The segment of one: treating every customer as a market
The average-persona logic is dying. Netflix saves over a billion dollars a year with recommendation (80% of what gets watched comes from it), and about 35% of Amazon's revenue comes from its personalization engine. This lesson shows you the shift from 'segmenting groups' to 'treating each customer as their own market', and the realistic path for an operator to start small with owned data (zero- and first-party), without building an agency and without crossing the invasion line.
Think of a customer of yours who bought three times this year. Now think of another one, who came in yesterday and hasn't bought anything yet. In old-school marketing, both fell into the same bucket: "man, 35 to 44, class B, interested in productivity". And they got the same email, the same offer, the same page. The question this lesson raises is simple and uncomfortable: why do two such different customers get exactly the same thing? The answer has always been "because treating each one their own way doesn't fit the budget". AI just changed that math.
You have a client base at your accounting firm. Today everyone gets the same "tax tips" newsletter. But the sole proprietor who registered last month and the company under the simplified tax regime about to switch brackets are two different markets, with different fears. Treating each one by their moment (not by the average revenue bracket) is the segment of one applied to your funnel. And the data you need for this, you already have: it's in your own records.
Your firm serves everyone from the startup that needs a founders' agreement to the family going through probate. Sending the same generic "legal news" content to both is wasting the relationship. The segment of one says: the probate client gets what speaks to their pain, at their pace; the startup gets something else. You don't need data bought from outside, you need to use what the client already told you in the intake consultation.
You run a store and blast the same campaign to the whole list: "20% off everything". Whoever never bought and whoever is a loyal customer get the same coupon. The segment of one flips this on its head: every person is a market. The newcomer gets an invitation for their first purchase; the loyal one gets the launch before anyone else. Same base, different messages, because you stopped aiming at the average and started aiming at the person.
You send the same onboarding email to everyone who joins the company. But the senior developer coming from a multinational and the intern in their first job are two different worlds, with different questions in week one. The segment of one applied to HR says: every hire gets the track that speaks to their moment, not to the average of the new-hire cohort. And you already have the data for this in the hiring process, in what the candidate themselves told you in the interview.
You have a thousand users on the product and blast the same new-feature notice to all of them. But whoever uses the app every day and whoever joined last week and hasn't even activated yet are in opposite moments of the journey. The segment of one, in product management, is treating each user by their real behavior inside the product, not by the base's average. The newcomer needs activation; the power user needs the advanced feature. And you already collect that usage data, it's in your own events.
You have a full pipeline and send the same template proposal to every lead that comes in. But the lead who requested a demo three times and the one who downloaded a piece of content and vanished are two different markets, with different objections. The segment of one, in sales, is treating each account by its stage and behavior, not by the funnel's average. And the data for this is already in your CRM: what they opened, what they replied to, what they ignored.
You have a customer base with SLAs and treat every ticket through the same generic queue. But the customer who opened the tenth ticket this month and the one who's never complained about anything have different operational needs. The segment of one, in operations, is looking at each customer by their actual service and consumption history, not by the contract average. And you already have that data in your ticketing and order system, no need to buy anything from outside.
You run the same LGPD training and the same policy communication for the entire company. But the finance team, which handles banking data, and the marketing team, which handles a lead list, have different risks and different controls. The segment of one, in compliance, is calibrating the control to each area's real risk, not to the organization's average. And you already have that exposure map in your own data-processing records.
You have several teams consuming the same internal platform and send the same deploy notice and the same alert to all of them. But the service that deploys five times a day and the legacy one nobody has touched in months have opposite monitoring needs. The segment of one, in technology, is treating each service by its actual production behavior, not by the architecture's average. And that data is already in your own logs, metrics, and incident history.
You design a single onboarding flow and assume every user comes in through the same door. But whoever arrives on their phone in the middle of the street and whoever sits at a desktop with time to spare go through different journeys. The segment of one, in UX, is designing for each user by their actual context and behavior, not by the average of a research persona. And you already collect this usage data in your own testing and product telemetry.
You send the same quarterly results report to the whole board. But the director focused on risk wants to see the exposure map, and the one focused on growth wants to see the new-market funnel, and both get the same generic thirty-page PDF. The segment of one, in strategy, is building the memo around what that specific stakeholder decides on, not around the average of what "the board" supposedly wants to know. And you already have the data for this: it's in the minutes of the last meetings and in the questions each of them always repeats.
Let me start with a number I think will give you the good kind of scare. Netflix calculates it saves over a billion dollars a year just from its recommendation system, and that about eighty percent of everything people watch on the platform comes from that recommendation, not from someone actively searching. At Amazon, estimates put around thirty-five percent of revenue coming from the "people who bought this also bought that" engine. Think with me about what that means: the two biggest revenue machines in the digital world don't sell to "the public". They sell to you, specifically. This lesson is about how that logic stopped being a giant's privilege and became something an operator like you can start doing with what's already in your hands.
The core idea of this lesson. For decades, marketing meant grouping people into personas and aiming at each group's average, because treating every person their own way was too expensive. AI knocks down that cost. The new unit of personalization stops being the group and becomes the individual: every customer becomes their own market, your segment of one. The fuel for this isn't invasive tracking, it's owned data (what the customer hands you on purpose and what they do with you). And the good news for the operator is that this starts small, with the base you already have, without turning into an agency. There's just one line you can't cross, and I'm going to show you where it sits.
01The average persona is dying (and why it existed)
Let's call it what it is. The "persona", that sheet with a made-up name, age, pains, and a stock-photo avatar, wasn't anyone's foolishness. It existed for a practical reason: you couldn't talk to a thousand people in a thousand different ways, so you grouped everyone into three or four buckets and spoke to each bucket's average. It was the best you could do within a human budget.
The problem is that the average is a fiction. Nobody is exactly the persona. When you speak to a group's average, you speak a little bit wrong to every person in it. And the customer feels it. That email that arrives "Hello, [name]" and offers exactly what you already bought last week, that's the average persona showing its seams.
Think with me about what changed. The reason the persona existed was the cost of treating each person individually. AI drops that cost to nearly zero: it can read a person's history, understand their moment, and put together the right message for them, and repeat that for ten thousand people at once. When the cost of speaking 1:1 falls, the average persona loses its reason for existing. It's the same move you saw in finance back in N.fin.1: when the mechanical part gets cheap, what was worth gold moves elsewhere.
Notice this doesn't mean "throwing the persona in the trash tomorrow". It means understanding it was a cost crutch, and that the crutch is being pulled out from under your feet. Whoever keeps aiming only at the average is going to look increasingly clumsy next to whoever aims at the person. Fair enough?
02What the segment of one really is
Now the clean concept. Segment of one means treating every customer as if they were their own market segment, with a message, offer, and timing thought out for them, not for a group they were shoved into. The unit of personalization stops being the group and becomes the individual. That's it, no mysticism.
And here I need to pull you out of a language trap. "Personalizing" isn't putting the first name in the email subject line. That's just a label with a nice name. The segment of one is deeper: it's the message changing according to what that specific person did, bought, opened, ignored, and said they wanted. Netflix doesn't call you by name; it shows you different covers of the same movie depending on what you usually watch. That's the level. The content changes, not just the header.
There's a piece of jargon you'll run into and that's worth knowing: next-best-action. It's the idea of a system that calculates, for every person individually, what's the best thing to offer right now: which content, which offer, on which channel, at this moment. You don't need to build this all at once. You need to understand that's where the game is heading, and that every small step of yours in that direction already sets you apart from whoever blasts the same thing to everyone.
For the board. Segment of one isn't "Hello, [name]". It's the offer, the content, and the moment changing according to what that person did and said. The name-tag is makeup; the segment of one is the whole outfit tailored to fit.
03The cheap fuel: owned data, not tracking
Here's the part that separates whoever understood it from whoever just thought it looked nice. To treat every customer as a market, you need fuel, and the fuel is data. But there's data and there's data, and the difference is what keeps you on the right side of the law and of common sense.
The world spent years running on third-party data: tracking that follows the person around the internet, cookies from outside, purchased profiles. That world is ending, by regulation and by the platforms' own choice. And, honestly, it was the wrong fuel: expensive, fragile, and the kind that scares the customer. Forget it as a starting point.
The fuel that matters to you comes in two types, both legitimate and cheap:
- Zero-party data: what the customer hands you on purpose, because they want a better service. The answer to a quiz, the preference they marked, the "let me know when size 40 arrives". They gave it to you willingly.
- First-party data: what they do in their relationship with you. What they bought, what they opened, what they clicked, what they returned, how long since they last came back. You already have this, it's in your system, and it didn't cost anything beyond already being in operation.
Did you catch the turn? The fuel you need isn't sitting with some expensive external data provider. It's inside your own house, in your registration form, in your order history, in the answers customers already gave you. Personalizing for real starts by organizing what's yours, not by buying what's outside.
04The creepy line: where personalizing turns into stalking
Now the caveat, and it's serious. There's a point where personalization stops being useful and becomes creepy. The jargon for this is the creepy line. On this side of it, the customer thinks "how nice, they get me". On the other side, they think "how do they know that? scary". Gartner even projected that a huge share of consumers would refuse personalization that feels invasive. Crossing that line isn't just ugly: it burns the trust that's your most expensive asset.
The difference between the two sides is almost always one thing: does the customer understand how you know that? If you use what they told you on purpose (zero-party) or what they clearly did with you (first-party), they understand. "You bought sneakers, here's the sock that matches." Makes sense, doesn't scare them. But if you use something they never gave you, that seems to come from outside or from snooping, then they get the chills. "How do you all know I was thinking about that?"
This is where everything you already saw on the Guardian track connects. LGPD (which you saw in G.8) isn't bureaucracy, it's exactly the contract that defines what you can use and why. Owned data, with clear consent, is safe ground. And if you're going to put an AI to read customer data and build messages, remember the badge and locked room from G.5.5: that AI needs scoped access and a contained environment, because customer data is exactly the kind of thing that can't leak. Personalizing well and protecting data are the same job, not opposing ones.
For the board. The creepy line's practical rule: only use what the customer handed you on purpose or clearly did with you, and that they can understand where it came from. If the personalization needs an embarrassing explanation ("we tracked you"), you've already crossed the line.
05Starting small, without turning into an agency
I know the fear kicking in here: "this is Netflix-level stuff, I don't have engineers, I don't have a data team, I'm not building an internal agency for this". Great, because you don't need to. The segment of one, for the operator, doesn't start with a system. It starts with one split and one message.
Think about the realistic path, from least effort to most:
- Step one, split your base into two or three moments, not demographic personas. Forget age and class. Use what you already have: whoever never bought, whoever bought once, whoever is a repeat customer. Three buckets, based on real behavior, are already worth more than ten made-up personas.
- Step two, write a different message for each moment. The newcomer needs confidence and a first purchase. The repeat customer needs novelty and recognition. AI helps you write the three variations in minutes, in your brand's voice (which you saw how to protect in the content track).
- Step three, ask for one piece of zero-party data at a time. A short quiz, a question at signup, a "what are you looking for?". Every answer refines the person's bucket and gets you a little closer to 1:1, without buying anything from outside.
Notice that none of these steps require code, a recommendation engine, or a giant's budget. It requires you to stop aiming at the average and start aiming at each customer's moment, using what's already yours. It's the same spirit as the rest of the course: you're not going to "learn about AI", you're going to install a new logic into the way you already work. Market research already shows most executives saying 1:1 AI marketing has stopped being optional. The advantage is in starting now, small, with what you have.
Do it now
Take your customer base (or your email list, or your CRM, your real task or any one). You're going to build your first segment of one in three moves, on a single sheet.
- THREE MOMENTS, NOT THREE PERSONAS. Split your base into three buckets based on real behavior, not made-up demographics:
- Whoever never bought (or never engaged). - Whoever bought/engaged once. - Whoever is a repeat customer. Count roughly how many people fall into each. You can already feel where the opportunity lives.
- ONE SENTENCE PER BUCKET. For each of the three, write in one line what's the main pain or desire of that person AT THAT moment. What's the newcomer afraid of? What does the repeat customer want that the new one doesn't? Don't think about the base's average; think about each group's moment.
- THE FIRST PIECE OF OWNED DATA. Write ONE zero-party question you could ask the customer to refine this (a quiz, a signup field, a "what are you looking for today?"). Something they'd answer gladly and that tells you which bucket they're in.
At the end, run the creepy-line test: if you used each of these data points in a message, would the customer understand where you got it? If yes, you're on the right side. If any of them gives you the chills ("they'll think we spied on them"), cross it out and swap it for something the customer gave you on purpose.
Why Netflix's and Amazon's recommendation is worth billions
Netflix estimates its recommendation system saves over a billion dollars a year, and attributes about eighty percent of everything watched on the platform to that engine: most people don't search, they accept what was recommended specifically to them. At Amazon, market estimates put around thirty-five percent of revenue coming from the "people who bought this also bought that" engine. What matters to the operator isn't the giant technology behind it, it's the logic: both companies stopped selling to "the public" and started building a different storefront for every single person, fueled almost entirely by first-party owned data (what each person watched, bought, rated). They don't buy outside data for this; they use the behavior that happens in-house. The replicable lesson is this: the wealth isn't in some external data provider, it's in organizing and using what your own customer already does with you. The scale is theirs; the logic is yours, and it starts with three buckets in a spreadsheet.
Practice
1. Which sentence best describes what the segment of one is?
2. You're about to start personalizing your base's communication. Which fuel is the right, cheap, and legitimate starting point?
3. How do you know if a personalization crossed the creepy line?
Let's close the point together. The average persona was a cost crutch, and the crutch is being pulled out from under your feet: when speaking 1:1 gets cheap with AI, aiming at a group's average starts to look clumsy. The segment of one treats every customer as their own market, and the fuel for that isn't expensive data bought from outside, it's the owned data you already have at home, what the customer handed you on purpose and what they do with you. Netflix and Amazon prove that logic is worth billions, but the scale is theirs and the logic is yours. You start small: three buckets by moment, one message for each, one piece of owned data at a time. And you stop before the creepy line, using only what the customer understands where it came from, with LGPD on your side and AI in a scoped badge. Whoever does this stops shouting at a crowd and starts talking to people, one person at a time, at scale. That's the new marketing game, and it fits your budget. Shall we?
Thanks for the feedback. It helps sharpen the next lesson.