When used effectively artificial intelligence could help retailers accomplish a perennial priority—cutting costs.
Among retailers that adopted enterprise-grade, AI-driven applications for assortment planning, 76 percent reported cost savings of at least 3 percent or more, according to a June survey from Coresight Research. The cost benefits were even greater for 36 percent of these businesses, at 5 percent or more savings.
But despite the numbers tilting in the right direction, more retailers than not are still losing sales and carrying more inventory than they should due to a lack of AI-driven forecasting and promotion optimization platforms.
More than 80 percent of the 170 respondents surveyed estimated lost sales of at least 3 percent due to allocation issues such as infrequent stock replenishment—an issue that has become even more of a concern as supply chains remain constrained. Fifty-seven percent of the respondents said that at least 3 to 6 percent of total sales were lost, while 18 percent believe allocation issues negatively impacted as much as 6 to 9 percent of sales.
“Honestly retailers are still stuck in that ’80s and ’90s mentality around allocation,” Prashant Agrawal, CEO of Impact Analytics, said during a recent webinar. “You want to make sure that you’re getting that right product to that right place, and you can’t do that in Excel, it’s just not humanly possible. You’ve got to rely on the AI and machine learning, and once you do, you actually see the results like this. The reality is, and you can see it in the earnings reports for some of them, whether it’s Target or Walmart, AI is actually helping their bottom line. The reason to do this is that it makes you a lot of money.”
And despite the industry’s pursuit of lean inventory to prevent overstocking and reduce markdowns, excess inventory remains a widespread problem. As many as 73 percent said they had at least 5 percent excess inventory on average after a selling period, with a minimum of 10 percent overages at 30 percent saying they had 10 percent extra at minimum.
During the presentation, which was hosted by Coresight founder and CEO Deborah Weinswig, Agrawal said that lost sales and excess inventory are indicators that retailers, particularly at the chief merchant level, are doing too much and preventing AI from fully taking the wheel.
“Lost sales should be minimized. Will they happen? Yes, but if you’re using true science, it should be very much minimized,” Agrawal said. “The same thing with excess inventory. When you look at that, the allocation is pure science. If you’re touching this too much, you’re already lost around this, and that’s what we see with teams. You really want them to make this the autopilot part of the process.”
Analyzing holiday planning at JoAnn Stores, Agrawal pointed out that the specialty retailer changed a decision to allocate more beach ornaments to coastal markets after learning from the Impact Analytics platform that beach motifs actually sell better inland.
“If I’m at the beach, I don’t need that on my Christmas tree,” he said. “If I’m sitting in Chicago, I want to remember my beach holiday that I had in San Diego or Miami or Tampa. And that’s what helped them actually get more sales, because they got the beach ornaments to the right place. Again, pure science.”
Heightened demand resets promotional strategies
Markdown mitigation has been a common theme for this industry this year, but changes have largely been a byproduct of crushing demand, rather than better inventory management practices. That means retailers have to rethink how they prep their promotions, particularly given the upcoming holiday season.
When it comes to the approach, Agrawal had one piece of advice: reset consumers’ expectations around pricing and promotion, and stop repeating deals from yesteryear.
“Guess what? Sales are good, margins are good,” Agrawal said. “We’re seeing and working with retailers to reset that expectation. If they’re used to having everything 40 percent off at an outlet, let’s do 20 percent. Let’s make 40 percent the new ‘Oh my god.’ You got that chance and if you don’t do it now, you’re not going to be able to do it next year or the year after.”
Only 15 percent of execs reported that they employ real-time promotion optimization with dynamic strategies based on category targets such as margin, sell-through and revenue. Of the remaining respondents, Coresight said that 28 percent use long-term planning focused on static promotional campaigns—meaning those that are not updated based on either SKU-level performance or external changes in consumer demand.
Twenty-seven percent plan on a short-term, or weekly, basis with limited long-term clarity. Nearly one-third (31 percent) focus on medium-term, usually monthly, planning strategies yet have no strategy for promotion optimization.
“Merchants are often very focused on their plan, their day-to-day or their week-to-week, but are not understanding of the product lifecycle, that this is a 12-week game, it’s a 10-week game, and how to actually maximize the promotions throughout that lifecycle. Retailers are leaving a lot of money on the table,” said Agrawal.
Retailers prefer gut feel or Excel to machine learning
When it comes to forecasting, there’s still plenty of room for improvement. Like many areas that need improvement in retail, much can be done on the technology front to enact change, but organizations are typically slow to adopt new solutions given the cost and time involved in making changes.
According to the report, only 17 percent of surveyed retail execs indicated they use more advanced forecasting techniques such as real-time machine learning-powered models. The highest percentage (37 percent) use Excel-based forecasting to calculate formulas, while 26 percent use statistical models such as time-series forecasting. Another 20 percent, somehow, still use manual “gut-based” forecasting models.
Coresight estimates that advanced demand forecasting could measure at 90 percent or greater accuracy, making it even more jarring that so few retailers leverage the technology. On top of that, the Covid-19 pandemic showed that using the prior year as a base model to make future projections was sorely outdated, according to Agrawal.
“Nobody predicted Covid, but when you use AI and ML models—we have 15,000 models running with different stores, different SKU counts, all of that—and actually pinpoint the best way to predict what will happen in the future based on the recency of what’s happening with that SKU, that’s proven incredibly accurate for our clients,” Agrawal said. “The challenge, though, is that machine learning is still machines. You actually have all of this data and algorithms being computed, and now you have a result, so you’re not actually understanding which model is being used at any given point. It takes retailers that slight adjustment of, ‘Oh, I’ve got to trust it to get to the next level.’”