Jonathan Landy
At VarietyIQ, we’re obsessed with optimizing complex assortments – catalogs in categories like apparel, wine, and beauty, where many products compete for the same customer attention. In these cases, a product should only be carried if it satisfies demand that would otherwise go unmet, lifting net sales. But that kind of uplift impact is harder to measure – and far less intuitive – than a product’s quality in isolation. As a result, even experienced merchandising teams can struggle to get the full assortment mix right, leading to redundant purchases and missed opportunities to better serve customers.
By addressing these inefficiencies, VarietyIQ helps retailers to simultaneously improve customer experience and operate more efficiently.
This post kicks off a two-part series exploring the challenges and opportunities around complex assortments. Today, we’ll define what makes an assortment “complex,” explain why it’s difficult for humans to optimize without algorithmic support, and review a real-world example through this lens – Baskin-Robbins’ current lineup of ice cream flavors. In our next post, we’ll consider why some product categories support complexity while others do not.
Complex vs simple assortments
In a simple assortment, products target different segments of demand and so rarely compete for customer attention. For example, at the hardware store, screwdrivers don’t compete with hammers. Even among hammers – if more than one is carried – the options usually sit at different price points or serve distinct use cases, which limits overlap. Optimizing a simple assortment still takes work – you need quality products at the right prices and right time – but interactions between items are minimal.
Complex assortments are different. Here, many products vie for the same purchase. The wine aisle at a grocery store is a perfect example: hundreds of subtly different options competing side by side. The same holds true in many other grocery aisles like juice, cheese, and cereal (but not all aisles – we’ll explore that in our next post). In these cases, optimization becomes significantly more challenging – and is often poorly handled in practice.
Figure 1: In categories like salty snacks, many similar products compete for the same customer attention. Because of this, sales often cannibalize each other, and a product’s individual sales don’t reflect its true value. To optimize assortments like this, we need to estimate how much each product adds to net sales by addressing otherwise unmet demand.
How to optimize a complex assortment
When you add a product to a complex assortment, it’s not enough for it to simply generate sales. It needs to lift net sales. The right product to add next is the product that lifts net sales the most. This is the challenge when optimizing a complex assortment – and it’s a tough one:
-
Although it’s easy to measure sales of a product, it’s generally not easy to measure what fraction of those sales are cannibalized from the rest of the assortment – the key measure. To be additive, a product doesn’t just need to be good, it must be better than the existing options – at least for some customers.
-
The products that compete vary by customer. In apparel, one shopper may care about color, another about fit, another about novelty, another about price. Products don’t have to look similar to compete for attention.
-
Customer segment sizes matter, too. To decide whether a product is worth adding, you need to consider both the value it adds to its target audience as well as the size of that audience. This is because the segment size acts as a lever arm, amplifying the revenue impact of improvements for that segment.
This means we can afford to offer more (less differentiated) variants to larger customer segments. For example, a 10% lift to sales in one segment is just as good as a 50% lift to another, provided the former is at least 5x the size of the latter.
Similarly, after sufficient investment in support of larger segments, it pays to begin investing in smaller customer segments. These options are often overlooked because they sell at lower volume than those aimed at larger segments. However, since we won't carry many products like these, the products aimed at the smaller segments are less cannibalistic, and so can often drive relatively large increases to net sales.
To drive profitability, delight customers, and reduce waste from unsold inventory, we have to focus on the products that meet new demand. Intuition can help, but it’s rarely enough to fully reason through all the interactions needed to estimate the true uplift of each product. As a result, strong inefficiencies persist in most real-world complex assortments (we’ll share examples in upcoming posts). The silver lining is that these issues are solvable – typically with algorithmic support – and tackling them can unlock meaningful gains in efficiency, customer satisfaction, and growth.
A review of the menu at Baskin Robbins
To explore how these ideas apply “in the wild”, let’s now take a look at Baskin-Robbins – the well-known American ice cream brand. Their website currently lists 42 flavors. A few observations:
-
A number of the current flavors cater to chocolate lovers: Non-Dairy Cookies and Cream, Oreo Cookies ‘n Cream, Chocolate, Chocolate Almond, Chocolate Chip, Chocolate Chip Cookie Dough, Chocolate Fudge, Chocolate Mousse Royale, Peanut Butter ‘n Chocolate, Rocky Road, and World Class Chocolate. That’s 11 of 42 flavors with heavy chocolate presence.
Given our concern for redundancy, is this overkill? Including the eleventh weakest chocolate flavor here hardly seems likely to drive much marginal benefit over the prior ten. E.g., if World Class Chocolate were to be dropped, I wouldn’t be surprised if upwards of 90% of its sales simply transfered elsewhere. Taking this to be true, its inclusion would only make sense if the chocolate-loving segment is very large – probably a reasonable guess. This would then offer a large lever arm, justifying a deep bench of slightly varied offerings that cater to different tastes within that group.
-
In contrast, only a handful of flavors seem aimed at the bright-color, fun-loving segment. It’s hard to believe that any of these can match the sales volume of any of the chocolates (I’d definitely take any of the chocolates over Wild ‘n Reckless Sherbert). Nevertheless, these probably make sense to include here because they’re truly additive – helping to meet the demand of an otherwise neglected segment.
-
Baskin-Robbins balances fixed classics with seasonal rotations, as noted on their site. As we discussed in a prior post, this is smart: New flavors keep things fresh for returning customers, while maintaining fan favorites supports new customer conversion and repeat purchasers.
Baskin-Robbins has been around for 80 years (founded in 1945 in Glendale, CA). I’d wager they’ve done a lot of experimentation to get their lineup dialed in. But is it perfect? Could they swap the weakest chocolate for a fruit-forward option and lift net sales? To know for sure, we’d have to carry out a careful analysis of their sales data. How about your local bakery’s pastry assortment? Or your favorite apparel shop’s hoodie wall? Are those assortments as carefully optimized?
Figure 2: Baskin-Robbins currently lists 42 flavors on their site. Of these, 11 are chocolate-heavy, while a few lean more playful. In the post, we consider whether this mix makes sense from an incrementality perspective.
Summary
Complex assortments appear everywhere: in ice cream shops, wine aisles, bookstores, and apparel catalogs. The subtle interplay between customer segments, overlapping product attributes, and finite attention makes this a rich area for optimization. Improvements here boost business performance, lead to better customer experiences, and reduce waste – removing redundant products that often end up unsold. They’re also fun to think about!
This is what excites us at VarietyIQ. We believe thoughtful algorithms, paired with human expertise, can unlock major gains in how products are selected, curated, and sold.
Thanks to Jaireh Tecarro for creating the images for this post.
