Find Segments That Work
Download the workbook including all prompts here.
Among marketers there is always this suspicion that there is not “one” customer but that there are different “kind of animals” that need to be treated differently. A though that is very powerful. But it turns out its is a hard nut to crack to make it work.
I once work as a marketing director for a packaging company and global headquarters had hired McKinsey to introduced a segmentation. The resulting five core segments made perfect sense. Still it lack everything a good segmentation needs to have, like a proof for distinct needs or a mode to access and service the customers differently.
My team and I later developed a segmentation that worked much better based on the principle I will describe here. Bear with me to see what we found.
What is segmentation really?
When providing products and services its turns out that some customers have other needs (Product), use other sales (Place) and communication (Promotion) channels or have other willingness to pay (Price). Just differing in one P can make a segmentation be worthwhile.
A segmentation tries to find customer groups behind the different product, place, promotion and price needs.
Product: If we find needs clusters we can design products that cater this needs.
Promotion: If we find psychological preferences we can create product variants that are communicated differently.
Place: If we know that those groups have sales channel preferences, we can optimize them to their needs
Price: If the groups have different willingness to pay we can differentiate prices 1. Per product or product variant if they have different product or communication needs or 2. Per sales channel if groups show distinct sales channel preferences.
A good segmentation is like tuning a radio to the right frequency—each customer group picks up the signal (product, price, place, promotion) best when it's broadcast in the way they’re wired to receive it. How many radio stations are out there in your market?
The issue – as consumer we are more complex. Our needs are not just evolving but depend on the context and situations. I prefer a sparkling water at an Italian restaurant for lunch, a Kölsch at a bar with friends and an Aperol Sprits at a beach bar date. Needs, psychical traits and willingness to pay differ by context and situation. This is one reason why it is nearly impossible (except in textbooks) to find truly distinct clusters in data.
We can design product variants and optimize its communication and pricing. That is what marketing is referring as “innovation research”.
We can optimize communication and pricing per sales channel, if we realize channels differ in needs and/or willingness to pay
A true segmentation has a deeper strategic implication. It finds “Ideal Customer Profiles” (ICP), sometimes referred and expressed by a “Persona” which the brand wants to understand and nurture over time, life cycles and bandwidth of applications.
Consumers are like chameleons—changing their colors depending on mood, setting, and surroundings—so while innovation tailors the outfit to each occasion, true segmentation tries to understand the chameleon itself: its nature, its rhythm, and its habitat.
Targeting VS All Category Buyers
In the early 2010s, Gatorade rebranded itself to focus tightly on serious athletes as its Ideal Customer Profile (ICP). The brand launched its “G Series,” repositioning the drink as a science-backed performance product. This shift came with increased marketing spending on elite sponsorships (e.g., pro teams, training centers), complex messaging, and lower overall reach. While it resonated deeply with core athletes, sales began to flatten as casual consumers—those who bought Gatorade for hydration or lifestyle reasons—felt alienated.
Eventually, PepsiCo recalibrated. They retained the athlete narrative but widened the lens, reintroducing broader campaigns to appeal to everyday consumers (e.g., “Sweat It to Get It” with everyday workout scenarios).
Focusing on one or more ICPs seems plausible, and we would expect a higher impact. But it comes with costs too:
Overhead costs – operating a marketing organization that caters more than one ICP has higher costs
Higher marketing costs – going after higher value target groups often requires higher marketing budgets
Lower reach – with ICPs you address a smaller market
Whether focusing on ICPs should be recommended is a matter of trading off higher impact with higher costs. Most companies fail to evaluate this tradeoff.
The world’s largest marketing science institute, Ehrenberg-Bass is denying the use of targeting. They argue the default should be to go for reach, not segments. The most cited fact about this is that Coca Cola gets 50% of its revenue from consumers who buy 3 bottles or less per year.
This tradeoff needs to be made. It’s a task outside marketing insights. The concept of “segments of one” already appeared in marketing in the 1990. Even the digital economic was not able to create it and it will even take AI a while to figure it out.
Focusing on target groups is like cooking a gourmet meal for a few—it’s impactful but can be expensive—while mass marketing is serving a buffet. You need to do the math and find a way to cook gourmet taste at acceptable costs.
Finding Segments
Step 1 - Gathering Wisdom Qualitatively
In 2009, Tropicana launched a $35M rebrand based on quantitative segmentation, aiming to appeal to a more premium audience. But by replacing its iconic orange-with-straw image, the brand overlooked the emotional meaning tied to the original design and its impact on the traditionally served segments. Lacking qualitative insight, the change confused loyal customers and led to a 20% drop in sales within two months—forcing a costly reversal.
The task is to identify the customer groups that your brand should be focusing on. This is a highly complex question which should not be answered directly with quantitative research. It requires an in-depth qualitative understanding of the customer first. Ideally you have been that customer yourself and experienced what he experiences. First-level experience has more depth than reported experience.
These are the research steps you can perform from easy to advanced:
AI’s Expert Judgement: Use the prompt you find in the workbook.
Expert Judgement: First interviews with 10-20 category experts to get a great overview and the breadth of hypothesis. Use the interview guide in the workbook as a start.
Deep Implicit: The method is using specialized audio meditation to quite the (post)-rationalizing mind of industry experts and enable them to hear their inner wisdom and intuition. It takes them 10 minutes and is an experience to be joyful experience. Ask SUPRA for tool access.
Deep Psychology Interviews at least 10 from each hypothesized customer segment. This requires a human moderator.
The outcome of the qualitative phase is a holistic understanding of customers. But it can be noised with post-rationalization and covering storytelling.
Qualitative research provides a storyboard for the final movie production of quant research.
Building and describing solid and distinct target groups requires more rigor. This rigor is tested on criteria that define a powerful segment.
Step 2 - Find distinct needs groups quantitatively
The more groups differ in the following criteria plus the more criteria show differences, the greater the need to serve those groups separately:
Product needs: why do those customers buy, what’s important to them?
Communicative needs: some personalities prefer balanced communication and other thrills. Do proposed groups truly differ in those psychological needs?
Media and Purchase channels used: Some groups consume linear TV others only stream. Is there a large enough difference? How do we effectively communicate and sell to different groups if channels largely overlap (which is the standard)?
Willingness to pay: Pricing is a great way to differentiate a market. Its also a great tool to recuperate the costs of segmentation. It requires the design of segment specific product/service offerings.
A close look at the first two criteria reveals that for segmentation we need to understand the NEEDS of customers. When asking customers “what they need” we face all kinds of biases. A causal analytics approach does not ask for what drives outcomes, it derives it from data. In other words, don’t ask “what’s important to you” but ask “how do you assess brand X” and “do you consider buying it”. Causal AI then finds the perception items that best predict purchase and willingness to buy.
This is the structure of a questionnaire that serves the purpose best:
Screen for category buyers, Measure which brand they know –
Personal context (demographics, values), situational context
Measure for two brands the following:
Brand preference and consideration. Optionally: willingness to pay - measurement for a key product
Perception on functional brand and/or product features (capturing product needs)
Perception on brand archetypes (capturing psychological needs)
Perception of brand at touchpoints (capturing communication and purchase channels)
Now the product needs are those features that explain preference, consideration and willingness to pay. Psychological needs are those brand architypes that best foster the same thing.
Driver analysis is designed to find those items that impact outcomes most. The modern “driver analysis” known today is called Causal AI. You can use the base version of SUPRA Causal AI for free at www.supra.tools/causalai
But at first the analysis runs on the whole market and there are two distinct ways to find and compare your potential segments.
“Apriori” Segmentation
At SUPRA we worked for an Automotive brand and developed a system to not only predict each customers lifetime value but also find meaningful segments. We included available properties of customers into the dataset such as which car they own, if it is leased or own and much more. Causal AI found it highly predictive for the lifetime value if a customer had a high horsepower car or if it was a leased company car. We found five different properties who on its own not only indicated high customer value but also had been associated with a very different profile of customer needs. This can be revealed by segment specific driver models. It made it possible to find and address each segment on the individual level out of the existing database.
Based on qualitative expert knowledge we often already have a hunch about what could differentiate segments. Integrating this information into the Causal AI driver modeling will give us the information whether the criterion really explains why and how customers buy.
Apriori segmentation is like setting the stage with characters you suspect are important—Causal AI reveals who’s really driving the plot, uncovering the hidden motives and dynamics behind each role.
“Aposteriori” Segmentation
Coming back to the packaging brand mentioned in the beginning. We interviewed the customer and prospect base an let them evaluate vendors on the market on different criteria, let them assess their purchasing strategy and asked who they choose and consider. Running this thru Causal AI found a handful of topics to be important.
Because Supra Causal AI models relationships in a universal nonlinear multivariate way, the model can actually report for every case, every respondent, a different “importance” for each product feature or emotional trait.
This individually derived importance figures then can be used to run a clustering exercise like k-means. The process then runs like typical textbook clustering exercises.
The hope is to find truly distinct cluster that are not just cutting one cluster into slices like in this picture.
Hopefully distinct clusters then need to be assessed. Assessed whether you find channels that distinctly communicate and sell to only this segment. The more it overlaps, the weaker the segment strategy.
Aposteriori segmentation is like listening to hundreds of people describe their perfect song—Causal AI picks up the subtle rhythms each person responds to, and then DJs use that data to form real music tribes, not just by genre, but by the beat that truly moves them. Its like watching how guests move through a buffet—Causal AI observes who picks what and why, then groups them not by assumed preferences, but by the hidden patterns in their choices and tastes.
The Truth About Segmentation
Segmentation only works when it’s grounded in real needs, not just textbook logic.
Too often, companies slice the market neatly—only to find the segments don’t differ much in what really matters: product needs, communication styles, channels, and price sensitivity. Or they focus on shiny customer profiles which sabotage business with the general category buyer
I learned this firsthand. True segmentation blends human insight with causal analytics to uncover who drives value and why. Whether guided by experience or intuition (apriori) or discovered through data (aposteriori), the goal is clear: find segments that don’t just look different but behave differently—so you can serve them better, smarter, and profitably.
This is how you 10x insights.
Download the workbook including all prompts here.
Book a FREE strategy call at https://supra.tools/