We’ve all read and heard the many recent stories of A.I.-based computers driving cars, defeating the human champion at GO (a profoundly complex board game that originated in ancient China) and beating humans at ATARI games.
The branch of A.I. for autonomous decision-making with the object to achieve a long-term goal is referred to as “reinforcement learning.” By way of example - long-term goal for self-driving car could be to stay on the road without crashing. A long-term goal for playing GO is to win the game. A long-term goal for a retailer is to maximize net income growth over a year while maintaining positive monthly net income. Predictive analytics is not capable of solving these kinds of problems.
A.I. systems based on reinforcement learning can assist retailers with many aspects of merchandise planning, i.e., decisions around what products to promote on any marketing channel, what prices to charge, how much inventory to allocate, the optimal assortment, the optimal space plan, etc.
When it comes to merchandise planning, the current volume of decisions is enormous. Consider that a retailer with 10 price zones, 10 ad zones, 50 stores, and 50,000 products is approximately 3 million per week (1,000 promo items at 10 ad zones plus 50,000 prices at 10 price zones plus 50,000 inventory decisions at 50 stores). If one had 50 merchants, that means the merchants would have to make 125 decisions per minute, for 40 hours per week. The number of potential decisions to evaluate would be 10>3,000 i.e. far, far beyond human capability. What these numbers illustrate is the practical impossibility for humans to execute optimal decision making in retail -- without assistance.
This is where autonomous merchandise planning A.I. comes in. A.I. offers the capability to make decisions at that volume, and evaluate a very large number of possible options each week, with a view to achieving sales or net income growth over the course of a year or more while minimizing monthly volatility. The technology to do this exists today, and we can report firsthand that some forward-thinking retailers are deploying this capability.
This does not mean A.I. will put merchants and merchandising professionals out of work. On the contrary. Human beings assisted with A.I. will be able to more closely focus on strategic priorities. Activities such as negotiating with vendors and identifying new and interesting products, and improving store design and customer service. Also, computers do not handle ambiguity very well. Humans are needed when making decisions where there is limited data, making decisions when last minute exceptions occur, and providing oversight to autonomous A.I. systems, in addition to many other situations where ambiguous decision making occurs.
Autonomous A.I. merchandise planning systems can transform a 1% net margin industry into a >5% one, and retailers using A.I. to grow net income will be able to invest more in price and innovation to stay competitive and maintain historical net margin levels. And, assuming incomes don’t actually decline and stay flat, or, continue to grow, consumers will see cost of groceries decline as a percentage of their disposable income.
Some additional capabilities that A.I. systems need to offer is fault tolerance and automated quality control. Self-driving cars, drones, and spacecraft have backup computer systems to ensure that the system continues to operate in all conditions. When 2 out of 3 computers agree, there needs to be a vote to decide what the system should do. With A.I., automated quality control is a necessity, in order to ensure the all the decision making is sane and behaving within some historic norms. Autonomous systems should not be allowed make extreme mistakes. These types of issues have long been part of the mainstay conversation in engineering and control systems theory. In fact, control theory and optimal control are in the foundations of A.I. when discussed in academia but these have yet to make it to the mainstream media as topics of interest. Maybe accidentally ordering one million pounds of bananas too many is not as interesting as a Mars lander crashing into the planet.
All in all, we see that artificial intelligence can and will have a profoundly positive and exciting impact in the retail industry, and those companies who lead with A.I. will survive and remain competitive. And it’s not just the retail industry, or even just business and commerce. Consider society as a whole. Consumers will enjoy favorable pricing, higher quality and better service in the coming years, and have more disposable income, A.I. applied to its maximum in all retail (and all industries) will minimize poverty and improve quality of life for people.
Gary Saarenvirta is the founder and CEO of Daisy Intelligence, a Toronto-based provider of artificial intelligence solutions designed to improve business decision-making.