You Don't Need a Segmentation Model - You Need (at least) Four

This article is going to start with a story, and you'll probably find the ending uninspiring (how's that for a hook?). But bear with me; I'm writing this to help you avoid some of the letdown I felt at the time.

Back in my full-time employee days a new CEO took over. His tenure followed years of what I would broadly call 'optimization-focused' management. The company had built its business on relentlessly analyzing marketing programs, products, pricing, logistics, and just about anything else you can imagine. The new CEOs' vision was to make the company more customer-centric, and he invested heavily to bring about that change. In addition to hiring a new CMO and other executives, he approved a multi-million dollar series of initiatives with the ultimate purpose of transforming not just how the company did business - but how it thought about doing business.

One of the initiatives involved engaging an outside consulting firm to build out a new customer segmentation structure. The project was extensive, using both data analysis and primary market research. When all was said and done each of the company's mainline brands had a new segmentation scheme, including customer profiles, key purchase decision drivers, and journey maps. The team also incorporated statistical models that, using a third-party demographic & behavioral data append, could predict a segment for each individual who had purchased from us. We scored the entire file of existing customers, and every time a new buyer came in we'd score them as well.

Based on the segments with the highest potential value, the company identified one or two primary target groups per brand and began to re-orient marketing accordingly. All employees were instructed to include this lens when reporting results - acquisition marketers would report on whether campaigns were delivering the right segments of customers, and retention marketers would detail how they would adjust messaging and other features of the offer when communicating with the different segments. In a relatively short span of time, the customer-centric thinking was permeating the entire organization.

Since my role at the time was primarily focused on optimizing the performance of existing customers, particularly through the email and direct mail channels, I was excited about the opportunities offered by the new data. In theory, we could tailor the marketing messages to each customer based on his/her segment, improving overall performance once we had tested our way into the best possible versions for each.

The results, at least in the Retention space, were less than impressive. We ran many tests of different product, offer, and messaging combinations over a period of many months, and most of the time it was difficult to discern a significant positive impact. That isn't to say the project was a waste; there did appear to be some benefit on the customer acquisition side and the new approach sharpened and aligned the thinking in the marketing organization overall, but those of us tasked with improving the performance of the existing customer base were struggling to see major gains.

Why was this? The pieces appeared to be there:

• The segmentation identified groups of people with different values and behavior patterns, relative to our product lines
• These behaviors and values suggested that the segments would respond differently to value propositions (e.g., strong preference for "speed & ease" vs. "quality and selection" when choosing where to buy)
• The statistical model that assigned every customer to a segment was very accurate when compared to segments assigned using a thorough survey analysis

Yet, the segmentation scheme just didn't help us very much when we were communicating with our existing customers. It took me a while (years, even) to fully understand and internalize the issues, even though I probably should have seen them far earlier. Ultimately, the segments we rolled out were not especially useful for our retention efforts for the same reason a Toyota Camry isn't an ideal choice for moving furniture - the tool employed wasn't designed for the job at hand.

I'll circle back at the end with a bit of math to illustrate one specific - but very common - problem, but first I'll address the primary issue. In short, you don't need one segmentation scheme for your business - you need (at least) four.

One Segmentation, Two Segmentation, Three Segmentation, Four…

Anyone who works with customer data is going to face questions about how to segment either the customer file, the universe of potential buyers, or both. I'd say at least 80% of my clients since I've become an independent consultant have involved a segmentation component in one way or another. The specifics vary quite a bit, but the goals broadly fall into two primary categories:

1. Improving Acquisition - Who are our best customers, and how can we (efficiently) find more of them?
2. Improving Retention - How can we improve the performance of the customers we already have?

Customer segmentation frameworks are key to informing these efforts, but businesses must keep in mind that there is no one 'silver bullet' that can cover all needs. This article describes four different types of segmentation. Each provides a specific lens through which to view the business, serves a distinct purpose, and is defined using different types of data. These segmentation types are primarily defined by the ways the business will use them; the means with which the segmentations are created is secondary to this overall goal.

In the table above, The first two categories (Strategic & Acquisition) apply to everyone - customers and non-customers alike. The last two (Retention and Planning) apply only to existing customers, both current and former. The "Primary Data Source" and "Primary Data Type" columns indicate how the segment is derived. "Specificity & Visibility" refers to how accurately we can assign an individual customer (person or business) to a segment, and "Stability" refers to how frequently a customer's segment is likely to change. The above are general guidelines - specifics will vary from business to business - but the key thing to remember is that there are continuums and tradeoffs. As you move from abstract potential customers to actual buyers, you are able to employ more and more direct observation of their behavior (at least with you) rather than being forced to rely primarily on inference.

While the table outlines four types of segments, you may actually have more than one segmentation scheme in each category. For example, if your company has multiple distinct business units, each of these may have independent models for Strategic and Acquisition segments. You may also have multiple overlaying Retention segmentations, each focusing on different types of customer interactions.

The final thing to remember is that no segmentation scheme should be viewed as permanent - the market changes, and your business must change with it. The effort invested in continuously evaluating and maintaining the segment structures will help ensure continued utility and relevance over time.

Strategic Segmentation

The strategic segmentation, as the name implies, ties to the firm's overall strategy by defining the broad part(s) of the market you are trying to serve. These are the people or businesses you are going after, and you'll build and/or align your operations to them. This will touch just about everything you do - from product offerings to marketing campaigns to customer service policies. For the strategic segmentation, it's critical to focus on what the customers want and how they buy - rather then who they are. For example:

• Are they more influenced by price or convenience?
• Are any product features or attributes particularly important?
• How much does brand matter? 
• How do they make purchase decisions?
• How frequently do they purchase?

Gathering this kind of information requires direct market research, which for most companies requires enlisting the help of a firm that specializes in that kind of work. These can be major endeavors, depending on the market and size of the company involved.

While the Strategic Segmentation is very important, it has significant limitations. Most notably, it's built upon factors which in many cases are unknowable at the individual customer level, especially at scale. Furthermore, in many cases the Strategic Segment is situational - individual customers may represent multiple segments, depending on the circumstances.

For a simple illustrative example, consider my gasoline buying habits. Most of the time, I fill up at my local Costco. The prices are excellent, but because of the distance to the store and the time spent in line, it's usually more time consuming than other options. I typically plan ahead and append the gasoline purchase to other errands in the area to cut down on wasted travel, but even so on balance I'm willing to sacrifice convenience for a significantly better price.

There is an exception to this, however. When I rent a vehicle, either locally or when traveling, I'll frequently find myself ready to return the car but only needing to put a few gallons into it. Unsurprisingly, this makes my preferences flip - I will almost always seek out the most convenient option, since the small size of the purchase negates all but the most extreme price differentials. What's key to understand is that while I may be "Price" focused 80% of the time and "Convenience" focused 20% of the time, I'm not looking for a single supplier that's somewhere near the 80/20 point on the Price/Convenience continuum - rather, depending on the situation, I'm fully in one camp or the other.

Because of the ambiguity and inherent 'fuzziness', strategic segmentation models are better at describing the behaviors of large (and somewhat abstract) groups of people than of specific individuals. More to the point - the real purpose of the strategic segmentation is not to precisely slot each customer into a segment. The purpose is to define the business's area of focus relative to the market - relative to both customers and competitors. If done properly, the exercise will identify areas where the company believes that it can establish and/or maintain some kind of competitive advantage.

That said, while the output of a strategic segmentation exercise are very helpful in setting up a product roadmap and defining an overall approach to marketing, they don't directly translate into something to find new customers. That's where the Acquisition segment comes in, and the bridge that helps us get there is the Persona.

Acquisition Segmentation

If you've ever been involved in, or even a witness to, a major segmentation initiative you've almost certainly encountered the concept of Personas. These profiles, usually including stock photos of bland but attractive people in various tangentially-relevant settings, might describe "Brad" as married man in his mid-40s with two children in elementary school, living in the suburbs, who gets most of his news online. "Brad", in this case, may serve as the class representative of a Strategic segment of customers who have very stable buying habits (for whatever it is we are trying to sell) and is generally far more focused on the availability and ease of purchasing existing products than the frequent rollout of new offerings. 

Keeping this Persona in mind can help marketers focus their efforts to create a communication strategy that is consistent with the strategic objectives. The characteristics of this theoretical person (or business) serve as an example of the variables that define the Acquisition segment - translating attitudes, preferences, and habits into more directly targetable and measurable demographics and observable buying behavior. The overlap will never be complete, and one Strategic segment may map to multiple Acquisition segments, but the two schemes should always be complementary to each other. 

Most of the segmentation projects that I've seen in my career (including the project I mentioned at the beginning of this article) have been a combination of these first two types. They tend to work very well in polished executive-level presentations, and good consultants can blend the Strategic (market needs and positioning) and Acquisition (demographics and other targeting/messaging) elements into a cohesive narrative that can help focus an organization's thinking in a productive, actionable way. 

Retention (or Tactical) Segmentation

Once a customer engages with the company, either through a purchase or some other action (sign up, free trial, etc.), the information available expands dramatically. This may still include demographic or attitudinal indicators, but now we'll have access to much more data. The Retention segmentation governs how the company interacts with the customer at any given time, particularly with marketing but also potentially with fulfilment and customer service. Content can be customized to reflect each customer's relationship with the company, and discounts and other offers can be targeted to where they have the most incremental impact.

At a minimum, companies should be able to break customers down into five basic categories, based on the customer's purchase history:

• Newly Acquired
• Recently Reactivated
• High Value
• Other Active
• Lapsed / Former

The exact thresholds for these will vary from business to business, with the average time between purchases being especially important.

In addition to the basic purchase history, businesses may also use other events to trigger segments, such as:

• Signing up for an account
• Purchasing a particular product or service
• Interacting with an email
• Visiting the site
• Contacting customer service
• Requesting a refund / return

The possibilities are nearly endless; if data allows, the company may be able to tag apply multiple segmentation frameworks simultaneously, identifying (for example) High Value Customers who recently returned a product.

Ideally, any changes in these segments would propagate through all systems in real-time, so that customers logging into the website or contacting customer service will be treated appropriately. This speed is important, since some of the segmentation variables can change very quickly.

Reporting (or Planning) Segmentation

The final segmentation scheme is somewhat similar to the Retention segment, with one major difference. A Reporting segment, while driven primarily by customer purchase activity, will not change during the planning period - be that period a month, quarter, or year - regardless of what happens during the period. For that reason, Reporting segments won't usually include short-lived indicators like site visits or product returns. When they include elements like "Recently Acquired", the term "Recent" will likely be interpreted more broadly than it would be otherwise.

Reporting segments are important for two primary reasons. First, they enable consistent reporting of retention performance parameters (see Intro to Strategic Customer Performance Measurement ), which in turn allows for effective comparisons over time and, potentially, across business units. Second, these segments are the foundation for Customer-Centric budgeting because they allow the isolation of the various performance parameters in such a way that goals can be set and any drivers of variance clearly identified (see So You Want to Build a Budget - Part 1 and So You Want to Build a Budget - Part 2).

The Right Tool for the Job

Returning to my original example, the output of the segmentation project clearly fell into the Strategic and Acquisition categories. In that context, it was an immensely valuable tool for formally defining (and refining) the company's value propositions and identifying the best customer populations to target. It brought a level of cohesion to marketing efforts that simply wasn’t there before. More than anything else, it put the customer first in everyone's thinking.

But the results on the retention side were lackluster, and the reason wasn't that the segments were faulty. The problem was attributing the abstract Strategic segments to individual customers was not a perfect process - and barring some form of clairvoyance, it never could be. The practical implication is that some fraction of the population would be assigned to the "wrong" segment, and the question at hand is whether the attempt at targeted segmentation beats the "use what's best for most" approach. The answer will depend on how accurate the model is at assigning individuals to their 'true' segments, the relative performance differentials of the different treatments, and the overall prevalence of each segment in the overall population.

Here's a highly simplified example that shows what can happen. The numbers are fabricated but generally reflective of the situation we faced at the time. 

• Total population of 100,000 people
• This population is composed of two broad customer groups - a group that cares about Speed (how easy/fast it is to place an order) and a group that cares about Selection     (how many different products are available)
• We have a statistical model that correctly identifies which segment customers are in 80% of the time
• For most of the company's history, the Speed customers have been the primary target. True Speed customers make up 80% of the file.

An omniscient marketer could build a table that looks like the following:

In reality, however, we don't know the "True" segment. Instead, what we observe is in the shaded section: 68,000 people whom the model classifies as "Speed" focused, and 32,000 classified as "Selection" focused. The key point, of course, is that half of the 32,000 is misclassified.

This line of thinking may feel familiar, since discussion of this particular phenomenon has come up quite a bit over the past year or so when talking about how to interpret positive COVID-19 test results. It's also taught in just about any entry-level statistics course. We don't need to get into a deep discussion of Bayesian inference here, but we need to keep in mind that the factor we care about (the customer's own individual habits and preferences) is not the same as the factor we can observe (the Segment, as assigned by the model). 

Continuing with the example, we'll further assume that we have two major marketing choices available - one targeted at "Speed" customers, and the other targeted at "Selection" customers. Customers receiving the correct treatment (e.g., Speed customers getting the Speed version) generate $10 in revenue. Customers receiving the wrong treatment (e.g., Selection customers receiving the Speed version) generate $7 each. Prior to building the model, we tested our way into giving all customers the Speed version as standard practice:

As with the previous table, all we can directly observe are the shaded sections, specifically:

• Total Revenue / Customer is $9.40
• Total Revenue / Customer for modeled "Selection" customers is $8.50
• Total Revenue / Customer for modeled "Speed" customers is $9.82 (+15.6% vs.     "Selection")

There's a clear performance differential between the two segments, but possibly not as much as we would expect.

Now, what happens if we change our strategy so that we assign customers based on their modeled segment? 

Perhaps surprisingly, not that much changes. Most importantly, the total generated from the portfolio ($9.40 per customer) is the same as before. In this particular case, the reason is that the two rows in the table above where the treatment changed (the red/green highlights) exactly offset each other. This obviously won't always be the case in practice, but it should serve to illustrate how modeling errors can blunt the benefit potential of using segmentation to deliver targeted messaging.

The graph below shows how the yield changes, both for the modeled "Selection" customers and the total pool, with the model accuracy when the company shifts to a targeted approach. I didn’t include a line for modeled "Speed" customers, since they will always get the same treatment under both the "standard" and "targeted" approaches (i.e., such a line would be horizontal at +0%).

The upshot is that increasing the accuracy of the model makes a huge difference, but limiting ourselves to the same structure and data we use for the population as a whole won't get us there. This is where the multiple segmentation schemes come in, and where we fell short when trying to adapt to the new segmentation methodology. We (the analytics team, and the marketing organization more generally) were hyper-focused on the modeled Strategic/Acquisition segments (those assigned prior to incorporating any internal data) that we didn't even make a serious attempt to build new retention segments. If we had, I believe we would have been able to refine the segments in such a way that would have tightened up our targeting significantly - for example, by incorporating product/occasion purchase data. These segments would likely not line up with the Strategic or Acquisition segments on a 1:1 basis, but that's OK - they would be tools better suited for the job at hand.

Conclusion - The Right Tool

Each of the four segmentation types provides a specialized window through which you can view business performance. While it's certainly tempting to try to create one universal solution, business leaders that recognize the different strengths and weaknesses will avoid the frustration that can come from misapplying the models. Instead, focus on what you want each segmentation scheme to do and build separate (but complimentary) models that allow you to:

• Define the scope of your business, relative to customers and competitors
• Identify and attract profitable customers
• Optimize the performance of existing customers
• Set and monitor performance targets

Taken collectively, the suite of segmentation types will provide the tools necessary to truly understand, and optimize, customer performance. Build them - and keep them updated - and your business will be better for it.

-------

Copyright 2021 Krepps Customer Insights, Inc. All Rights Reserved.

Originally posted 5/25/2021 on www.defininggrowth.com