Find them by description, win them by need

Customers who look the same on paper can need different things, and need is what decides whether an offer wins them. Description (age, city, subject) helps you find people; it rarely tells you what to sell them or when.

Picture two 16-year-olds in Kyiv, both doing exam math at Klasna. One started in October. Families like hers, the steady exam families, average 7.61 lessons a month for almost ten months: they plan ahead and want steady progress. The other started in February. Crammers like him average 9.71 lessons a month and are mostly gone by June: they need an intensive plan to a date close at hand.

Same age, same city, same subject, yet a different offer, timing and message. That makes them different segments. A district, a phone model or the teen's gender explains none of this; the start month and the deadline behind it explain almost all of it.

There are two strong ways to form segments, and one way to reach them.

• Needs-based segmentation groups customers by their goal and situation: "families with an exam in June who start late", "adults who need English for weekly client calls".
• Behavioral segmentation groups them by what they do: pace, how long they stay, whether they come back. "About ten lessons a month, gone by June" is a behavioral definition of Klasna's spring crammers.

The two often describe the same group from two sides, and the best segments are clear on both: a need that explains the behavior. Description (age, country, who pays) comes after, as the way to find each group in a channel.

This order is the heart of the lesson: find them by description, win them by need. Starting from description ("parents aged 35–45") gives an audience without a reason to buy.

Sort what you know about customers into four kinds; each does a different job in a plan.

• Description: age, country, who pays. Helps you find a group in a channel.
• Need and situation: the goal and the deadline behind it. Tells you what wins them.
• Behavior: pace, how long they stay, whether they come back. Shows what they actually do.
• Value: what they pay over time. Shows what they're worth to the business.

A diaspora family from Klasna's export, sorted: lives in an EU country (description); the daughter keeps up Ukrainian math in case they return (need and situation); about four lessons a month for nine months (behavior); around 30,000 UAH over that time (value). In the export, diaspora families average 4.21 lessons a month for 9.58 months and 30,227.69 UAH per account.

Mixing the kinds is the usual mistake: describing a segment only by age and country gives you an audience to target, but not a reason anyone would buy.

Compare the value of groups precisely, then remember it's one input to the choice of whom to serve, not the choice itself.

Klasna's two exam groups: steady exam families paid 53,960 UAH per account on average; spring crammers 36,472 UAH. The difference: 53,960 − 36,472 = 17,488 UAH per account. Most of it comes from time, not price: the steady families stay about twice as long (9.67 active months against 4.78).

It's tempting to conclude that the steady families are simply better customers. But the crammers are more numerous (40 accounts against 24), arrive with an urgent need that Klasna can meet, and 30% of them come back in September. Which group is "better" depends on what Klasna wants and can deliver: growth, profit, tutor capacity in spring, the strength of its offer. That choice, with every input on the table, is the next module's job.

A synthetic persona is a customer profile generated by an AI assistant from text such as interview notes. It can suggest groups worth checking. It can't tell you what real customers need or how many of them exist, so it never replaces talking to them.

Uliana pasted interview notes into an assistant and got five personas with names, hobbies and favorite coffee. The useful way to treat them: for each persona, ask whether a group with that need and behavior exists in the export, and how big it is. The hobbies and the coffee are decoration; a vivid detail makes a persona feel real without making it true.

So AI can speed up the first draft of segments, but the data and real families decide. When time is short, the shortcut is fewer interviews, not invented customers.

Before customer material goes into an outside AI tool, apply two checks: minimize and consent.

Minimize: send only what the task needs. For finding segment hypotheses in interview notes, the assistant needs the goals, situations and quotes, not names, phone numbers, emails or any detail that identifies a family. Strip those first.

Consent: use notes only from people who agreed their answers could be used this way. A family that agreed to talk to Klasna's team didn't automatically agree to have its words processed by a third-party tool.

Two tempting shortcuts fail. Pasting the whole CRM "for accuracy" adds risk and doesn't fix the method: the personas still need checking against data and real conversations. And asking the assistant to "keep it private" isn't a safeguard: what you don't paste can't leak.

Klasna applies the same rule everywhere AI touches customers: public pages in freely, customer data only minimized and with agreement.

A one-person business segments the same way a school does: by need and situation, not by description. For an adult learner the two most telling facts are the goal and the deadline behind it.

Dana first sorted her eight adults (three students and five September enquiries) by age and job: thirties, forties, IT, design, sales. It told her nothing about what to offer. Sorted by goal and deadline, three groups appear:

• A fixed date: the exam candidate (exam in January), the designer (job interview in five weeks), the master's applicant (exam score by March).
• English for work every week: the QA engineer and the sales manager (calls with foreign clients or partners), the support lead (writing to foreign clients every day).
• No deadline: the bakery owner and the retired engineer, who simply want to speak someday.

Each group calls for a different offer, which a list by age and job could never show.

Once segments are defined by need, the offer follows from the need. Recall Dana's three groups, sorted by goal and deadline:

• Fixed date (the exam candidate, the designer, the master's applicant): a short sprint to the date, planned backwards from it.
• English for work every week (the QA engineer, the sales manager, the support lead): a steady weekly habit around real work tasks: their calls, emails and messages, not a general syllabus.
• No deadline (the bakery owner, the retired engineer): a gentle start with no pressure, small steps that beat putting it off.

A single generic offer ("English lessons for adults, any level") would fit none of them well. For one person with a few evenings, two groups are plenty; the point is that each gets an offer shaped by its situation.

How it fits together: description finds customers, need and behavior win them, value is one input, AI personas are hypotheses fed only minimized data, and a solo practice segments by goal and deadline. The next lesson asks who uses, who pays and who influences, and what a marketer may never target on.