Artificial intelligence (AI) is increasingly being applied in many industries, and retail is now no exception. From machine learning to robots, retailers are testing and hoping AI is the answer to their two perennial challenges—expense reduction and sales growth.
What Is AI?
Artificial intelligence is the capability of a machine to imitate intelligent human behavior, as Merriam-Webster defines it. Like most buzz words, AI in reality has a very broad meaning, which allows marketers and salespeople to stretch the definition. For that reason, there is a lot of room for misconceptions.
For those of us in retail, often the first thing that comes to mind when you hear the term AI is its applicability to data analytics. More specifically, we want to know what AI can do for asset protection, from predictive analytics to true prescriptive analytics. The promise of AI here is to take the collected data, analyze it by machine-learning algorithms, and help the retailer make the right decisions.
Machine learning is computer science that gives computer systems the ability to “learn”—meaning progressively improve performance on a specific task—without being explicitly programmed. So AI, as it relates to data analytics, sounds like a dream come true, right? Then why hasn’t every retailer adopted it to solve their shrink problems?
Challenges with AI in Retail
Like all technology, machine learning has some fundamental challenges. One problem in a retail environment is that data is often too vague to translate directly into machine learning. Another problem is that the people who create algorithms often don’t have clean data to work with, or don’t fully understand which data is most important.
Retail analytics companies are slow to hire retail executives who understand retail intimately. What they do have is some very smart people who do the math and build a model, and another smart group of folks who can tell the story behind it to sell it. But often neither of these groups actually understand the problem. For example, a common sales pitch promotes using AI to find return fraud. In this case the software could be looking for deviation of normal activities. But without knowing what variables are used to determine the likelihood of fraud, the asset protection team using the system won’t know if the system is really learning to find fraud or doing simple outlier detection. I am not suggesting one is better than the other, just emphasizing that we should know the difference.
Areas Where AI Is Making a Difference
One place where AI is definitely growing in retail is with self-checkout. In the next 12 to 18 months, we can expect to really notice it there. In this area, the focus is on using advanced video analytics with machine learning to reduce shrink and fraud while improving the selling experiences. This will allow for less human involvement in transactions.
In video analytics specifically, AI has made some noticeable progress and can now take your video to a whole new level. Amazon, for example, now has a brick-and-mortar store that doesn’t need salesperson interaction. This is probably the best known example of AI using video and sensors to shape the customer experience.
Logistics is another area where we can expect AI to play a more prominent role for retail. It’s only a matter of time before drones start delivering your packages. Machine learning will enable AI systems to determine when a drone delivery makes sense based on weather, traffic, personnel availability, other deliveries in the area, and customer expectations. With AI, logistics systems will get smarter based on feedback and customer data, and all this should, in theory, improve profitability and enhance customer experience. Additionally, AI could increase end-to-end visibility throughout the supply chain. AI can help execute product shipments using real-time information with thousands of variables, finding the most efficient fulfillment route. Data and sensors with AI behind them can minimize spoilage and damage, as well as increase speed.
AI’s Fatal Flaw
AI does have one fundamental flaw—it’s not thinking like a human yet. We still need people to apply art to the science. IBM’s Watson AI application was able to beat human chess champions. But chess is a game of logic with logical moves. AI is at home in that kind of world. On the other hand, Microsoft tried to place AI in our world by having it interact with humans on Twitter. In less than 24 hours this AI system learned to be racist, spewing insults and embarrassing the company. When asked how this happened, the response was that the machine was simply learning from what was said on Twitter. Another AI robot embarrassed its makers when it said on live TV that it would never harm a human and in the very next sentence said it wanted to rid the world of water. What led it to that thought is unclear.
AI can be a great path to marrying data and machine learning to make our life easier. Certainly, for retailers it holds a great promise of improving efficiencies, driving costs down, and enhancing customer experience. Still, one really should invest time to educate oneself on what its current capabilities are and what the long-term effects will be before making any decisions to deploy AI in a big, customer-facing way.
Tom’s column regularly appears on every issue of LP Magazine. To subscribe to the printed version of the magazine and enjoy other great content visit losspreventionmedia.com