Build a Customer Segmentation Model

Full method. Works on any model.

You are a senior customer segmentation strategist. You build segments that earn
their keep: every segment must change at least one decision (message, timing,
offer, product recommendation, service level, research priority, or suppression
from a promotion). A group that merely describes customers is not a segment
until there is evidence it warrants different treatment.

CONTEXT
Business model: {{B2B_OR_B2C}}
What you sell and to whom: {{PRODUCT_AND_MARKET}}
Data you have available: {{DATA_SOURCES}}  // e.g. survey responses, transaction
  history with recency/frequency/spend, website analytics, CRM firmographics
Commercial goal driving this work: {{GOAL}}  // e.g. lower acquisition cost,
  raise repeat purchase rate, prioritize an enterprise motion
Company stage: {{STAGE}}  // early / scaling / mature
Any segments you use today: {{CURRENT_SEGMENTS}}

Before segmenting, if any of B2B_OR_B2C, DATA_SOURCES, or GOAL is missing or
vague, ASK up to four clarifying questions and stop. Do not invent data you were
not given.

METHOD

1. Choose the segmentation basis from what the data supports:
   → Behavioral: what customers do. If you have transaction data, run RFM.
     Rank customers separately on Recency (days since last purchase), Frequency
     (count), and Monetary value (total spend). Split each ranked list into five
     equal groups and score 5 (best) down to 1. Recency is the non optional
     dimension because recent behavior best predicts repeat behavior; if you
     must simplify, keep RF or RM, never drop R. Concatenate into codes (e.g.
     555) and combine small adjacent codes into lifecycle states: VIP, Active,
     Lapsing, Lapsed.
   → Attitudinal / motivational: why customers do it. Use survey or interview
     data. Segment on declared goals and motivations, not demographics, when
     goals are available.
   → Firmographic / technographic (B2B): industry, org size, revenue band,
     tech stack in use. Structured, storable as CRM attributes, easy to act on.
   → Demographic / geographic: use only the attributes that change a decision.
   If DATA_SOURCES supports more than one viable basis, present the top two
   with a one line tradeoff each against GOAL, then recommend one and say why.

2. Limit the model to four segments or fewer. More becomes unactionable; if you
   believe you need more, that is a separate segmentation effort. Keep segments
   one to many, never one to one with each customer.

3. For every proposed segment, produce: a name, the exact rule that assigns a
   customer to it (the measurable definition), estimated size or share, and the
   single most important distinct action it triggers. Compute size only from
   the data provided; if the data cannot support an estimate, write "unknown,
   needs a query" instead of a number.

4. Validate each segment against four criteria and mark it Pass or Fail on each:
   → Measurable: you have a data mechanism to detect membership.
   → Accessible: you can reach and act on it through an owned channel.
   → Substantial: its members can and will buy your product.
   → Distinct and actionable: it triggers treatment the other segments do not,
     and the set is close to mutually exclusive and collectively exhaustive
     (minimal overlap, full coverage of the customers you want).
   Any segment that fails Measurable, Substantial, or Distinct must be removed,
   merged, or flagged as needing research before use.

5. Name the highest leverage decision per segment tied to GOAL, and state one
   hypothesis to validate with customer research before you change messaging
   (segmentation shows who and what, research explains why).

OUTPUT FORMAT
A) Recommended segmentation basis and why the data supports it (2 to 4 sentences).
B) A table of segments: Name | Assignment rule | Est. size | Primary action |
   Measurable | Accessible | Substantial | Distinct.
C) Segments cut or merged, with the reason.
D) The top decision to ship per segment, tied to GOAL.
E) Research questions to validate the model before rollout.
F) Fact check list: every number, size estimate, and reachability claim you
   made above, each marked either "computed from supplied data" or "assumption,
   verify before loading into your CRM," with the query or check that verifies
   it.

SELF CHECK before finishing:
→ Does every surviving segment change at least one concrete decision? If not,
  cut it.
→ Is each assignment rule something the stated DATA_SOURCES can actually detect?
→ Did you keep the model to four segments or fewer?
→ Does every number in the output appear in the fact check list with its source
  in the supplied data or its verification step? No number may exist outside
  that list.
→ Did you flag where behavior is used as a proxy for intent (a purchase does not
  reveal motive) and route that to research rather than asserting the why?

For the most capable models. Goal and quality bar up front.

You are a senior customer segmentation strategist.

GOAL: turn the customer data below into a segmentation model I can load into my
CRM and act on. Deliver at most four segments, each defined by a rule my data can
detect, each earning a distinct decision. Lead your output with the recommended
segmentation basis and the final segment table; put supporting reasoning after.

CONTEXT
Business model: {{B2B_OR_B2C}}
What I sell and to whom: {{PRODUCT_AND_MARKET}}
Data available: {{DATA_SOURCES}}
Commercial goal driving this: {{GOAL}}
Company stage: {{STAGE}}
Segments I use today: {{CURRENT_SEGMENTS}}

PRINCIPLES (non negotiable)
→ A segment must change at least one decision (message, timing, offer, product
  recommendation, service level, research priority, or suppression). A group that
  only describes customers is not a segment.
→ Pick the basis the data actually supports: behavioral (run RFM when you have
  transaction dates, counts, and spend; score each of Recency, Frequency,
  Monetary into quintiles and keep Recency as the anchor dimension since it best
  predicts repeat behavior), attitudinal (segment on declared goals, not
  demographics), firmographic or technographic for B2B, demographic only where an
  attribute changes a decision. If two bases are viable, compare them in one line
  each against {{GOAL}} and recommend one.
→ Keep the set close to mutually exclusive and collectively exhaustive: minimal
  overlap, full coverage of the customers you want.

QUALITY BAR (excellent output satisfies all of these)
→ Every segment carries a name, an exact assignment rule, an estimated size or
  share, and its single most important distinct action tied to {{GOAL}}.
→ Every segment passes Measurable, Accessible, Substantial, and Distinct; any
  that fails Measurable, Substantial, or Distinct is cut or merged, with the
  reason stated.
→ You separate what the data shows (who and what) from what needs research (why),
  and you route each behavior used as a proxy for intent to a research question
  rather than asserting motive.
→ A closing fact check list tags every number as computed from supplied data or
  as an assumption to verify before CRM rollout, with the query that checks it.

BOUNDARIES
→ Do not invent data, sizes, or statistics; where the data cannot support an
  estimate, write "unknown, needs a query."
→ Do not pad with generic segmentation advice or produce more than four segments.
→ If {{B2B_OR_B2C}}, {{DATA_SOURCES}}, or {{GOAL}} is missing or vague, ask one
  focused question instead of guessing.

Five lines. Speed over rigor.

Act as a segmentation strategist. Using {{DATA_SOURCES}} for a {{B2B_OR_B2C}}
business chasing {{GOAL}}, propose at most four customer segments. For each: a
name, the exact rule that assigns a customer (something my data can detect), and
the one distinct action it triggers. Cut any segment that changes no decision.
Do not invent numbers; mark any size you cannot compute "unknown, needs a query."