Claude Prompts for Data & Analytics

Turn raw A/B test output into a stakeholder-ready writeup with effect size, confidence, segment behavior, and a clear ship/kill/iterate recommendation.

Produce a structured investigation plan for a sudden metric anomaly: triage steps, candidate causes, ownership, and the order of cuts to run.

Frame a churn prediction problem clearly: target definition, prediction horizon, label edge cases, and the tradeoffs between competing formulations.

Turn a raw cohort retention table into a clear narrative that explains what's happening across cohorts and why it matters.

Produce a deprecation comms plan for a legacy dashboard — audience map, replacement, timeline, and the messages that prevent angry Slack threads.

Produce a dashboard design spec — audience, decisions, metrics, layout, and what to leave out — before any tile is built.

Produce a tiered data quality test plan for a new ingest or pipeline — covering source freshness, schema, volume, and business-logic checks without over-testing.

Triage an incoming ad-hoc data request: clarify what's being asked, scope, prioritize, and respond — turning a vague Slack ping into a tractable piece of work or a redirected ask.

Produce a slide-by-slide outline for a data-storytelling deck — built around a single argument, with charts that earn their place and a clear ask.

Take a dbt model's SQL and generate the schema.yml description, column docs, and recommended tests an analytics engineer would actually approve.

Produce a one-page exec narrative that sits on top of a dashboard data dump — the story leadership reads instead of squinting at the dashboard.

Generate a structured set of feature engineering candidates for an ML problem, organized by feature family with leakage and stability flags.

Take a conversion funnel with a notable dropoff and produce a ranked set of hypotheses with a concrete investigation plan for each.

Produce a metric definition spec for a metrics layer (LookML, Cube, dbt Semantic Layer, MetricFlow) that closes the ambiguity that causes downstream disagreement.

Produce a model card that documents what an ML model does, how it was built, where it works, and where it does not — for stakeholders, auditors, and the team that will inherit it.

Produce a structured ML experiment plan: problem framing, hypotheses, baselines, success criteria, and the path to a deployable model — before writing any training code.

Explain what a propensity model does, what its score means, and how to use it — for a non-technical executive who will sponsor or rely on the output.

Produce a structured retention analysis writeup that goes beyond the cohort table — segments by behavior, isolates drivers, and recommends concrete retention bets.

Produce concise, accurate glossary entries for metrics and terms that appear across dashboards — written for self-serve users who want to know what a number means before they cite it.

Produce a naming convention spec for tables, columns, metrics, and dimensions in a semantic layer or dbt project — covering prefixes, casing, and the awkward edge cases.

Produce a structured review of a SQL query — correctness, readability, performance, and dbt-fitness — with concrete refactor suggestions.

Turn a weekly metrics CSV into a tight written commentary that explains what moved, why, and what to do about it.