re able to generate both predictive and prescriptive information from it, as well as the ability to interpret that information in ways your leaders can use to make decisions. That means powerful machine learning capabilities coupled with intuitive user experiences to make decision-making easier.
The learning curve: Correlating retail data
Today, we’re actively creating machine-learning algorithms to help retailers incorporate all these sources of data and solve new business challenges. Case in point: returns forecasting. Think back to 15 years ago, and you probably saw around 7 to 10% of your items returned. In 2018, pureplays see up to 60 or 70% of items coming back. That’s a logistical and forecasting challenge that can’t be ignored.
Since a return only happens when a sale has happened first – and since we can understand not only the attributes of items being returned, but the how, when and where of the original sale – we’ve got the ability to analyse the correlations between sales and returns, then figure out the probability of an item being returned in the future. Analytics can also unlock the customer segment with the highest propensity to return.
This one example shows how we can use new technology to transform retail from relying on forecasting only what we can see, to include patterns we could never recognise on our own.
Forecasting based on the outside world
Now, go a step further and think about how sources of data from outside your stores can make other kinds of predictions easier. For instance, if you could know that warm weather will come to one area earlier than usual, you could change out your seasonal apparel offerings in those stores sooner and be ready when your customers want to shop for spring and summer.
What about new product launches? Besides daily and weekly sales information, what if you could gauge the social sentiment around a new product and use that insight to guide your sales forecast?
But the possibility of correlating outside information from social media, news, events and weather is only possible when your internal data house is in order. For many retailers, a lack of detailed product attribute information is a problem that could make other kinds of decision-making easier. The more you know about product attributes, the better you can understand the decisions your customers might be making – for example, why a person who ordinarily buys product A might decide to choose product B instead.
The competitive challenge
Online mega-marketplaces like Amazon collect tremendous information about products and customers: what people are buying, what purchases get abandoned, what items are browsed and how often. This level of data analysis still isn’t common across the brick and mortar landscape.
The easiest solution today is to use loyalty information to track how often customers buy, the products they choose and the promotions that attract them. With that information and the right software, you can create robust customer profiles that can drive product selection, ranging, pricing and other strategic planning.
But just like self-driving cars could change the way we live and travel, the promise of unsupervised AI is the ability to look at transaction and customer interaction information and create better customer segments, more localised product assortments, optimised pricing and better promotions.
We’re not that far away from self-learning retail supply chains that can analyse correlations in data and answer many of our questions: Did we have a stock-out because an inbound order was late? If so, was it late due to a traffic issue? Did weather cause that traffic issue? Or some other factor?
If we can bring all that information together and learn from it – can we be smarter about predicting stock-outs? Can we build retail systems that can learn based on past interactions, then look ahead to predict issues that might impact our stores, and our customers, tomorrow – or in a few days – or next week?
These are a few of the ways that machine learning could break retail free from relying only on historical data and the old ways of making decisions. The future of retail will be data-driven, and with these emerging technologies, it’ll be more efficient and predictable than ever before. And the data divide between the haves and the have-nots will ultimately determine the retail landscape of the future.