Cognitive Computing Pricing Tool Protects Profits

In a weak economy with sluggish sales, retailers often cut prices to attract cost-conscious buyers. The challenge is devising markdowns that generate sales without gutting profits.

The Ermes Group, which operates more than 70 retail shops in Cyprus and Greece, is succeeding despite the difficult economic climate of the past few years. “It’s a vicious cycle,” says Sophocles Sophocleous, purchasing manager of Ermes Department Stores. “We need to be aggressive with promotions but find ways to protect the bottom line.”

His team of 20 merchandisers and product managers was adjusting prices manually, a time-consuming process. “They would go through spreadsheets line by line to decide which items to mark down and by how much,” he explains. Timing was critical. If markdowns were too late, when demand was declining, merchandise wouldn’t move, forcing deeper price cuts.

“We realized we had to be more tactical with markdowns at the right time and at the right percent,” Sophocleous says. They also wanted a more efficient process.

Lessons Learned From Other Retailers

After considering various technology solutions, the company’s IT team chose the cloud-based IBM Watson Price Optimization service. Sophocleous, a 17 year-veteran of Ermes, traveled abroad to visit other retailers using this technology. “Our research convinced us that this was the right tool for us,” the purchasing manager says.

Ermes began discussions with IBM in June 2015 and spent about six months on prep work, submitting two years of data on sales, stock, costs and profits so IBM could build a statistical model. Sophocleous’ team created a hierarchy with mapping for 200-plus brands, each with its own hierarchy. One brand might comprise tops, outerwear, accessories and more, for example.

In January 2016, Ermes launched a trial involving a few departments. “The system saved us a lot of time and had more pros than cons, so we decided to move ahead,” Sophocleous recalls.

The IBM Watson tool proposes a list of items to mark down, the percent and the timing. The staffers who used to doing this manually weren’t immediately on board. “I needed to convince them to rely on and trust the tool,” he adds.

For every recommendation, Sophocleous’ team could provide feedback. He had to decide whether an objection was logical or whether sentiment was coming into play. “For example, our merchandisers might be inclined to protect a certain brand,” he says. “But if an expensive designer bag isn’t moving because of its color or style, we need to address that.”

There were valid reasons to overrule some recommendations. “An experienced manager wouldn’t mark down heavyweight knitwear while the weather is still mild, because sales are likely to pick up once colder temperatures arrive,” he explains. “And contractual obligations with premium brands might not allow us to be aggressive in markdowns.”

Some adjustments were necessary. In the initial mapping work, the Ermes team had to group disparate products to provide enough data in some categories. “We found large deviations in forecasts and actual sales, mostly in groupings that didn’t make sense to us,” Sophocleous says. “We drilled down and corrected those groupings. It was a major lesson learned.”

Automation Saves Time and Money

It took about six months to fine-tune the process and build trust among the team. “Now, a year later, we rely fully on the tool,” Sophocleous reports. “We apply human intelligence by exception, based on facts rather than emotion.”

Early-warning reports enable the retailer to adjust prices at the right time. “If sales of an item are slow and won’t clear inventory by the end of the season, we are better off giving a 20 percent discount early in its lifecycle rather than waiting three months and then selling it for 60 or 70 percent off,” he explains.

Time savings have been significant. “It used to take at least two weeks to prepare a sale; now it takes three days or less,” Sophocleous declares. Those time savings free up staff members to focus on other aspects of their job.

Perhaps more important, profitability has improved. “Compared to last year, our markdown spending is much less,” he reports, noting that Ermes has reduced purchase volume to reflect slower sales.

Sophocleous likens the IBM Watson tool to “an automated merchandiser working on our behalf.” He adds, “We are happy and satisfied. This has limitless capabilities, and we are still learning. We see room for continued improvement.”