Q&A With WWD | Mexico's Largest Department Store Chain Coppel Expands AI Merchandising Pilot Program to Footwear
Coppel is in the midst of a $4.6 billion transformation strategy, and using artificial intelligence (AI) as a tool to help with merchandising is at the top of the list of priorities.
Mexico’s largest department store chain has tested predictive AI using First Insight’s platform in women’s private label apparel across a product range of 462 items. Coppel is now expanding that pilot program to include private label footwear across men’s and women’s. The platform predicts how likely customers are to purchase a product based on direct consumer input, making it an analytical tool to guide Coppel merchants on which private label footwear products to develop, and how to position and price the items before bringing them to market.
Footwear News interviewed Daniela Orduña, divisional merchandise director at Coppel, and Viki Zabala, chief strategy and growth officer at First Insight, to get their perspective on using predictive AI in footwear merchandising.
FN: Has Coppel used AI technology before working with First Insight for other operations, such as supply chain management or sourcing?
Daniela Orduña: Yes. Coppel has integrated AI capabilities across multiple parts of the business. We’re a large omnichannel retailer with a significant financial services business — Coppel stores, BanCoppel, and Afore Coppel — and AI plays a role in areas like eCommerce search optimization, WhatsApp conversational commerce and AI in the supply chain modernization. What First Insight brings is a different application of AI than what we’ve used historically. The AI we’ve deployed to date has largely been operational — optimizing what already exists. First Insight’s predictive AI works on what doesn’t exist yet: it forecasts how customers will respond to products before we produce them. That’s a new capability for us, and it’s where we believe the next layer of competitive advantage lives.
FN: What was the reasoning behind deciding now to use a predictive AI tool for product development and merchandising?
DO: Our commercial teams have always relied on a combination of professional expertise, historical sales data, and direct customer feedback to make product and merchandising decisions. As customer expectations continue to evolve and the retail environment becomes increasingly dynamic, we saw an opportunity to strengthen this decision-making process with predictive AI. By integrating First Insight into our merchandising and product development workflows, we can make more informed decisions, deliver products that better meet customer needs, and create more relevant and engaging shopping experiences.
FN: Has Coppel done any testing on the use of predictive AI for product development and merchandising? And if so, what have been the results?
DO: We started with a pilot in women’s apparel, which is our largest category, where we tested an assortment of hundreds of styles. It has already provided our commercial teams with valuable additional insights for decision-making regarding brands, products, and trends. We’re now expanding the work into additional categories — including men’s apparel, and women’s and men’s footwear — using the same predictive consumer intelligence approach to guide assortment and pricing decisions at every step. We expect to have the definitive results of this pilot by the end of this year.
FN: What is the feedback from consumers? Is it based on what they buy and how much? Are focus groups used? Is there some section somewhere where consumers can post commentary about a product? How do we know that the data is representative of the targeted consumer base?
Viki Zabala: For any product a retailer is considering, we collect direct consumer feedback in a closed digital research environment from a targeted audience that matches their customer base. Consumers see the product and answer structured questions on purchase intent, perceived value, and pricing — plus open-ended commentary, which more than half of respondents provide.
Our predictive AI validates and weights each respondent’s signal using methodology built over 20 years of correlating consumer feedback with actual market outcomes. It then converts the validated signal into a set of predictive outputs: value score, model price, demand curve, and price elasticity curve. Those outputs are predictions of how the market will actually respond, not survey results.
Retailers tell us who their target customer is — whether that’s a certain age range, region, income, shopping profile, lifestyle, or any mix of those — and we build that audience from our network of 360+ million real consumers across 180 countries and 67 languages. We can also overlay multiple consumer profiles in the same study, so the [system] accounts for how different segments behave.
FN: What exactly is included in the First Insight platform? How does it know about fit? For footwear, is the platform asking for direct consumer input or is it looking at what shoes are returned and what is reordered and kept? And what are some of the examples of data points on shoe styles that are collected? Is it open toe v closed toe? Heel height?
VZ: The platform is built around direct consumer input on products before they go to market rather than returns data, reorder analysis, or historical sales alone. We capture the consumer signal [for] pre-production, then our predictive AI converts it into a set of outputs retailers use to make merchandising decisions.
For any product, the platform produces four core predictive outputs: value score: a 1-to-10 predictive ranking of how likely a product is to sell at full price; model price: a forecasted average selling price, often projected 8–16 months in advance; demand curve: [total market potential] demand at the product level, and price elasticity curve: [what can actually be fulfilled] across price points, with markdown cadences modeled in.
For footwear specifically, retailers decide which attributes they want graded. Consumers see the product — photography, renderings, [computer-aided designs], etc. — and answer structured questions on silhouette, colorway, material, toe shape, heel height, sole construction, brand cues, and any other attribute the retailer wants feedback on….
Physical fit testing still happens at sampling and wear-testing. What predictive AI catches in pre-production is whether consumers expect a particular silhouette to fit, whether a width or strap detail creates concerns, and whether the design signals quality or cheapness. Those early reads often catch issues that would otherwise surface as returns later.
FN: How do you ascertain upcoming trends using predictive AI in footwear [and] how does predictive AI also inform Coppel on what the pricing threshold should be for any given shoe style?
VZ: We’re measuring what consumers are telling us right now and using that to help retailers make better decisions about what’s coming next.
When thousands of consumers consistently respond positively to certain product attributes, styles or price points, patterns start to emerge. Retailers can see which concepts are gaining momentum and which ones don’t have the demand they expected. That’s how you start to see something like “consumers are responding more strongly to dress shows right now than fashion sneakers” and it shows up in the signal before it shows up in sales.
Pricing is a big part of what the platform does. For a new product, we can identify the optimal price point based on consumer feedback, essentially the price at which demand is strongest. We can also ask consumers directly what they’d be willing to pay, and if their purchasing intent changes if the price is raised or lowered by a certain amount. That gives retailers a much clearer picture of pricing thresholds and margin opportunities than waiting to see how a product performs at markdown.
FN: Tell me more about how you can use predictive AI for merchandising in footwear? What about how to determine early which styles to put in production?
VZ: Most retailers today are making merchandising decisions using a combination of historical sales, trend forecasting, merchant expertise, and competitive analysis. Those are all important inputs, but they all look backward, meaning they can only tell merchants about what already happened. Predictive AI adds a forward-looking input layer: what real consumers say about a product before that product is produced.
This matters more in footwear than in almost any other category. Footwear has the highest return rate of any DTC (direct-to-consumer) category at around 31 percent, the lowest online conversion rate of major retail categories, and markdowns punish missed bets disproportionately. Every product decision before production carries more economic weight in footwear than in apparel — which is exactly where pre-production predictive signal earns its keep.
Mechanically, here’s how it works for a footwear merchant. The merchant has a line review with 10, 50, or 100 candidate styles competing for production slots, which they then run through First Insight. The AI returns a ranked value score for each, a model price prediction, demand depth, plus a reach-and-penetration analysis that tells them which combinations of styles work best together as an assortment. The system also documents [the products that] will underperform, which is often the call merchants care about most, because killing the wrong bets early frees capital for the right ones.
FN: And if you are looking at what to produce, given the lead times of about one year for footwear, is the predictive analysis mostly for within that one year timeline or can it make predictions that go out a bit further (let’s say you want to buy raw inputs early to capture advantageous price points for use much further down the line)?
The sweet spot is helping retailers make decisions during the active product development and merchandising cycles when they’re deciding what to source, produce, and design. Our predictive AI forecasts demand and price elasticity 8 to 16 months in advance, which is the same horizon a footwear merchant is operating in when they make production decisions. So the signal is timed exactly for the merchandising cycle.
The platform isn’t a commodity-pricing tool, but the forward-looking consumer signal we collect is directly useful for longer-horizon sourcing decisions. The earlier a retailer can detect a durable shift in consumer preference, [such as] a specific silhouette, material, color direction, or use case, the more confidently they can pre-commit to raw materials and capacity.
Read on WWD.
