VarietyIQ: Smarter Personalized Merchandising

Introducing VarietyIQ’s Personalized Curation Engine: Our solution helps retailers curate product assortments that efficiently serve every target customer segment. This is especially critical in sectors like apparel, grocery, and beauty, where uneven support across segments creates significant inefficiencies. By integrating VarietyIQ into existing workflows, merchandisers can seamlessly build well-balanced assortments, reduce redundant spending, and drive meaningful sales growth.

The heavy cost of status quo buying methods

Every retailer is aware of the large expense needed to maintain an inventory – and that losses due to unsold goods can be even more substantial. Some well-circulated stats on this:

• 20-40% of annual revenue is generally locked up in inventory at any given moment for a typical retailer.
• In some industries (such as apparel), a whopping 30% of units go unsold each year and need to be cleared.

These numbers are not just lines on a spreadsheet, but actual physical reality. E.g., products that go unsold often end up in landfills, some of which have gotten so large that they can be seen from outer space! These losses aren’t good for the environment, and they’re also not good for business.

While these large costs are widely appreciated, it is less well-recognized that standard-practice buying strategy unintentionally drives these issues. As we explain below, this is because standard practice results in uneven customer support:

• The legacy method overbuys for core customer segments. This redundant buying leads to unnecessary expense and also drives the unsold goods issue.
• At the same time, legacy underbuys for the long tail of smaller customer segments. Taken together, these smaller segments represent significant missed growth and revenue potential.

We see then that uneven buying leads to two sources of inefficiency: The visible expense and loss, and also the unobserved missed upside – often the larger of the two factors.

Tyranny of the majority

The goal of merchandising is to develop well-balanced assortments that cater to each customer profile, with investment set across segments so as to maximize net sales. At scale, this is easier said than done. With hundreds or even thousands of products in a catalog, it’s impossible for any human to fully grasp all the details necessary to maintain balance.

To attempt a reasonable solution, the standard approach is to first consider the products within each "class" separately (e.g., dress shirts might represent one class in an apparel company, trousers another, etc.). Next, to simplify the problem further, products are ranked within each class by expected demand. A buyer will then curate from options at the top of each list, applying their expertise to select likely winners for the next season. As a final pass, the team will review the resulting full assortment by eye, identifying needed adjustments to ensure a good color and price point mix.

These methods are certainly a step in the right direction. However, they go wrong in two important ways:

First, retailers often have a “core customer” segment – a large group of customers that share a common set of preferences. Their choices dominate the expected top-sellers list. By primarily purchasing inventory from the top of these lists, retailers then inadvertently end up allocating ~100% of their budget to just ~60% of their customers - the core group. This leads to intense competition among the products targeted at these customers, with many contributing minimally to net sales increases. Further, even core customers have a limited budget and so can only sample from the options available to them - this leads to excess inventory and clearance.

Second, the effort to ensure a good color and price-point mix is an admirable nod to attempting balance. However, inspecting actual customer preference data shows that tastes vary across a surprisingly large number of qualities (more on this in a future post). As a consequence, this step is not nearly enough to ensure “there’s something there for everyone who shops here”.

Given these challenges, there remains a consistent, structural inefficiency across retail catalogs. To address this, we must equip our buyers and planners with new metrics that complement demand forecasts, providing visibility into where investement is uneven and where opportunities exist to meet new demand.

Precision Buying to Match Inventory to Demand

In our past roles as senior data-scientists at Stitch Fix, Greg and I – the co-founders of VarietyIQ – were exposed to the ubiquitous challenges I’ve reviewed above. We created VarietyIQ’s Personalized Curation Engine as a solution to directly resolve the structural inefficiency associated with uneven buying. Although the math behind this is somewhat involved, the high-level ideas are simple:

• Step 1: We apply AI / ML methods to develop a personalization model that understands the preferences of individual customers.
• Step 2: Making use of that model, we then develop metrics to quantify whether a product meaningfully lifts the experience of many individual customers, driving new sales. If so, great! If not, the product is flagged as redundant.
• Step 3: Finally, we leverage these metrics to create first pass buy-sheets that maximize net business sales (buyers can then work atop these, editing as desired).

Two wonderful things happen when you take this approach.

First, some products are surfaced that may not top the charts in direct sales but that do lift net sales significantly. When we present these options to buyers, we always see heads nodding. These products are clearly of high quality and complement the overall assortment well. In other words, these are precisely the products that align with buyers' instincts and the core goals of merchandising - yet they were previously overlooked by the legacy data approach. Sales data does actually support and validate these instincts, but only when viewed in an appropriately segmented way.

Second - by design - our approach ensures your budget will drive maximal lift to net customer experience: The core client is supported just as well as through standard practice, but redundant spend is cut and a small portion redirected to better support smaller customer segments. As a result, we've seen this drive more sales, at lesser expense, and with fewer goods going unsold, all at scales important to the business’s bottom line. Again, we’re able to achieve all three of these outcomes simultaneously because we’re correcting a large, structural issue in the current approach.

Optimize for Balanced Growth

Retailers today face a major challenge: how to curate assortments that effectively serve all relevant customer segments while minimizing waste and maximizing sales. Traditional buying methods rely on broad demand forecasts that overserve core customers and underserve smaller segments, leading to both lost revenue and excess inventory.

VarietyIQ’s Personalized Curation Engine solves this by applying AI and machine learning to understand customer preferences at a granular level. Our solution identifies which products genuinely expand demand versus those that merely compete with existing options, ensuring a more balanced and profitable assortment strategy. By integrating this approach into existing workflows, merchandisers can reduce redundant spend, increase sales, and cut down on unsold goods—optimizing both business performance and sustainability.

That's VarietyIQ - Smarter Personalized Merchandising.

---

More on this topic in future posts. We'll also be writing on other aspects of the art & science of merchandising as they catch our attention. Interested in keeping up with our latest insights? Sign up for our newsletter below to receive alerts when new posts are published.

Thanks to Callie Ryan, Kat Faley, and Greg Novak for feedback on an earlier draft of this post, and to Bernie Chu and Jaireh Tecarro for supporting our site refresh.