Customer Segmentation Playbook

Segment your market by the problem customers are trying to solve and how they buy, not only by company size or industry. Pick the one segment you can win first, win it, then expand.

Many early-stage teams can describe their perfect customer in detail but cannot say which group of companies shares the problem, buys the same way and can be reached through the same channels. That gap is why the same product can win easily in one corner of a market and stall everywhere else.

What is customer segmentation, and why does it come before an ICP?

Customer segmentation divides a market into groups whose needs differ enough that each group calls for a different offer, message or channel. You then choose one segment to win first, your beachhead. An ideal customer profile (ICP) describes the best-fit accounts inside that segment, and personas describe the people inside those accounts.

The order matters. An ICP written before you have chosen a segment usually describes your favorite customer, not a market.

What makes a real segment? The research has not settled it: Thomas Ritter and Carsten Lund Pedersen, reviewing the field in Industrial Marketing Management in 2024, conclude that business-to-business segmentation “is still not fully developed as a general theory”^[1]. The test I use is the needs-based one. Grouping customers by size, region or how profitable they are to you, without tying the group to a shared need, produces tidy lists that do not explain why anyone buys.

So define each segment by a shared need, then describe it with traits you can see, so you can actually find those customers. Using several layers of traits goes back to the “nested approach” Benson Shapiro and Thomas Bonoma set out in Harvard Business Review in 1984: start with what you can see from outside, such as industry, size and location, and work inward to how the company buys, the buying situation and the people involved^[2].

The buying process deserves more attention than it usually gets. In a Gartner survey of 632 B2B buyers, 74% of buying teams showed unhealthy conflict during the decision, and groups that reached consensus were 2.5 times more likely to report a high-quality deal^[3]. Two segments with the same need but different buying groups need different sales motions.

Selling to consumers? Segment by situation before demographics. The same person buys differently on a rushed weekday and a relaxed weekend. A 2025 paper in the Journal of Business Research argues that segments are better described by consumption situations, the people, spaces, devices and routines involved, than by consumer types alone^[4]. The jobs-to-be-done idea makes the same point: in the authors’ words, “the circumstances are more important than customer characteristics”^[5].

Start with a long list and cut it down. Write every group you could sell to, then merge the ones that share the same problem and split the ones that do not. Keep three to six. Fewer, and you have probably skipped a real option; more, and you will not have time to test them.

“Mid-market education” is a label; “tutoring centers that must show parents reading progress every week” is a segment you can test.

Early on your data is too thin to show segments, so you find them by talking to people. Speak to 10 to 15 people across the candidates: customers, lost deals and prospects who took a call. Ask what they were trying to get done, what they used before, what made them look, who else was involved and who paid.

Then work through five moves. Write each person’s problem in their own words, without translating it into your features. Group the problems, not the companies. Check the groups against evidence you already have, such as win-loss notes, support tickets and trial usage. Name each group by its need. Finally, note which of your Step 1 candidates survived, merged or split.

If you already have enough data to cluster customers with software, check the clusters against business results, not only statistical neatness. A 2023 review of 172 studies on algorithmic segmentation found that only 4.1% had subject-matter experts judge whether the segments made sense, and recommended judging segments against business outcomes, such as the company’s own performance indicators^[6].

Size each surviving candidate by counting the buyers you could actually sell to and multiplying by what each would pay in a year. Aulet’s check on a beachhead’s size is the yearly revenue you would earn with 100% of that segment^[7]. A top-down figure from an industry report tells you the market exists; a bottom-up count tells you whether the part you can reach is worth the effort.

Use a public count wherever one exists. In the US, the National Center for Education Statistics counts 13,335 regular school districts in 2023–24^[8]. For businesses such as tutoring centers, the Census Bureau’s County Business Patterns counts establishments by industry code (tutoring falls under NAICS 611691, exam preparation and tutoring), which is a better starting point than a guess^[9].

Score each candidate from 1 to 5 on six criteria. Three describe how attractive the segment is: pain (how costly or urgent the problem is), willingness and ability to pay, and size. Three describe your ability to win it: reachability (can you name the buyers and reach them affordably), competition (including spreadsheets and doing nothing) and fit today (can your current product solve it without custom work).

Weight the criteria before you score, so the answer you want does not shape the weights. Write down the evidence behind every score, and mark guesses as guesses; the Segment Scorecard has a confidence column for this.

The biggest segment, districts, scores highest on attractiveness and lowest on ability to win: long buying cycles, entrenched suppliers and requirements the product cannot yet meet. Plotting the two halves of the score makes that trade-off visible.

Choose one segment to win first, your beachhead, and keep marketing, sales and product focused on it until you are clearly winning there. Geoffrey Moore’s Crossing the Chasm, first published in 1991, builds its advice on this: a small team cannot win a broad market at once, but it can dominate a narrow one and use it as the base for the next^[10]. Aulet’s tiebreaker helps when scores are close: select the segment where you have “the highest odds of success while also giving you legitimacy and strategic assets to grow further”^[7].

In my view, a good beachhead has six things: a painful problem the group already spends money or time on; budget and a buyer who can say yes; reachability through one or two channels; a product you can deliver today; no entrenched rival that owns the relationship; and word of mouth inside the group, so each win makes the next one easier.

What two well-known companies’ filings show

The public filings of two companies show the same pattern: start with one clearly defined group, then expand to the next. These are what the documents say, not an account of how either company made its choices internally.

Amazon started with books. Its 1997 IPO prospectus describes a store that opened online in July 1995 and offered “more than 2.5 million titles”. The reasoning was a need physical stores could not meet: “the extraordinary number of different items precludes even the largest physical bookstore from economically stocking more than a small minority of available titles”, while an online store has “virtually unlimited online shelf space”. The prospectus also called the book industry “large, growing and relatively fragmented”^[11]. Books came first but not alone; the same sentence mentions “a smaller number of CDs, videotapes and audiotapes”. Amazon then launched a full music store in June 1998 and a video store in November 1998, and cumulative customer accounts grew from about 1.5 million to about 6.2 million during 1998^[12].

ServiceNow started with IT departments at large companies. Its annual report says that “In 2004, ServiceNow pioneered the cloud-based delivery of information technology (IT) service management applications”^[13]. Its 2012 IPO prospectus defines the target precisely: “We target our sales efforts at large enterprises, which we define as companies with over $750 million in revenues and a minimum of 200 IT employees”^[14]. By 2015 the same 10-K lists solutions for “IT, human resources (HR), facilities, field service, marketing, customer service, security, legal and finance”.

Both moved to the segment next door, keeping the customers they had won: Amazon added categories for the same shoppers, and ServiceNow added departments with service workflows like IT’s.

Once the beachhead is chosen, describe it in four parts: the ideal customer profile (ICP), the buying group, the message and the channel. The ICP is the list of traits you can see from outside, which is how you find the right accounts. ServiceNow’s definition above, a revenue threshold plus a minimum number of IT staff, is a good model: two observable facts anyone on the team can check.

Write the buying group down person by person, because the consensus Gartner measured has to form among these people. Forrester’s 2026 buyer research, from a survey of nearly 18,000 business buyers, puts the average at 13 internal and 9 external people influencing a purchase^[15]. Smaller buyers involve fewer people, but rarely just one.

Decide in advance what “winning” means, in numbers you set yourself, and review every quarter. Watch five signals: win rate against qualified opportunities; sales cycle length; retention and expansion; referrals and inbound from inside the segment; and whether buyers describe the problem back to you in your words.

If win rates stay low and cycles stay long after a fair trial, check your execution before you blame the segment. In a 2025 survey of 259 B2B managers in four countries, the way firms chose and worked their target segments changed how much segmenting paid off^[16]. A good segment with half-hearted execution looks exactly like a bad segment.

When you reach the target, move to the adjacent segment that shares the most with the first: the same need in a new group, or the same group with a related need.

What changes by stage?

What mistakes should I avoid?

• Segments defined only by size or industry. Name each candidate by its need first (Step 1).
• Too many segments at once. Three half-served segments lose to one fully served one.
• Segments you cannot reach. Score reachability honestly in Step 4; a segment you cannot name or reach affordably is theoretical.
• Choosing by market size instead of odds of winning. Plot both halves of the score, as in Step 4.
• Never revisiting the choice. Put a review date on it (Step 7).