Data & BI

Verified against ChatGPT · 2026-08-14

Build a customer segmentation that tells your team what to actually do differently for each group

Designs a customer segmentation scheme tied to a specific business action, rejecting descriptive-only segments that are interesting to look at but don't change what anyone does for any given group.

ChatGPT (GPT-5.1)4 fillable variables

The prompt

Ready to copy — highlighted parts are example details you can swap.

Help me design a customer segmentation. The test I want you to hold every segment to: if this segment didn't exist, would the team behind it actually treat that group of customers any differently? A segmentation that's interesting to look at but doesn't change a single action isn't done yet.

BUSINESS GOAL FOR SEGMENTING
Reduce churn among mid-market accounts by targeting retention efforts where they'll have the most impact.

DATA AVAILABLE FOR SEGMENTATION
Monthly spend, tenure in months, product usage frequency (logins/week), support ticket count, plan tier.

WHO WILL ACT ON THESE SEGMENTS
The customer success team, who can assign a dedicated CSM, offer a training session, or flag an account for an executive check-in.

EXISTING SEGMENTATION IF ANY
Currently just split by plan tier (Basic/Pro/Enterprise), which the CS team says doesn't actually predict who's about to churn.

Do this:
1. Propose 3-5 candidate segments based on the available data and business goal, each defined precisely enough (specific thresholds, not vague labels like "high value") that a person could look at any customer record and assign them to exactly one segment without ambiguity.
2. For each candidate segment, state the specific different action the team named would take for that group versus another group — if you can't identify a concrete different action, cut the segment or merge it into a neighboring one rather than keeping it for descriptive completeness.
3. Check for segments that sound different but would actually contain almost entirely the same customers in practice given the data available (e.g., "high spend" and "long tenure" segments that are 90% overlapping in this business) — flag any such overlap and recommend collapsing them.
4. If an existing segmentation is already in use, explicitly compare your proposal against it and say whether it's worth the disruption of changing, or whether the existing one is fine and doesn't need replacing.

WHAT NOT TO DO
Do not propose a segmentation scheme purely because it's a common industry framework (RFM, personas, life-stage) unless it passes the different-action test above for this specific business — a well-known framework that doesn't change anyone's behavior here is still not worth shipping.

OUTPUT FORMAT
1. Segment definitions (precise, threshold-based), 3-5 total after merging
2. Action mapped to each segment, stated as what the named team does differently
3. Overlap check results and any recommended merges
4. Verdict on replacing the existing segmentation, if one exists, with a one-line reason

Customize

Optional — swap in your own details for the highlighted parts above.

Why this works

Holding every candidate segment to an explicit different-action test is the load-bearing instruction here, because segmentation is one of the analytics tasks most prone to producing output that looks sophisticated and complete while being entirely inert — a model asked simply to "segment these customers" will happily produce a clean, well-labeled scheme (power users, casual users, at-risk users) that reads as thorough analysis but that no one downstream actually changes their behavior based on, since nothing in a generic request forces the connection back to a concrete action; requiring that connection explicitly, and requiring segments that fail the test to be cut rather than kept for descriptive completeness, is what turns a taxonomy exercise into an operational tool. Requiring precise, threshold-based segment definitions rather than descriptive labels addresses a related failure: "high-value customer" sounds like a segment but isn't actually usable until someone defines the exact spend or usage threshold that puts a customer in it, and without being pushed for precision a model will happily generate segment names that sound rigorous while leaving the actual boundary fuzzy, which means two different people applying the scheme would sort customers differently. The overlap check specifically targets a subtle but common segmentation failure: a business's data often has enough natural correlation between different-sounding dimensions (long-tenure customers often are high-spend customers) that two segments proposed independently turn out to describe almost the same population, adding apparent granularity without adding real distinctions, and this only gets caught by explicitly checking correlation against the actual available data rather than assuming distinct-sounding labels imply distinct populations. Requiring an honest comparison against an existing segmentation, including the option to conclude the existing one is fine, keeps the model from defaulting to always recommending a change just because it was asked to design something — genuine analytical honesty here includes the possibility that no change is warranted.

What you get back

Segment: 'At-risk mid-market' = plan tier Pro or Enterprise, login frequency down more than 40% versus their own 90-day average, tenure over 6 months. Action: CS assigns a dedicated CSM outreach within 5 business days, versus no proactive outreach for stable accounts. Overlap check: a proposed 'low engagement' segment based purely on login frequency alone overlapped roughly 85% with this at-risk segment once tenure and tier were factored in — recommend merging rather than keeping both. Verdict on existing plan-tier-only segmentation: worth replacing, since plan tier alone doesn't capture the usage-decline signal the CS team says actually predicts churn.

Verified against

ChatGPT GPT-5.1 · 2026-08-14

Changelog

  • 2026-08-14 Initial publish, verified against ChatGPT GPT-5.1.

Need this built into your business?

If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.

EXPLORE WHAT SCULT BUILDS
All Data & BI prompts

Check your AI visibility

One URL in, a 0–100 score and the exact fixes out.

RUN THE CHECK

Browse all the tools

15 tools across six categories
13 of them never send your data anywhere

Free · No signup · No trial clock

SEE THE DIRECTORY