Benchmarking your B2B customer segmentation
(Disclaimer: Some of these pointers might apply to B2C customers too, but some won’t)
Big corporations have B2B customers. They serve their end customers via an intermediary such as a retailer, and hence they need to win the retailer first, before their competition does. To win them, they need to segment them first.
If you are a big corporation, the way you would engage and incentivize businesses like retailers is likely a three-step process — segmentation, targeting and positioning (STP). Here is how it works:
a) Segmentation —Creating clusters of similar customers based on some agreed dimensions like recentness of purchase, frequency of purchase etc. Let us call these clusters as segments, and they are the protagonist of this article.
b) Targeting — Identify your sweet spot segment. Make sure it is reasonably sized, and juicy enough when squeezed to give you an ROI on your marketing spend.
c) Positioning — This is where you hit the sweet spot and extract the juice, with the right intervention at the right time, through the right channel.
For decades, data analysts have toiled to create the perfect segments. Customers are perceived differently. Perception is subjective. One can argue over how many segments, what dimensions to segment on, and how cohesive the segments should be. So, let us benchmark a good clustering model, if your customer is B2B (for example, a retailer).
Here are thumb rules to conclude your segmentation well. I have classified these thumb rules as What is often done/What works better/What is best. Based on my research, if you are doing the better/best, you are likely ahead of the industry curve in the context of using data and analytics maturity.
Here we go.
- Think big picture in your segmentation.
What is often done: There are companies which go the RFM-RT route (the 5 dimensions being Recentness of purchase, Frequency of purchase, Monetary spend, Range of assortment and Tenure of customer). This is not necessarily wrong, considering they have looked at dimensions outside this bucket. Am I doing what everyone does, or can I do more?
What works better: It’s hard to see things when you are too close. Take a step back and look at the bigger picture. The FPB (Firmographic, Psychographic and Behavioral) route helps teams look at a larger picture for B2B customers. These dimensions outstrip the transactional persona and paint a more holistic picture. Do not hesitate to explore syndicated data either — there are simple data points like “retailer revenues by geography” available quite easily with third parties today.
What is best: Customer Segmentation is an evolved field, and the problem is often not the lack of available data, but the lack of nuance. Merging customer data with syndicated data can change the game. A simple example — while companies consider “sales” generated by a retailer for them in the Firmographic dimension, it would be great to augment it by how much sales they are creating for competitors, along with growth patterns for me versus for others (consider overall sales of retailer as a data point).
2. Merge your dimensions effectively:
What is often done: Using >4 dimensions and running a K-means clustering with an elbow curve is a norm. Using 5 or more different dimensions can complicate clustering, not just because usual and first-hand calculated dimensions like R, F and M often correlate, but also because increasing dimensions blurs what constitutes a segment. Too many cooks spoil the broth.
What is better: Principal Component Analysis (PCA) is a mechanism to reduce dimensionality. The 5 different dimensions can be reduced to 2, called PC1 and PC2 — with PC1 capturing the most variation, and PC2 capturing the next most, and both sharing the data so that they give the full picture when they go together. This reduces the scenario of using multiple dimensions that correlate with each other. The lesser the better.
What is best: Derived Dimensions have worked really well, when they are quantifiable/understood. For example, RFM-RT index can merge 5 dimensions to form 1 score to describe a Loyalty Score of the customer. Similarly, a combination of recency, frequency, seasonality and channel engagement/purchase behaviors can predict Churn Propensity using ML classification. These limited dimensions tell a wider story and are explainable. Add story and meaning to your dimensions!
3. Evaluate homogeneity and heterogeneity:
What is often done: Segments (clusters) are visually demonstrated and the if they look satisfactorily clustered, it is deemed appropriate. Sometimes, the ability to corner a cluster using a straight line or a simple curve is considered satisfactory. I only believe what i see, but am I right?
What is better: A Dunn Index or a DB (Davies-Boulder) Index calculates inter cluster distances and intra cluster distances. The larger the ratio of inter/intra, the better the clustering. Keep your friends as close as possible and everyone else as far as possible.
What is best: Augmenting a Dunn Index with a Silhouette plot. While Dunn Index will validate the cohesiveness and separation as mentioned above, Silhouette will validate the location of key data points (sample ~20) within their clusters. You are proving that a few key sample customers are actually sitting where they truly belong.
4. Optimize drift:
What is often done: Segments are designed in a volatile fashion. We often don’t realize the volatility in the process of segmentation, but it hits us later like a brick. It means the data points (customers) could be drifting often from a “loyalty” to an “about to churn” segment too fast, before you take any action. This creates inconsistent Targeting and Positioning strategy leading to a spray and pray journey for the customer.
What is better: Assuming you use Principal Components or Derived Dimensions, generate them using medium-long horizons over short horizons. This will ensure that a small seasonal lapse in customer engagement or sale will not immediately impact the segment. That way, loyalty is hard earned over time, and loyalty will be championed despite small lapses.
Happy Slice and Dice!
There’s more to the story.
I have seen and heard companies' remorse about how analytics is underused in Positioning stage. Use the segmentation benchmarking above to quickly release time and efforts from your Segmentation to focus on Positioning. That is where you extract the juice, after all. I would be releasing another article in that context.
