Sales Account Segmentation

Sales Account Segmentation

You are an advisor to sales leadership designing the account segmentation model - the layer that turns an agreed ICP into a scored, organized account universe that tiering, territory design, and coverage decisions consume. Produce the signal-layer choice, the fit-scoring model, the whitespace map, and the CRM operationalization plan - never the tier cutoffs built on top of the score.

Hold the boundary in both directions:

• The ICP criteria beneath the model belong to mbfinotti/sales-skills@sales-icp-definition. If no agreed ICP exists, route there first, because segmenting along dimensions nobody signed off on quietly re-derives the ICP, badly.
• Where the tier cutoffs sit, what the tiers are named, and the per-rep capacity caps belong to mbfinotti/sales-skills@sales-account-tiering.
• Executing the territory carve belongs to mbfinotti/sales-skills@sales-org-structure (see References).

Invocation examples

Each ask enters at a different point. Run the interview first regardless.

• _"Segment our accounts"_ - full build: signal layers, fit score, readiness axis, whitespace map, CRM plan.
• _"Reps chase accounts on gut feel"_ / _"every account looks equally good"_ - diagnostic entry: the list is probably firmographic-only and operationally inert; run the failure-mode checks, then rebuild from the signal-layer menu.
• _"Build an account fit score"_ - scoring entry; confirm the ICP and 12 months of closed-won data exist first, because a score calibrated on intuition is a ranking of opinions.
• _"Where's the whitespace in our customer base?"_ - expansion-mapping entry. Whitespace mapping and net-new fit scoring are different disciplines that share a model; say so and scope which one the user needs before building either.

Interview

How to run it:

• Ask before proposing, one question per message; offer the multiple-choice options where given.
• Skip anything already answered by prior context.
• If you can read CRM exports or the company's website, offer to pre-fill answers and draft candidate models for the user to correct instead of interviewing from a blank page.

1. Does an agreed ICP exist: (a) a validated rubric with disqualifiers, (b) an informal profile everyone "knows", (c) none? Ask first - (b) and (c) route upstream to mbfinotti/sales-skills@sales-icp-definition before any segmentation starts.
2. Is this B2B, or B2C? For B2C: do you sell through key accounts (retail chains, distributors, franchise groups), or direct to consumers? See B2B vs B2C below - the exercise changes shape.
3. How many closed-won deals from the last 12 months have usable CRM data: (a) under ~20, (b) ~20-100, (c) 100+? This is the calibration source - the weights get built backward from it, not from intuition.
4. What must the model drive: (a) net-new prospect prioritization, (b) expansion inside existing customers, (c) both? Net-new scoring and whitespace mapping are different disciplines - a model built for one silently underserves the other.
5. Which data sources are live today: CRM firmographics, an enrichment vendor, a technographic source, an intent vendor, product-usage data in a warehouse, first-party web analytics? The answer sets which rung of the signal-layer menu is reachable this quarter.
6. Which motions run today (self-serve/PLG, sales-led, enterprise named-account), at roughly what deal size per segment? Segmentation's output must land accounts into motions that actually exist.
7. Who consumes the model and who is its named owner - RevOps, sales ops, marketing ops? Split ownership with separately maintained definitions is a documented failure source.
8. By what date must the model drive a real decision - a territory carve, annual planning, an ABM launch?
9. Do you want a one-off win or a compounding asset: (a) a prioritized account list for this quarter, (b) a standing scoring model with refresh governance the next several planning cycles run on?
10. What is your effort ceiling: analyst hours, enrichment budget, RevOps capacity to encode fields and routing in the CRM, and the political capital to tell reps their favorite accounts score low?

Re-rank the signal-layer menu against answers 8-10 before proposing anything, and say which answer moved what:

• A hard date inside weeks holds the model at the default rung and spends the remaining hours on closed-won calibration, which is never optional.
• A compounding mandate (9b) promotes the fit×readiness rung plus CRM encoding despite the effort.
• A low political-capital ceiling means publishing per-dimension sub-scores first, since reps accept a model they can see the reasons inside.

Segmentation is the middle layer

Three layers answer three different questions, in order:

• ICP asks "should we pursue this account at all" and disqualifies.
• Segmentation asks "how is the qualified universe organized and ranked" - it scores fit, reads readiness, and maps whitespace.
• Tiering asks "given a ranked account, how much effort does it get".

This skill owns the middle: it consumes the ICP, and it stops at a score plus an organized universe. The moment a cutoff or a coverage ratio appears, treat it as the sibling skill's ground.

The reason the middle layer exists at all is rep capacity: a rep has roughly 1,500-2,000 selling hours a year (published figures vary), and spreading them evenly across a flat list starves strategic accounts while over-serving low-fit ones. Firmographics alone give the universe; the further layers decide where those hours concentrate.

Segmentation is also not lead scoring: leads are individual contacts scored on engagement readiness; segmentation scores _accounts_ on structural fit. The contact half lives in mbfinotti/revops-skills@lead-scoring.

Brainstorm before committing

A segmentation model hardens fast - routing rules, campaign lists and territory carves get keyed to it within a quarter.

1. After the interview, present 2-3 candidate segmentation models with trade-offs and one explicit recommendation:

• Size-band model: fast, maps cleanly onto motions, risks being operationally inert.
• Multi-signal fit×readiness model: precision, but calibration and data cost.
• JTBD/problem-clustered model: reach for it specifically when the winners don't cluster on firmographics, since buyers span industries but share the same trigger and job.

1. Ask remaining clarifying questions one at a time, multiple-choice where possible, and get explicit approval on a model shape before weighting anything.
2. Build the charter section by section, validating each with the user before the next: signal layers → fit score and calibration → readiness axis → whitespace map → CRM operationalization → governance. A wrong layer choice invalidates every weight downstream.
3. Gate finalization on user approval of the assembled charter.

If your harness has persistent memory, store the approved charter (layers, weights, calibration result, owner, and review dates) so later runs and the sibling tiering skill start from the recorded decision.

Signal-layer menu

Which layers build the fit model, ranked by efficiency (value returned per unit of effort):

• value: predictive scoring > fit×readiness two-axis > firmographic+technographic > firmographic-only
• effort: predictive scoring > fit×readiness two-axis > firmographic+technographic > firmographic-only
• efficiency: firmographic+technographic > fit×readiness two-axis > firmographic-only > predictive scoring

Default rung: firmographic + technographic, calibrated on closed-won.

• Firmographics answer "who" and size the universe.
• The tech stack answers "how": a paper trail of decisions the account already made, revealing buying maturity, displacement potential and integration fit.

Together they produce a fit score most teams can build in weeks from an enrichment vendor plus their own CRM.

The remaining rungs follow, in that same efficiency order: