A weekly KPI briefing that explains the numbers, not just shows them

Every Monday the CEO gets one page: the dozen numbers the leadership team agreed matter, each compared to last week, the trailing trend, and plan, with a plain explanation of what moved and the evidence behind it. Dashboards still exist for digging. The brief is for knowing where to dig.

01the problem and who owns it

The CEO has five dashboards, three of which disagree, and a weekly meeting where the first twenty minutes go to reconciling them. Department heads each present their own slice in their own format, and the number that quietly turned bad sits on a tab nobody opened.

The chief of staff or finance lead owns the weekly pack and rebuilds it by hand every Sunday night: exporting, pasting, writing commentary. The commentary is the valuable part and the first thing cut when the week is busy.

02what the AI does, step by step

1. Lock the metric definitionsLeadership agrees a short list of KPIs with one written definition, source, owner, and target each. The briefing reads from that list, so a metric cannot change meaning without someone editing it on purpose.
2. Pull and reconcile the numbersOn schedule, the workflow queries the warehouse or source systems, computes each metric, and runs sanity checks: row counts, freshness, and totals that should match across systems. Failed checks are reported instead of hidden.
3. Find what movedEach metric is compared with last week, the trailing average, the same week last year where relevant, and plan. Movements beyond a threshold the owner set are flagged for explanation.
4. Drill into the driversFor flagged metrics, the system breaks the change down by segment, channel, product, or region to find where it came from, using queries rather than guesses.
5. Write the briefThe model drafts one page: headline numbers, three to five things worth attention with the driver analysis behind each, and suggested questions for the leadership meeting. Every figure in the text is pulled from the computed data, not written freehand.
6. Deliver and capture questionsThe brief arrives by email or Slack with links to the underlying dashboard views. Replies and questions from the CEO are logged so the next week's brief can answer them.

03systems it connects to

• Data warehouse. Snowflake, BigQuery, Redshift, or Postgres, ideally with a modeling layer such as dbt.
• Source systems. CRM, billing such as Stripe, accounting such as QuickBooks or NetSuite, product analytics, and support tools.
• BI tools. Looker, Tableau, Power BI, or Metabase for the drill-down links.
• Delivery. Email, Slack, or Microsoft Teams.

04human checkpoints

• Metric list sign-off. The CEO and leadership team approve the KPI definitions and targets, and changes go through the chief of staff.
• First month review before send. For the first several weeks, the chief of staff reads each brief before it goes out and notes every correction.
• Owner annotation. Metric owners can add a sentence of context before the brief sends, and that context is labeled as theirs.

05what to measure

• On-time delivery. Whether the brief lands by the agreed time each week.
• Data check failures. How often sanity checks block or caveat a number, by source.
• Corrections after send. Errors found after delivery, which should trend toward none.
• Meeting time on reconciliation. A simple tally by the chief of staff of how much of the weekly meeting goes to arguing about numbers.

06risks and guardrails

• Hallucinated figures. A model writing commentary can invent a number. Generate the numeric text from the data directly and have the model reference values by name, then check every figure in the output against the source.
• Access beyond role. The briefing service sees compensation, margin, and pipeline data. Give it a read-only account limited to the modeled tables it needs, and control who receives the brief.
• False precision. Driver analysis finds correlations. Label explanations as likely drivers and keep the CEO's questions flowing back to the owners who know the real reason.

07build vs buy

BI platforms now offer scheduled reports and AI summaries, and if your data is already clean in one warehouse with one BI tool, those may cover a basic weekly email.

A custom internal tool is worth it when the numbers come from several systems that need reconciling, when the CEO wants the driver analysis rather than a chart, or when definitions must be enforced in one place.

08related playbooks

Browse every executive and leadership playbook or the full library.