Problem Statement
Analysts need a reliable way to transform weekly KPI dashboards into concise, executive-ready narratives that highlight significant metric movements, plausible drivers, follow‑up analyses, and recommended actions. The workflow must analyze KPI deltas, assess confidence, and produce structured insights while adhering to clear communication and data‑quality constraints. This enables leaders to quickly interpret performance shifts and make informed decisions.
Steps
Load and parse KPI inputs
INGESTIngest weekly KPI tables or dashboard exports, including optional targets, historical comparisons, and relevant contextual events. Normalize metric names and ensure numerical fields are properly typed.
Compute deltas and assess confidence
EXECUTECalculate week-over-week and optional month-over-month or quarter-over-quarter deltas. Identify noise or small-sample issues and assign LOW, MED, or HIGH confidence levels with explanations.
Rank and select top KPI movements
EXECUTEDetermine which KPIs exceed a justified movement threshold (e.g., 5–10%) and rank them by magnitude and business relevance. Select the top three changes for narrative generation.
Generate hypotheses and follow-up questions
EXECUTEFor each significant change, generate 3–5 clearly labeled hypotheses, ensuring they are framed as correlations rather than causation. Produce follow-up questions or next-step analyses to validate each hypothesis.
Draft executive-ready narrative and recommended actions
EXECUTECompile the selected KPI movements, hypotheses, validation steps, and action items into a concise narrative tailored to the specified audience and medium. Ensure the executive summary is under 90 words and includes at least one data-quality caveat.
Prompt Configurations
KPI Narrative Generation Prompt
You are an analytical, crisp, decision-oriented reporting assistant. Using the provided KPI deltas, confidence levels, targets, contextual events, audience type, and desired output format, produce: 1) Top 3 KPI changes above a reasonable threshold (justify threshold). 2) 3–5 hypotheses per change (clearly labeled as hypotheses; never claim causation). 3) Follow-up questions or next analyses to validate each hypothesis. 4) 1–3 recommended actions with role-based owners and urgency. 5) A 3–5 sentence executive summary under 90 words. Include at least one caveat about data quality or measurement limits. If data is small or noisy, explicitly mark confidence as LOW/MED/HIGH with reasons. Do not invent metrics; only use provided data.
Example Inputs/Outputs
Weekly KPI export: - Sessions: 102,000 (prev 95,000) - Conversion rate: 2.4% (prev 2.9%) - AOV: $54 (prev $50) Context: Minor pricing test launched mid-week. Audience: Exec, format: Slack update.
Top 3 Changes: 1. Conversion rate down 17% (threshold: 10% justified due to recent volatility). 2. Sessions up 7%. 3. AOV up 8%. Hypotheses (example for conversion rate): - Hypothesis 1: Pricing test increased user hesitation. - Hypothesis 2: Traffic mix shifted toward lower-intent channels. - Hypothesis 3: Mobile latency increased mid-week. Follow-up questions: Channel mix breakdown? Pricing test cohort impact? Device-level performance? Actions: Growth PM to run channel-level diagnostic (high urgency). Marketing Ops to verify tagging accuracy. Executive summary: Conversion rate dropped sharply while sessions and AOV increased. Early evidence suggests pricing test or traffic mix shifts may be contributing, but confidence is medium due to limited sample windows. Further analysis is required before decisions. Data-quality caveats apply due to recent tagging updates.