Artificial intelligence, or more accurately machine learning, holds huge theoretical value for businesses across retail. It has applications from inventory management to personalisation to strategic planning and store location; but brands and retailers are in danger of letting this utopic vision obscure some of the important groundwork that comes first.
Per Forrester, successful implementation of AI has been “held back” in 2018 because “AI is data-dependent and data hungry. But most firms struggle with basic data governance issues.” The vast majority of businesses just aren’t set up to deploy the exciting techie stuff right now. In spite of this, almost every major retailer has been hiring or attempting to hire data scientists in a competitive jobs market, in the hope that they can work some of their machine learning or algorithmic magic to turn things around.
Retailers are therefore in real danger of putting the cart before the horse on data science, especially if their new data teams aren’t empowered to make sweeping changes to technical setups, procedures and technology stacks.
Here are the problems they need to tackle and how they can start to sort them out.
Big data in a big mess
Retailers have access to a huge wealth of potentially valuable data on their customers, their products, their past performance, et cetera. However, these data sets tend to live in isolation and in various formats. Centralising and standardising these into a comprehensive source of information is a great way to set up for success with any machine learning project.
Where to start with machine learning?
Often the complex nature of new technology and techniques can make it hard to understand where they can most effectively be applied. The same is certainly true of machine learning for retail.
One of the most often-cited examples of machine learning is in product recommendations, which can be intelligently adapted based on user data such as previously browsed, saved and purchased products; search terms; et cetera. Personalisation doesn’t have to be limited to a niche use case like product recommendations however – this data could be used to influence the products highlighted on a retail site homepage to a returning user.
Perhaps the biggest opportunity for brands and retailers is in applying the lessons of one machine learning implementation to bigger and bolder use cases like dynamic homepage content.
Regardless of the scope of the potential implementation, it’s essential to test on a small scale to minimise initial disruption. Once the concept is proven and the results are starting to work, the available data set can be scaled up and the algorithm can start improving faster.
Connecting the tech stack
One of the obstacles for brands and retailers is often legacy systems which obstruct the efficient transfer of data between departments, services or to other systems. Adopting new technologies with API layers which allows them to communicate with any other system is essential to scaling a data-driven, machine-learning enabled business.