By Kerry Liu

Retailers are pouring billions into AI technologies in a bid to catch up to Amazon. But even as it becomes clear that AI will power the next wave of retail competition, many executives are grappling with how, exactly, they should deploy it. AI is fundamentally different from the technologies their IT teams are used to dealing with, while a thick layer of hype makes it difficult to separate the smart use cases from the dumb ones. Even in a relatively specific area like the supply chain, AI could mean anything from robotics in distribution centres to demand-forecasting algorithms.

So how should retail executives approach deploying AI? Here’s some key areas to focus on:

Look under the hood of your AI vendor

With corporations ready to write big cheques for cutting-edge tech, some software companies have struck on a profitable idea – sprinkle some algorithmic fairy dust over old-school data analytics and rebadge it as AI. So it pays to scrutinize potential AI vendors carefully.

Engineers, data scientists and mathematiciansare the people who are actually creating and testing algorithms and working with terabytes of data. However, they should be backed up by a team of people with first-hand knowledge of the retail industry and can bring this perspective to everything from the user experience to the data science. An AI vendor that lacks this experience is likely to ship a generalized product designed to roughly suit multiple industries. But a bank trying to forecast mortgage delinquencies obviously has very different needs from a retailer aiming to manage excess inventory.

Start small, then move fast

The best deployments start by clearly identifying a specific issue or a set of related problems that AI can solve with an implementation in one or two quarters and build from there.

Take Wal-Mart. Through its partnership with Microsoft, the retailer used AI-powered natural language processing to surface important information in its post-payment auditing processes. Using the lessons it learned there, it built out a platform that powers a chatbot to answer questions from the firm’s employees about their benefits. The company has now created an AI lab in one of its New York stores to test technologies like smart sensors that alert staff when stock or shopping carts are running low and monitor the store for spills that need cleaning up.

Companies like Amazon have moved on to even more complex implementations to actually understand what customers want to buy, when they want to buy it and which channel they prefer to engage with. Amazon uses that information to predict how much stock it needs and where it needs to position it to rapidly deliver to customers. That insight could be worth billions in incremental revenues.

Amazon is able to do this now because it moved decisively into using AI in early 2014. Its algorithms have already been refined by three years of data from its vast user base.

Free your data

Though retailers often have masses of data accumulated from decades in business, it’s usually trapped in separate legacy systems like sales platforms, inventory trackers and marketing databases. For an AI implementation to be most useful, the data that powers it needs to be combined into a single architecture. Though that would be an enormous task for humans, it can be accomplished relatively easily by – conveniently enough – AI.

Many cloud computing companies and AI vendors have invested heavily in machine learning systems that rapidly ingest, clean and integrate large amounts of data. Most can get a client’s up and running in this area in a couple of quarters.
The scope of AI to transform the retail industry is enormous. Accenture predicts the technology could raise industry profits by 60%over the next 15 years. Capgemini thinks AI could save the global retail industry up to US$340 billon. But AI isn’t an all-purpose retail savant. It has to be deployed smartly to drive efficiencies and sales – and that will still take human intelligence.

Kerry Liu is the CEO of Tornto-based AI company Rubikloud.