With COVID-19 creating both supply and demand issues across the value chain, retail decision-makers are dealing with too many problems at once to monitor everything manually. While artificial intelligence (AI) has become a catch-all buzzword of sorts, the technology when implemented effectively can be the difference between a retailer either competing in a post-COVID environment or falling into obsolescence.
During a recent webinar from Llamasoft, Retail Systems Research (RSR) managing partners Brian Kilcourse and Steve Rowen stated the case for an AI-enabled supply chain based on how retail’s “winners” thrived in the lead-in to the pandemic.
“Even before COVID-19, retail winners felt much more strongly than average and underperforming retailers that they had the ability to monitor supply chain capacity to model contingency plans and to stimulate the effects of supply chain changes before they implement them,” Kilcourse said.
RSR defines retail winners, which comprised 55 percent of companies studied, as generating at least 4.5 percent year-over-year sales growth in 2019. These retailers are far ahead of their peers when it comes to leveraging AI to identify network bottlenecks (58 percent vs. 19 percent), gain visibility into available-to-sell inventory anywhere within the enterprise (58 percent vs. 32 percent) and getting alerts on critical inventory situations anywhere in the supply chain (56 percent vs. 30 percent).
Successful retailers also identify different challenges within the supply chain that they believe AI can address. While 53 percent of winners are interested in creating more flexible sourcing strategies due to geopolitical issues including tariffs, only 35 percent of the others feel AI would help them address this. On the other hand, 62 percent of the average or underperforming retailers feel rapid consumer demand changes are constantly undercutting their ability to buy big and lower costs, whereas 49 percent of winners think this is a challenge.
Related Stories
These issues are more important than ever for retailers, especially as they try to navigate an environment upended by the pandemic’s spread. Kilcourse noted that today’s AI-driven demand forecasting solutions are different than the demand forecasting systems of the past because the AI engine learns from itself. Simply put, the more data it gets and the more time the engine runs, the more it learns to adjust the modeling, leading to increased precision, he said.
“This turns out to be awfully important when you’re thinking about hyperlocal demand, which COVID-19 has really exemplified,” Kilcourse said. “We’re talking to a friend of ours in the industry who is using AI forecasting in seven- and 14-day timeframes in order to direct the replenishment either to increase or decrease product flow to a store depending on the near-term need of people who were stricken by a COVID hotspot. This is a real example of how it’s being used now. Your traditional algorithmic demand forecasting isn’t going to pick that up.”
Given the unpredictability of the remainder of 2020, with uncertainty over whether stores will remain open in the event of another major outbreak, Rowen recommended that retailers model everything they can now that there may be variance on a week-to-week basis. Based on RSR data, Rowen doesn’t believe retailers are using predictive modeling enough to measure modern necessities including labor management and transportation costs across the supply chain, which he described as “a clear miss.”
The RSR report describes AI as “winners’ country” since there is such a significant difference in the manpower behind it for retail winners, compared to average and underperforming competitors. While 53 percent of winners say they have one or more data scientists on board who are competent with mathematical data analysis and predictive modeling tools, only 22 percent of other retailers can say the same. And while just 2 percent of winners say they haven’t decided on a direction for AI implementation even after internally discussing the need for experience, this number jumps to 19 percent for the others.
“This is not a small ask,” Rowen said. “We understand this is a very expensive thing. But if you have the means to afford this person, they can provide you the focus to get the kind of actual data that you need out of these tools because they’re incredibly powerful tools. But as we all know a tool is only as good as the user.”
Vikram Murthi, vice president of industry strategy at Llamasoft, told Sourcing Journal that he believes the retailers that are truly successful at forecasting do so at a granular level—ideally down to the store and even SKU—and leverage a blend of techniques comprising of machine learning and statistical modeling such as regression.
He cited the RSR survey, which noted that 56 percent of retail winners are actually using and satisfied with demand forecast modeling using external factors like macroeconomic indicators, weather, consumer trends, social media data and competitive information. Only 30 percent of the other retailers surveyed were using (and satisfied with) their demand forecast modeling capabilities.
“[The winners] also have a clear understanding of what factors drive demand—store location, demographics in the area, promotions, markdowns, seasonality, price bands to name a few,” Murthi said. “They pick solutions that have the ability to try multiple models and pick a subset of models and then blend them together with weights to reduce forecast error over time. Also, retailers that are most successful realize that it is not just about improving accuracy but leveraging this forward projection to drive inventory and replenishment strategies, warehousing, transportation and sourcing decisions.”
Murthi said fashion and specialty retailers in particular would most benefit if they had AI technologies that could power capabilities including demand forecast modeling, data cleansing and data robustness, automated forecasting processes, capacity planning and predictive analytics for supplier selection, among others.
During the webinar, Murthi highlighted a case study of how an undisclosed “multi-department” retailer improved operational decision making with responsive near-term demand forecasts. Leveraging the Llamasoft solution, the retailer used a machine learning-based forecasting algorithm across millions of SKU/store combinations at a weekly and sometimes even daily level to not only reduce stockouts, but also improve replenishment strategy with a 25 percent increase in forecast accuracy.
“This is a wake-up call for retailers, because we’re in the era of the ‘never know,’” Murthi said. “Retailers really need to monitor the impact of supply chain disruptions and create contingency plans so they can respond to those disruptions.”