The Workshop Everyone Applauds
I've run something like forty segmentations for large companies over the last decade. About half of them never took root.
The dead ones failed the same way. A team spends a few months on customer research and a segmentation, cutting the market the way the data lets them cut it for free: industry, company size, age, job title, the ICP from last quarter. They name the personas, add stock photos, and present. The room lights up — "the clearest picture of our customer we've ever had." Then the deck becomes the answer. Segmentation was the Q1 deliverable; now the real work starts.
A few weeks later the first product decision lands — which feature to build, which buyer the page is for, which plan to price up — and the segmentation has nothing to say. The team falls back on gut and the loudest customer. The deck stays on the wall, and within a month it's dead. Nobody noticed the date of death.
What changed everything was the order of the cut. The segmentations that took root were the ones where I led with Jobs — cutting first by the work the customer was trying to get done, and getting the team to commit to that instead of demographics. Those stuck, and threw off real business value: sharper roadmaps, pricing, messaging.
But "segment by Jobs" only got me to the edge of the problem. Inside the Job the ground was still soft. By the Big Job above it? The context around it? The outcome the customer wanted? Each choice drew a different map, and I had no principle for picking. For years that was where method ran out and taste took over — the part that earns segmentation its reputation as a beast.
The answer arrived about a year ago. It was so simple, and had been so hard to see, that when it landed I got up and walked around the room. Ten years and forty segmentations to find something I can now say in a single sentence.
Real, Working Segmentations Are Rare
Start from the uncomfortable fact. Real, working segmentations are rare. Far more of them get applause than ever change a decision. So the interesting question isn't "how do I run a segmentation workshop" — anyone can run one. It's "why do so many of them die, and what do the survivors do differently?"
They don't die because the team is lazy or the data is dirty. They die because the cut carries no cause-and-effect a team can act on. A segmentation is worth exactly as much as the decisions it lets you make, and a demographic deck lets you make almost none. Knowing a buyer is a 200-person company, or 35 years old, doesn't connect to a single product move — what to build for them, what to charge, how to win them. Two customers with identical age, income, and title can want completely different things and judge success by completely different standards. Group them by demographics and the line runs straight through the middle of every real segment, mixing people who share a profile and nothing that decides what you do next.
That's the difference between a slice of the market and a segment you can run on. A slice tells you who's out there. A segment tells you what to do — which is exactly what the dead decks never could.
The survivors cut somewhere else, and this is the sentence it took me ten years to write. They segment by the set of Core Jobs people are trying to get done, the outcomes they expect from each, and the success criteria they judge those outcomes by. That is the answer to which part of the Job. The Big Job above supplies the motivation, and the context shapes the details, but the cut that makes the map hold is the expected results themselves — the specific things the person is hiring the product to produce, and the bar they hold each one to. That single move is the difference between a deck on a wall and a segmentation the whole team actually runs on.
What Anastasia Saw in the Data
Anastasia is a product manager on the Kotlin team at JetBrains. A few years ago she was trying to segment the people building cross-platform mobile apps with Kotlin Multiplatform, and she did it the standard way first: company type, team size, the platforms they shipped on. She turned the data every way it would turn. Nothing useful came out. The developers who picked Kotlin over Flutter, React Native, or Xamarin had no demographic thread connecting them. With no segment to point at, the team did the only thing left — a little of everything for everyone. The backlog swelled to hundreds of open issues with no principle for ordering them.
Then she cut by the Job instead, and two clean groups fell out of the same population. One group was hiring Kotlin to reduce the errors in the complex business logic shared across platforms. The pricing math in a ride-hailing app has to come out identical on iOS and Android, and a divergence is a real bug with real money attached. The other group wanted to keep a highly customized, native-feeling interface on each platform. Same surface activity, building a cross-platform app. Two different outcomes, judged by two different standards, so two different segments needing two different things from the same technology.
That sentence is the one to hold onto. The segmentation didn't change a slide in the marketing plan. It changed which bugs got fixed first. It moved the roadmap, the positioning, and the message at the same time. Kotlin Multiplatform was effectively unknown in 2019; within a few years it sat in the top three cross-platform technologies, and satisfaction climbed about 20%. The team stopped building cross-platform UI frameworks it didn't need and started shipping against two Jobs it could finally name.
Why This One Choice Decides Everything Downstream
Anastasia's line is not a flourish. It's the reason a dead segmentation is so expensive, and the reason a live one is worth the trouble.
I made the full case back in Part II, so I'll only point at it here. Everything in the product flows from the Jobs of the segment you choose to serve (II.1). Value, the landing page, pricing, the funnel, unit economics, operations, the roadmap — none of them are independent disciplines. Each one is a projection of which Core Job of which people you decided to compete for. And the error runs one direction: choose the wrong Job of the wrong segment, and every step below it inherits the mistake, no matter how cleanly you execute the steps (II.2). The most expensive single mistake in product is competing for the wrong Job of the wrong segment.
So segmentation isn't one stage among many that changes nothing. It's the orienting choice the whole product sits on top of. That's exactly why the demographic deck that dies is so costly. It looks like a harmless marketing artifact, so its death looks free. It isn't. While it sits unused on the wall, the team keeps choosing segments and Jobs anyway, silently and badly, one decision at a time. Every one of those choices ripples through value, pricing, and the roadmap.
Why the Dead Ones Die So Quietly
Two things make the death almost impossible to see while it's happening.
The first is that doing it right is slow, expensive, and doesn't feel like progress. Real segmentation means interviews, transcripts, arguing about which outcome is actually the Job, and sitting in uncertainty for weeks. None of that ships. Testing a fix for a problem you already understand feels wonderfully productive by comparison — you write the ticket Monday, ship Thursday, watch the number. Under deadline pressure the brain reaches for the work that feels like work. So the team drifts back to shipping against intuition, and the slow, correct work gets abandoned for the fast, satisfying kind.
The second is that the segmentation never dies in a meeting. It dies one reasonable-looking ticket at a time. This feature is for the customer who emailed yesterday. This price tier is because sales asked. This homepage is because the CEO wanted it punchier. Each call is locally sensible, and not one of them announces "I am ignoring our segmentation." The death is spread across dozens of small, defensible decisions, so there's never a single visible failure to point at. By the time anyone notices the deck is irrelevant, no one can say when it stopped mattering. There was no funeral because there was never a body.
Which is also why the applause in the room is worthless as a signal. A boardroom rewards a clean story: tidy boxes, recognizable names, a slide that presents well. The only test that means anything is whether the segmentation survives the first real product decision and actually routes it. Most never get tested at all. They get admired, and then they get ignored.
Wes
Where it leads
Cut by the outcome people want and the standard they judge it by, and the deck finally drives decisions instead of decorating a wall. But some of the cleanest-looking criteria are only labels, and telling a label from a real cause is what the next chapter is about.
Source · anchor
segmentation.md §1 (every product decision contains a segment-and-Job decision) · §2 (Core Jobs with success criteria are the segmentation root; demographics never first) · §17 (segmentation failures start upstream) · ajtbd-key-theses.md §2 (most expensive error: wrong Job of wrong segment), §12 (demographics as a second-order correlate) · cross-refs: II.1, II.2 · case: Kotlin Multiplatform with AJTBD (Anastasia Kapanina, JetBrains) · emotional arc: course-transcriptions.md L663 ("сегментация — очень хитрый зверь… только в июне удалось хакнуть")