Lesson 86.1 mapped the journey every lead walks. Now the path forks. The reader studying how models are evaluated and the builder deploying an application this week downloaded the same worksheet — but the same follow-up email helps one and annoys the other. Segmentation is how you stop sending everybody everything.

The vocabulary, in plain language

• Segment: a group of contacts who deserve the same next step — defined by what they did and need, not by who you guess they are.
• Behavior: what they observably did — finished the worksheet, attended office hours, opened pricing twice, stalled at checkout.
• Preference: what they asked for — weekly notes only, no calls, plain-text email, topics they subscribed to.
• Fit: whether your offer can actually serve them — right budget, right stack, right region, right problem. Low-fit interest is information, not pipeline.
• Intent: how close the behavior looks to a decision — comparing plans, inviting a teammate, asking about migration. Treat intent as a hypothesis, never a certainty.
• Suppression: the deliberate decision *not* to send — unsubscribed, bounced, already converted, already got the nudge, asked for a pause. Suppression is a feature, not a failure.
• Personalization: inserting the right next thing for the segment — the correct lesson, the correct offer — not printing their first name in a generic blast.

The mental model: a good segment is a hospital triage tag, not a horoscope. It says "given what we observed, this person needs *this* next," and it expires the moment the observation changes.

Real jobs, not guesses

Weak segments guess at identity: "young professionals," "tech-savvy moms," "SMB decision-makers." Nobody wakes up as a demographic. People wake up with a job: "I must cut inference costs before launch," "I need one credible filing check for Monday's meeting." Build segments from four observable materials:

1. The job they hired you for — which offer they took, which worksheet they finished, which question they asked support. 2. Actual content use — lessons completed, checklists downloaded, office hours attended. Behavior beats self-description. 3. Decision stage — learning, comparing, trialing, buying, renewing. Same person, different stage, different message. 4. Consent and preference — what they agreed to and how often. This is a boundary, not a suggestion.

ByeBuy makes the contrast concrete. A Models reader is studying Part IV: they open evaluation lessons, retake the leaderboard quiz, save explainers. Their job is understanding. A deploy-builder is shipping: they downloaded the deployment checklist, asked about environment variables, started a trial. Their job is launching. Same list, opposite next steps — the reader earns the "how to read a new model release" note; the builder earns the "ship checklist + office-hours invite." Send each the other's message and both conclude you do not understand them.

The three parts of every segment

Every working segment needs exactly three lines — no more to start:

• Signal: the observable fact that puts someone in (and the fact that takes them out).
• Useful next action: the one message, invite, or offer the segment receives.
• Stop condition: when sending ends — converted, unsubscribed, went silent for N days, moved to another segment.

The stop condition is where most teams fail. Without it, segments become permanent buckets that everyone mails forever — which is how "engaged subscribers" become spam complaints. A segment without an exit is a trap.

Tool note: the segment logic above is tool-independent — it works in any CRM or mailer. When you implement it, build segments as saved filtered views and suppressions as list exclusions, not as static copied lists. For the mechanics in one common CRM, see the HubSpot knowledge base; evaluate any interface against the three-lines rule here, not the other way around.

A small local example: the catering shop keeps three segments — "tasted at event, no quote yet" (signal: tasting card scanned; action: one quote email with two package options; stop: quoted or 14 days silent), "quoted, deciding" (signal: quote opened twice; action: one call offer plus a photo menu; stop: booked or declined), "booked this quarter" (signal: deposit paid; action: care notes only until the event; stop: event delivered, then move to referrals). Three segments, three exits, zero blasts.

Exercise: write three segments

Create three segments for your own project. Each gets a signal, a next action, and a stop — grounded in something you actually observed.

# SEGMENTS.md — [Project]

## Segment 1 — [Name, e.g. "Worksheet finishers, not yet building"]
- Signal (in): ___ (observable behavior + source)
- Fit check: ___ (who this excludes and why)
- Useful next action: ___ (one message/offer, channel, timing)
- Stop condition: ___ (converted / moved / silent ___ days / suppressed because ___)

## Segment 2 — [Name, e.g. "Trial starters, stalled"]
- Signal (in): ___ / Signal (out): ___
- Useful next action: ___
- Stop condition: ___

## Segment 3 — [Name, e.g. "Customers ready for referral"]
- Signal (in): ___ / Signal (out): ___
- Useful next action: ___
- Stop condition: ___

## Suppression list (never mailed from these segments)
- ___ (unsubscribed / bounced / asked for pause / already converted / teammate account)

Worked mini-example — creator-led publication: "filing-check downloaders, never opened follow-up" (signal: download, zero opens in 10 days; action: one plain-text "which check would actually help Monday?" reply-prompt; stop: reply, open, or 1 send then suppress), "repeat openers" (signal: 3+ opens; action: invite to the paid teardown; stop: joined or declined), "members" (signal: paying; action: member-only notes, never the acquisition pitch; stop: churned → move to reactivation).

Finish line: three segments, each with a real observed signal, one next action, and an explicit stop — plus a suppression list.

Verify quickly: for each segment, name one real person (or anonymized record) who belongs in it today. If a segment has zero members, it is a guess — replace it with behavior you have actually seen.

Common failure mode: the horoscope segment — "engaged millennials interested in AI." Nobody can enter it, nobody can leave it, and every message to it feels slightly wrong to everyone receiving it. Delete it and rebuild from one observed action.

Check your understanding

1. What four observable materials build a segment, and why is each stronger than a demographic guess? 2. Why does every segment need a stop condition? What happens without one? 3. How do the Models reader and the deploy-builder differ in job, signal, and right next step?

Next

Segments tell you *who* gets *what*. Lesson 86.3 turns that into a bounded test — a campaign brief with one objective, one audience, one promise, and a clock — so you learn something every two weeks instead of mailing forever.