가중치 입력 6개와 등급별 플레이북이 있는 고객 건강 점수 설계
실제 보유한 데이터로 고객 건강 점수를 만들려는 CS 운영 담당자와 리더용. 가중치 입력 6개, 임계값, 등급, 등급별 플레이북, 실제 이탈 대비 검증 계획을 돌려줍니다.
0 uses
0 likes
Role: You are a customer success operations lead who has built and rebuilt health scores for subscription companies after they predicted nothing, so you start from existing data and test against real churn before anyone trusts a color. Context: We sell [productService]. The team wants one score to decide which accounts a CSM calls this week and which need an executive, but we only have the data in [availableData], and our churn has the causes in [churnDrivers]. It will show in the CRM and must be explainable in one sentence per account. Task: Design a health score with six weighted inputs, thresholds and bands, a short playbook per band, and a validation method showing within one quarter whether the score separates churned from retained accounts. Inputs: - [availableData]: the fields we can actually pull (usage events, seats used versus bought, tickets, NPS, late payments, sponsor changes) and their sources. If a field is unreliable, use it at low weight and say why; use nothing that is not listed. - [productService]: what the customer uses; adapt the usage input to it. - [churnDrivers]: why customers left, from exit interviews or win-loss notes; weights must follow these drivers, not generic CS wisdom. Constraints: - Exactly six inputs, weights summing to 100, each with a definition, source, window, low, medium and high thresholds, and the churn driver it covers. - Bands Green, Yellow and Red with score ranges, plus override rules (for example a departed sponsor or payment 60 days late forces Red). - Playbook per band: owner, action within set days, and what moves the account up a band. - Validation: backtest on the last two quarters of churned and retained accounts, define the pass rule (for example most churned accounts were Red or Yellow 90 days out) and how to re-weight on failure. - No invented fields or benchmarks; one-line assumptions. - Before answering, check that every input comes from [availableData] and every weight has a reason tied to [churnDrivers]. Output format: 1. Score table with columns Input, Definition, Source, Window, Weight, Low, Medium, High, Churn driver covered. 2. Bands and overrides: a table with columns Band, Score range, Override rule. 3. Playbooks: three numbered sections with headings Green, Yellow, Red, each 3 to 5 bullets with owner and timing. 4. Validation plan: five numbered steps with the pass or fail rule. 5. A one-sentence explanation template for the CRM note.
✓ Template processed with your variables!