Segmentation is a cut of the customer base that changes a decision. The test is simple: if two customers sit in different segments, you would price, serve, or sell them differently. If you would not, you drew a slide, not a tree.
Do not name STP out loud. Say: “I would split customers by the variable that drives willingness to pay and cost-to-serve, then pick who we will not chase.”
What a useful cut looks like
A segment needs three properties:
- Observable in the case (or you can name the proxy you would ask for).
- Different economics — price, frequency, churn, or cost-to-serve.
- Actionable — a sales motion or product spec can target it.
“Engaged users” fails all three. It is circular, it overlaps with “high spend,” and it does not tell you who to hire for.
| Cut | When it works | When it is decoration |
|---|---|---|
| Size (seats, beds, tonnes) | Cost-to-serve and price power scale with size | Size does not change the pitch |
| Need / job-to-be-done | Product specs differ | You invent five “personas” with no numbers |
| Channel or payer | Who pays is not who uses | Channel is just a logo list |
| Geography | Regulation, density, or logistics differ | You slice 50 states with no thesis |
Pick one primary cut, then a secondary only if the first still mixes two businesses. Three simultaneous cuts (size × industry × region × persona) is a spreadsheet, not a case structure.
Worked example: clinic software, not a grocery promo
Prompt. Chartwell sells EHR-lite software to outpatient clinics. 2,400 accounts. Net revenue retention is 96%. The CRO wants to “segment for a new mid-market motion.” Marketing’s current grid is Champion / Neutral / At-risk — a recency score.
That grid overlaps with revenue: “champions” are the accounts that already expanded. Targeting them is not a segment; it is a tautology.
Better primary cut: number of clinicians (observable, drives both ACV and implementation cost). Secondary: whether the clinic is independent or owned by a health system (who signs, who pays).
Pull the book:
| Segment | Accounts | Avg ACV | Gross margin | NRR | Cost-to-serve / yr |
|---|---|---|---|---|---|
| 1–3 clinicians, independent | 1,520 | $4,800 | 72% | 91% | $1,100 |
| 4–12 clinicians, independent | 610 | $18,400 | 68% | 104% | $2,400 |
| Health-system outpatient (any size) | 270 | $41,000 | 51% | 99% | $14,000 |
Contribution after cost-to-serve:
- 1–3: (4,800 × 0.72) − 1,100 = $2,356 per account. Fine margin, negative NRR — this is a leaky bucket. A “mid-market motion” will not fix 1,520 small logos that churn.
- 4–12: (18,400 × 0.68) − 2,400 = $10,112. NRR 104%. This is the business. Density of clinicians makes workflow stick. They buy add-on billing.
- System: (41,000 × 0.51) − 14,000 = $6,910. Higher ACV, worse contribution than 4–12, because RFPs, security questionnaires, and a six-month IT steering committee sit in cost-to-serve. NRR 99% is not expansion; it is survival.
Recommendation. Do not staff a “mid-market” team as a third generic pod. Define mid-market as independent 4–12 clinician clinics. Put two SEs and one implementation lead on that band only. Stop spending CSM time on 1–3 except a self-serve success track (the $1,100 cost-to-serve is eating the book). Bid on health systems only when the contribution after RFP cost still clears $8k — most will not.
Risk. 4–12 independents get acquired by systems and flip into the expensive segment. Next step: tag which of the 610 sit in MSAs with active system roll-ups.
The so-what is a number: the 610 accounts produce more contribution per logo than the 270 “big” logos everyone wants to toast.
When segmentation is the wrong first tool
Profit fell and you have not isolated the driver. Mix might be a segment story, but start with a profitability bridge. Do not open with personas.
The decision is capacity or operations. A kiln does not care about buyer personas; it cares about tonnes and energy. Segment later if you are allocating scarce output to contracts.
There is only one buyer. A city granting a concession is not a segmentation case.
The mistake unique to this tree
Segmenting on the outcome you want to predict. “High-value customers” is defined by value, then used to explain value. In the Chartwell book, Champion / At-risk is NRR in costume. Cut on an input (clinician count, ownership) and measure value.
The second unique fail is overlapping cuts presented as MECE: SMB vs enterprise vs “digital natives.” Digital natives sit in both SMB and enterprise. See MECE.
How to use it in 45 seconds
- Ask what would change commercially if two customers switched segments.
- Choose one observable split; table economics by that split.
- Name who you will not serve, or not serve with humans.
- Tie the recommendation to contribution and retention, not to a persona nickname.
Cut the book on an observable
Structure drills punish overlapping personas. Practice one economic split and a kill list.