Marketing — the Atlas · ch.04 · segmentation & jobs
📣 Chapter 4 · Part I · The battle for the mind

Whom are you for?

Kotler said slice the market, pick your slices, plant your flag. Christensen said forget who people are — watch what they hire. Sharp said your buyers look suspiciously like everyone else's. This chapter makes all three fit in one head.

Here's the whole chapter in one line: people don't buy a drill bit — they hire a hole, and the more confidently you can describe who your buyer is, the more you should worry that you can't describe when and why they buy. Everything below is that idea — the textbook version, the milkshake version, and the version the buying data insists on.

1The org chart of marketing's oldest idea

The oldest play in the textbook is three moves and one acronym: STP. Segment — slice the market into groups that want different things. Target — pick the slices you can actually win, and put your money there. Position — plant a flag in each chosen slice's mind (that flag is Chapter 1). Wendell Smith proposed segmentation in 1956; Philip Kotler's Marketing Management canonized the full pipeline, and for half a century every marketing course on Earth has opened with it.

Why did it conquer every textbook? Because it turns a mob into a plan. "The market" is millions of strangers doing unknowable things; that's not something a company can act on. Run it through STP and the mob becomes a spreadsheet — rows you can name, size, price, and assign to a brand manager. Segments get budget lines. Budget lines get owners. The whole marketing org chart is downstream of the slicing.

Programmer's version: STP is sharding the monolith. You can't serve every user from one undifferentiated blob, so you pick a shard key, partition the users, and route resources per partition. Sensible architecture — with the same failure mode. Everything depends on the shard key. Pick a good one and the system hums; pick a bad one and every downstream query does perfect work on the wrong partition.

💡
The pipeline inherits its first stage. Targeting and positioning can be executed flawlessly and still fail, because they faithfully aim at whatever the segmentation step said was real. The next two sections stress-test the default shard key — and it does not go well for the default.

2Demographics lie

Try this segment on: male, British, born 1948, married twice, wealthy, famous, lives in a castle, fond of dogs. A tidy row. Media plan practically writes itself. One of the men in that row is King Charles III. The other was Ozzy Osbourne, the Prince of Darkness. Identical cells in the database — and you would not, under any circumstances, run the same campaign at both of them. That gag is decades old because the problem is decades old.

What do demographics actually predict? Category membership at the crudest edges: households with babies buy diapers, people with driver's licenses buy fuel, retirees buy more cruises than teenagers. Useful, real, and almost the whole list. What marketers pretend demographics predict is everything else — which brand, which moment, which message, what the purchase means. But "woman, 25–54, suburban" is not a psyche. It's a census cell wearing a trench coat.

Programmer's version: demographics are user-agent sniffing. The header is easy to read, so we built an industry on it — but it tells you what browser showed up, not what request it's making. The same person is five different buyers in one day: a commuter at 7 am, a snacker at 3 pm, a parent on Saturday. Traits are constant; behavior isn't. Situations beat traits. Watch it happen below — one customer, one product category, and a demographic profile that never moves while everything that matters swings wildly.

Interactive · same demo, different jobs Pick a moment · the customer never changes
Sam Kowalski · the customer (this card is frozen)
age34 income$95k cityChicago segment"urban professional, 25–44" categorymilkshakes & cold drinks
Pick a moment above. Sam's row in the CRM will not move; the job, the rivals, and the winning message will.
"Women 25–54, daytime, three networks."
US television upfronts · the 1960s
the targetAn age–sex cell. Audiences sold like acreage — so many thousand women per spot.
the dataNielsen diaries and a phone survey. The cell was the finest grain the pipes could carry.
the logicSoap ads between soap operas: a crude proxy, honestly priced.
the blind spotWhy any of them buys — the job — never entered the buy at all.
"Bid on the moment, not the person."
post-cookie adtech · 2026
the targetContext plus first-party signals: the page, the query, the hour, the logged-in relationship.
the dataYour own CRM, clean-room matches, modeled lookalikes — the third-party trail went cold.
the logicThe person is opaque again; the moment is legible. Ads match situations, not census cells.
the blind spotSame trap, new pipes: precision about who still can't see when and why.

3Jobs to be done

In the early 2000s, McDonald's wanted to sell more milkshakes. It did the textbook thing: profiled the milkshake segment, asked them how to improve the product — thicker? cheaper? chunkier? — shipped the improvements, and watched sales do nothing. Then Clayton Christensen's collaborators, Bob Moesta among them, tried something unfashionable: they stood in a restaurant for eighteen hours and wrote down what actually happened.

The log was strange. Nearly half the milkshakes sold before 8:30 in the morning — to people who were alone, bought nothing else, and drove off with it. Interviewed, the morning buyers converged on a story no survey had asked about: they had a long, boring commute, one free hand, and a stomach that would start complaining around 10. They were hiring the milkshake to make the drive tolerable and hold hunger off till mid-morning. A thick shake takes twenty minutes to pull through a straw — a feature, it turns out. And the competition wasn't Burger King's shake. It was bananas (gone in a minute), bagels (two hands, crumbs on the suit), doughnuts (gone fast, plus 8 am guilt) — and boredom itself, the incumbent that was winning most mornings.

The afternoon told a different story with the same product. Parents, worn down by a week of saying no, hired a shake as a small, affordable yes — a way to feel like a good parent for ten minutes. Their competitor was guilt (and the toy aisle). And the spec inverts: the afternoon shake should be small and fast, because a kid nursing a spoon-thick 22-ounce cup while a parent checks the time is a product failing its job. One SKU, two jobs, opposite requirements — which is why tuning the product for "the milkshake segment" moved nothing. There is no milkshake segment. There are two jobs that both happen to hire milkshakes.

This is the deep point, and it rewires Chapter 1: the job, not the customer, defines the competition. Remember positioning against the real alternative? The real alternatives are exactly the job's other candidates — whatever else could plausibly be hired for the same progress, including "do nothing." Programmer's version: the customer is the caller, but the job is the call site. You don't optimize a function by studying the biography of whoever invoked it; you profile the call site — what state was the program in, what was it trying to accomplish, what else could it have called.

Interactive · milkshake job explorer Drag the dots into the shaded job zone
commute zone treat zone
Drag the three dots into the shaded bands to build the shake this job would hire.
"Half the shakes were gone before 8:30 am."
Christensen, Moesta & crew · parking-lot interviews at dawn
the methodStand in the restaurant for 18 hours. Write down everything. Then ask buyers in the parking lot.
the questionNot "how's the shake?" but "what were you doing at 6:30 that brought you here?"
the findOne product, two jobs — the commute shake and the say-yes-to-your-kid shake.
the outputA spec per job: thicker and longer-lasting for mornings; smaller and faster for kids.
"What did you fire to hire us?"
continuous discovery · product & growth teams · 2026
the methodSwitch interviews with recent sign-ups and recent churns, on a permanent loop.
the questionThe four forces: what pushed you, what pulled you, what made you anxious, what held you back.
the findJobs become the segments — onboarding flows and ad angles per job, not per persona.
the outputThe milkshake study, running as infrastructure.

4Personas vs jobs

Somewhere on your company's wiki lives "Marketing Mary." She's 38, drinks oat-milk lattes, does yoga on Tuesdays, "values authenticity," and is played by a stock photo. Teams gather around Mary and ask what she'd want. Mary always answers, because Mary is a mock object — a hand-written stub that returns plausible values for whatever you call on it. Your tests pass because you wrote the stub to pass them. No real user was consulted in the production of that green checkmark.

A job spec is a different artifact. It has a shape: when [situation], I want to [make some progress], so I can [outcome]. When I'm facing a 40-minute solo drive, I want something to do that also counts as breakfast, so I can arrive fed and sane. No age. No income. No latte. And unlike Mary, a job spec is falsifiable — you can stand in a parking lot at dawn and check it against reality, one interview at a time.

To be fair to personas: they earn their keep as casting decisions. Tone of voice, vocabulary, media taste, which jokes land, which channels to shoot for — knowing your buyers skew a certain way genuinely helps you write and place the creative. The trouble starts when a persona is asked to do a job spec's work: predicting needs. That's when teams ship features for a fictional woman's imagined preferences, with total confidence, because the mock kept returning what they hoped to hear.

⚠️
Persona theater has a tell. If your persona document could not be wrong — if no interview, log, or purchase record could ever contradict "Mary values authenticity" — it isn't research, it's fan fiction with a budget. Keep the persona for casting and tone. Route every claim about needs, triggers, and competition through a job spec, where reality gets a vote.

5Evidence check — Sharp's cold shower

Now the cold shower. Chapter 1 introduced Byron Sharp and the Ehrenberg-Bass Institute as marketing's auditors; here they arrive with decades of buying panels and a bucket of ice for everything above that smells like "find your special tribe."

Finding one: rival brands' buyer bases are nearly identical. The duplication-of-purchase data is blunt — Coke's buyers also buy Pepsi, and every brand shares its buyers with every rival roughly in proportion to that rival's size. Profile the buyers of competing brands on demographics or attitudes and the columns come out almost interchangeable. There is no Pepsi psyche, no Colgate tribe. Brands in a category sell to the same people at different rates.

Finding two: growth comes from recruiting light and occasional buyers across the whole category — the people who think about you almost never — not from squeezing the heavy, loyal core. Which makes tight targeting arithmetically awkward: narrowing the audience mostly shrinks reach, and shrinking reach cuts you off from exactly the light buyers who would have been next year's growth. Meanwhile precision costs more per impression. You end up paying premium CPMs to re-greet people who already buy you — better CPMs on a smaller pond.

Run the arithmetic yourself. The dot field below is a category: a few heavy buyers, a sea of light ones, a fog of not-yets. Slide the budget between broad reach and precision targeting and watch where the money lands — then flip the B2B switch and watch the verdict invert.

📈
Evidence check. The duplication-of-purchase law holds with boring regularity across categories, countries, and decades of panel data: a brand shares its customers with every competitor in proportion to competitor size. And penetration — how many category buyers buy you at all — tracks growth far more tightly than loyalty metrics do. Both regularities are the empirical spine of How Brands Grow; Chapter 5 gives them a full hearing.
Interactive · reach vs precision Slide the budget · then flip the market
reach precision 70% reach · 30% precision
Move the slider. Dark dots are heavy buyers, pale dots are the light buyers growth comes from, ghosts aren't buying the category yet.

6When precision pays

Sharp's arithmetic is not a universal law of nature — it's a law of big ponds. It assumes a mass category where your next thousand buyers are scattered through the population and no list could name them. Break that assumption and the verdict flips. Three honest exceptions:

  • B2B account-based marketing. If you sell reactor-monitoring software, your total addressable market is 500 logos, and someone in your CRM can list them. "Reach the whole category" and "target a named list" are the same act. ABM — ads, content, and sales plays aimed at specific accounts — isn't a violation of the reach math; it's what reach means when the market fits in a spreadsheet.
  • Performance retargeting near purchase. The person with three items in a cart is hours, not years, from a decision. Precision here isn't guessing who might someday buy — it's finishing a transaction that already started. Small universe, short clock, measurable close. The math clears.
  • Genuinely narrow category membership. Insulin pumps, left-handed guitars, software for dental labs: when the category itself is small and definable, "broad reach within the category" is already precise. The sin was never precision — it was precision about the wrong thing, carving a mass category into imaginary tribes.

And when you do aim, aim at the right coordinate. The useful target is rarely a demographic; it's a category entry point — the situation that summons the category into someone's head. "People searching 'moving boxes'" beats "adults 25–40" every time, because the first is a job announcing itself and the second is a census cell. Target the moment the job occurs, and let the demographics fall where they may.

💡
The rule of thumb. Before paying for precision, ask two questions: can I actually name the universe? (a list of accounts, a cart, a search query — not "millennials") and is the buyer near a decision? Two yeses and precision is just efficiency. Two noes and you're buying expensive frequency against people who already know you, while your future buyers scroll past unbothered.

7Targeting after the cookie

For about fifteen years, the third-party cookie let advertisers pretend the Prince Charles problem was solved: never mind the census cell, we'll follow the individual across the whole web. Then the surveillance got embarrassing. Safari and Firefox killed third-party cookies years ago; Apple's tracking prompts gutted mobile ad IDs; regulators kept tightening; and Chrome, after half a decade of promising a full deprecation, landed on a shrug — user choice, restrictions, and a signal that keeps degrading either way. The follow-people-around era didn't so much end as bleed out.

What got rebuilt on the wreckage, and what you'll actually buy in 2026: first-party data — the logins, purchases, and loyalty sign-ups people gave you directly, now the scarcest asset in marketing. Contextual placement — the ad matches the page, not the person: espresso machines advertised on coffee reviews, back to 1964 but with machine-read pages and real-time bidding. Modeled and lookalike audiences — statistical guesses that someone resembles your buyers, replacing observed trails with inference. And clean rooms — neutral enclaves where your first-party data and a platform's data are matched under privacy constraints, so you can measure without either side handing over the raw lists.

Notice two ironies. First, contextual targeting is accidental jobs-to-be-done: the page someone is reading right now is a decent proxy for the job they're doing right now — the moment is legible even when the person is opaque. Second, signal loss quietly repriced the whole debate. When individual-level precision gets expensive and leaky, broad reach gets relatively cheap and reliable — which is to say, the infrastructure drifted toward Sharp's side of the argument without ever reading his book.

8The synthesis

So: Kotler says slice and pick. Christensen says the slice that matters is the job. Sharp says most slicing shrinks your future. These sound like three religions; they compose into one working position — what Sharp's own school calls sophisticated mass marketing:

  • Define the market by the job, and define it broadly. You're not "for urban professionals" — you're for everyone who hits the boring-commute job, or the small-yes-to-your-kid job. Jobs give you Kotler's structure without demographics' lies.
  • Buy reach across the whole category. Your next buyers are light buyers you can't name in advance, and they look like everyone else's buyers. Show up cheaply, broadly, and distinctively, so the memory slot from Chapter 1 exists when the job fires.
  • Tailor creative to situations, not to segments. One brand, many moments: a 7 am message and a Saturday message, not a "millennials" message. The scene changes; the flag doesn't.
  • Spend precision only where the math clears. Named accounts, loaded carts, genuinely small categories. Everywhere else, precision is a premium-priced way to talk to yourself.

One thread is still hanging, and it's load-bearing: this chapter kept asserting that growth comes from light buyers and penetration, and asked you to take the dot field's word for it. Chapter 5 opens the evidence locker — the panel data, the laws named after unglamorous statisticians, and the reason your loyalty program is not going to save you.

Check yourself

3 questions · instant feedback 0 / 3