Recommending Assortment to a Store— How Much Is Too Much

Introduction

Consumer goods companies typically indulge in B2B business. For them, retail stores are not only their customers but also facilitators who connect them with the end consumers. Hence, companies have the incentive to push more and more types of SKUs to these retailers, so that they can adhere to the requirements of more and more consumers. However, there are upper limits to this. Any additional SKU requires negotiations and discussions with the retailer. They have no reason to believe why the new SKU, which they have never bought, is good for their business. There is always a general sense of reluctance towards disturbing their steady assortment.

There are two types of SKUs that can be recommended — those from brands that the retailer already purchases, and those from brands not yet explored. The ones in the former are more likely to be accepted since they are relatable for the retailer. But the ones in the latter are more profitable for the company since they open completely new sales areas, which can be further expanded in future. Therefore, whenever recommending products, it’s important to reach a balance (which varies by retailer) between the two. This article demonstrates a simple approach to systematically decide this balance for each retailer.

Fig 1 — Types of SKU recommendations

Approach

Let’s assume we have one year’s order-level data for the retailers. The bare minimum fields any order-level data should consist of are Order Date, a unique identifier for the customer (Customer ID), a unique identifier for the product (Product ID), and quantity (of a particular product purchased by a particular customer on a particular date). Furthermore, we have the mapping of Product ID and brand, which can be merged into the orders data.

Fig 2 — Basic data structure

The above image shows a sample of the data structure. The dataset can be much larger, depending upon the type of industry, and the company. Irrespective of the size, the important thing is that the duration is large enough, preferably a few months at least. Now, using such data, we can easily determine the following two metrics for each customer.

1. Sensitivity — how much the retailer/client is open to completely new brands. This can be quantified by calculating the number of additional brands they have started to purchase in the recent past, say 2–3 months. It’s important to note that here we are talking about additional brands and not the total number of brands they regularly purchase at this point in time. The idea is to determine how much the client is expanding, and not what is their current level of expansion.
2. Portfolio depth — how much variety of SKUs does the retailer typically keep per brand. This can be quantified by calculating the average number of SKUs purchased per brand. For example, the portfolio depth of the retailer in figure 1 will be = (2+2+3)/3 = 2.3.

Once we have the above two for each customer, we can plot them and create a matrix as follows.

Fig 3 — segmentation of clients/retailers

The actual values of the two metrics can possibly have any range. Hence we make divisions on percentiles. The cut-offs can be adjusted as per the requirement. We can also increase the number of divisions to create 6 or 9 or even more, to reach a higher level of customization by retailer. For this demonstration, let’s keep things simple.

Now, it is quite intuitive that a client with high sensitivity will be open to additional brands, whereas a client with high portfolio depth will be open to additional SKUs within existing brands. If both are high/low, we can go for a balanced mix. The labels in figure 3 mention the suggested strategies for each bucket. If the number of buckets is higher, all we have to do is vary the proportions of recommendations coming from the two categories in a stepwise manner. The underlying principles will remain the same.

Conclusion

While recommending products to B2B clients, we should always avoid taking things too far. There is a serious risk of losing credibility if all the recommendations are too ambitious, and don’t make sense to the customer. Also, the level of experimentation that can be done varies by retailer. In this article, we went over a simple approach to segment the customers based on their buying behavior, so that different strategies can be implemented for each segment. In this way, we can expect to minimize both risks to credibility as well as untapped expansion potential.