Editor's note: Stephen Needel is managing partner at Atlanta-based Advanced Simulations LLC.

Sometimes, in our effort to make things simple, we go a bit too far in our explanations. A recent Quirk’s article on researching product assortment by Mark Travers (“Using consumer data to optimize product assortments,” December 2017) is an example of this situation.

Let me state at the outset that Travers’ article is well worth reading and offers some very good advice about doing this type of research. He makes a number of solid points, including: product assortment is a critical component of business success; experimentation is important in picking a good assortment; your sample needs to be representative of the population you are trying to understand; and you should do your research well.

But I want to extend his thinking about how we determine what is and is not a good assortment. Specifically, I want to talk about the choice context, the sample and the limitations of TURF analyses.

Complete control

All of the examples described at the beginning of the Travers article share a common feature: the choice environment is monopolistic. Whether it is the vending machine, the proprietary computer store or the bank, the “owners” are deciding the assortment because they have complete control over what it is. However, as he ventures into the example of which flavors of juice to put on the grocery shelf, there is no discussion of the retail context. CPG manufacturers understand all too well that whatever assortment they select for themselves is subject to a retailer’s whims and to competitive effects. To use Travers’ example, we might find that our orange juice-apple juice-grape juice configuration is optimal for us. But a retailer who is already heavily stocked in grape juices may reject that product – and may not replace it with our iced tea offering. Just because we offer a three-item line does not mean retailers will take all three items. A competitor who specializes in grape juice may go after our product in order to protect their franchise, diminishing the advantages of our offering.

Travers compellingly argues that our profit maximization should drive our choice of assortment and, on the surface, this seems pretty reasonable – let’s make as much money as we can. However, once we are no longer in control of the shopping environment, the goal of profit maximization loses some of its primacy. 

  • The retailer’s goal may be profit maximization and that may conflict with our goal. 
  • The retailer may want to stock a product because a competitive store stocks the product – there is value in defensively preventing uniqueness. 
  • We might alter our assortment recommendations based on the demographics of a store’s trading area, even though the new mix may not be as profitable for our brand. 
  • We may be recommending an assortment in order to keep or gain physical space. This is especially true if we have a new sub-line of products coming soon. We want more space now and then we’ll give it up for the new product line.
  • We may recommend an assortment that includes me-too products so that shoppers will look at our brand too, not just a competitor, for those variants.

Travers advocates subgroup analyses to make sure that we are not missing something in the data and we heartily agree. What we want to recognize is that we adopt assortments for more reasons than profitability. By adding in the shopping context and retailer goals, the assortment math may need to go by the wayside.

Not always right

A representative sample is not always right for assortment analyses. Or, more accurately, the population you want to study may not always be a representative sample of the population at large or even representative of category users. When you want to change your existing assortment, your primary concern is franchise alienation. You care much more about what your brand users think than what competitive buyers think. This is especially true when: your category shows high levels of brand loyalty; your product’s assortment factors are unique to you; or the assortment factor (flavor, scent, color, etc.) is not what drives purchases.

Let’s be honest – in most categories, all the good stuff has already been done; the most popular flavors are available, the most popular scents are being produced, etc. In most cases, you are looking to make changes in your assortment that are going to have a fairly small impact on revenue and will have a small impact on your buyer base. You want to make sure you are not alienating your franchise when you change your assortment; you’re not trying to bring in competitive buyers with your umpteenth new fragrance.

Of course, a new brand’s assortment analysis has a totally different sample requirement – the need for a representative sample of category users could not be more important. If it is a niche brand, then a sample, or at least a readable subsample, of niche users should be included in the sample. 

Often misunderstood

We stole TURF analyses from media researchers in the mid-1980s, although I suspect they’ve forgiven us by now. It is a remarkably useful tool but one that is often misunderstood and credited with too much analytical power. This comes from weaknesses on both the input and the output sides of TURF research.

We can ask the TURF question in any number of ways, although we usually are asking for respondents to state a purchase likelihood of some sort. This may be a standard purchase intent question (definitely would buy, probably would buy, etc., on a five-point scale) or it may be a sequential question (which variant would you most likely buy, if that/those weren’t available, which would you next most likely buy, etc.). These aren’t great questions from a psychometric viewpoint. Purchase intent suffers from being a relatively limited scale, driven more by the brand than the assortment variable(s), and is not a very good predictor. The sequential approach relies on providing the entire range of variants to the respondent – a choice array that is not likely to occur in the real world. 

On the output side, TURF analyses often have the problem of ties. At some level, you will see very small differences between the choices, yet you have to make the call as to which one you’ll choose in order to continue the analysis. When this happens near the end of the analysis, maybe in the fourth or fifth level of variant choice, it’s not such a big deal. You will usually find that the incremental value of each variant is relatively small. When items are close together and near the top of the hierarchy, however, the results can be problematic. 

We recently ran an analysis of a potential 10-item line where four of the items had first-choice preference between 33 percent and 38 percent. These preferences are all very close and which one you choose first leads to different ending assortments. Travers’ TURF-WAR approach (Quirk’s, June 2017) may or may not provide some relief from this problem depending on differential co-purchasing. In the category we studied, there is little co-purchasing, so his technique does not help us decide.

Exploratory rather than confirmatory

Most important when using TURF is to understand that it is an exploratory technique rather than a confirmatory technique (Needel, 2006). There are no statistics that give TURF a patina of scientific respectability – it is just a counting tool. Is it useful? Of course it is. And can it be done better? Yes, it can, and Travers shows us some ways to make it better and some of the pitfalls to avoid. But in the end, TURF only gives you ideas about what might be a good assortment for your product in isolation from the rest of the category. 

Our advice is to let TURF come up with a variety of answers that meet success criteria (profitability, retailer acceptance, etc.) and then test those ideas. Improving assortments is not easy and researching them is a very important endeavor but there are a number of pitfalls to avoid. Tread carefully, don’t be afraid to experiment and do your experiments with the appropriate validated tools.

References

Needel, Stephen. “Can you reduce product assortment? Of course you can.” In Excellence 2006: ESOMAR World Research Papers, pp. 277-292. ESOMAR. 2006.

Travers, Mark. “Using TURF to find something for everyone.” Quirk’s. June 2017.

Travers, Mark. “Using consumer data to optimize product assortments.” Quirk’s. December 2017.