As with any overhyped terms, AI and ML are thrown around in sometimes wildly inconsistent ways by various vendors, often making exaggerated claims. While the hype around the terms may be recent, any data scientist can tell you that the key concepts and algorithms have been around since the 1950s.
In simplest terms, AI enables machines to carry out tasks in a way that we would consider smart or intelligent. ML is an application of AI based on enabling machines to access data and learn for themselves without explicitly being programmed to do so, and they can continue to evolve as market and customer behaviors change. The reason they are being used much more in retail today is because of the disruption that has taken place. It’s no longer possible to succeed in retail without the supplement of science. The growth of e-commerce and hyper competitive landscape along with the explosion of customer, competitor and market data, means historical analytical tools are no longer getting the job done. The sheer volumes and variety of data along with the millions of possible scenarios, swamp a human’s ability to consume and analyze the data and make optimal decisions. Yet it’s critical for retailers to understand their digital shoppers who have global shopping and buying power with 24/7 transparency on prices and promotions.
It’s Not Your Father’s Retail Industry
We don’t have to look back very far to remember when long lead times existed, price changes were infrequent, and retailers could get away with long-standing everyday prices and repeating the same promotions year after year because “we’ve always done it this way.” By today’s frenetically paced standards, retailers once had the luxury of lead time to prepare price resets and promotional offers. They knew their shoppers and their degree of loyalty to favorite retailers, brands and products. In other words, their shoppers were well-defined and pretty predictable. And the Economics 101 law of supply and demand dominated: in times of high demand and low supply, retailers could raise prices and consumers would pay them to get their hands on a hot new item or label. This is no longer the case. Long gone are the days where retailers can succeed by employing these types of pricing policies and assumptions.
It all sounds so quaint in today’s harsh retail reality. First, we can say goodbye to traditional demand/supply economics. A recent Forrester global shopper study commissioned by Revionics revealed that when retailers raise prices due to limited availability, 59% of shoppers said they would not purchase, wait for a better price, or shop at another retailer. That makes supply-based price hikes incredibly risky for retailers – in terms not just of lost sales but also in terms of losing long-time customers completely.
For today’s hyperconnected shoppers, retailers must reach them on their own terms, across a variety of channels, selectively using each promotional vehicle (circulars, endcaps, websites, mobile, social networks, etc.).
The good news in all that data is that retailers have access to unprecedented detail on their shoppers’ behavior, competitive data and market evolutions. But the data is only valuable if you can mine and analyze it meaningfully.
AI, ML and the Ethical Dilemma
Technology has caught up with data volumes and compute power to give retailers the opportunity to mine all of this data, run billions of possible scenarios dynamically, and systematically deliver optimal prices and promotions that will best serve their customers as well as their strategic and financial business objectives.
The ethical dilemmas fall broadly into two categories: one relevant to vendors providing the AI- or ML-based solutions to retailers, and one relevant to how a retailer uses the tools. With respect to ethics in designing AI pricing solutions, a critical element is complete transparency – retailers should be able to clearly understand how the technology arrives at its decisions and recommendations.
The reason the first generation of price and promotion optimization failed to take root long-term is because they were black-box technologies: they did their calculations in the background, offering up recommendations to retailers with no context, thus creating uncertainty as to their effectiveness and accuracy and the science was not productized (among many other reasons). Fortunately, the new most advanced generation of AI and ML has learned from the anemic adoption of the earlier approach, and today the technology proudly leads with complete transparency and AI models can dynamically learn and regenerate themselves. There is also an important consideration that organizations should be able to adjust sensitivities and settings as needed. In today’s solutions, retailers can “turn the knobs” for themselves, making their own trade-offs between, say, unit volumes and margins to enable different strategic price and promotion approaches for different categories, items, zones or customer segments. They can also run detailed simulations of various approaches and the impact of each, zeroing in on the most effective combination “before” committing to that price or promotion.
Another key aspect of AI is the notion that systems should continually learn and improve. The beauty of this flexible technology is that it senses real-time demand signal changes, shifts in market factors, and evolving competitive practices to respond before humans can detect them. It’s critical to look for AI that has proven its ability to detect and respond to these signal shifts.
AI Ethics in Action: Dynamic Pricing
At the end of the day, AI ethics require that humans retain ultimate control. An interesting application of the intersection of self-learning and human control is dynamic pricing in retail. Since this is another term that’s thrown around with casual definitions, let’s start with the definition of dynamic pricing, which is:
Targeted and smart, updating those items where shoppers are most sensitive to prices and to competitive offerings.
Flexible in frequency, allowing price changes to happen at a speed that matches your business parameters.
Structured for fast, automated processes leveraging self-learning, science-based algorithms and predefined workflows.
How can retailers take advantage of automation while still ensuring appropriate human oversight? By setting bumpers, which are defining tolerances so that any recommendation falling outside these preset limits are flagged for manual human review. With this approach, retailers get the benefit of being able to respond with agility while still having the assurance that any recommendation that violates business rules or other guidelines will trigger manual intervention to either accept or reject the recommendation.
The Retailer Factor: Shopper-Facing Ethics Considerations
Well-applied, transparent technology gives retailers confidence in the power of an AI-based pricing solution, but how do shoppers view ethics with respect to these systems? First, it always comes back to price: the Forrester research found that price was the #1 factor shoppers cited when deciding where to shop, across all retail sectors including grocery, apparel, DIY, and convenience.
The studies also showed that shoppers are more than ready for dynamic pricing, with 78% saying they are comfortable with the use of data science to determine prices as long as they receive a fair price for the product they are purchasing. In fact, only 6% of respondents say they don’t think it is fair at all for prices to change dynamically. The idea of fairness in pricing is critically important, with 59% of shoppers reporting that they would refuse to purchase an item if they perceived the price as arbitrary. They accept price increases or decreases that remain within the “fair” range if they are based on data science – that is, driven logically and not arbitrarily.
A key ethical consideration is when and how to apply personalization in pricing. Although 65% of shoppers appreciated personalized prices, 47% of those shoppers also stated they would be angry if someone else received a better price, according to the Forrester research commissioned by Revionics. Further, 59% of shoppers reported that they would refuse to purchase an item if they perceived the price as arbitrary. Retailers may be best off starting off with logical customer segmentation to give prices and offers that, again, seem fair and relevant to shoppers – for example, a pet supply retailer could logically provide different prices and promotions to cat owners than to dog owners.
Clearly, retailers who want to remain relevant to shoppers while maintaining a healthy business can and should rely on contemporary AI-based price, promotion and markdown optimization technologies. But as Spiderman’s Uncle Ben cautioned, “With great power comes great responsibility.” Retailers should absolutely tap into the power of optimization solutions, but give careful consideration to the importance of ethical transparency in both their technology selection and in how they apply it to constructively engage their customers with fair, relevant pricing.
Marc Hafner is Chief Executive Officer of Austin-based Revionics, a leading provider of pricing, promotion, markdown and space solutions.