Automate Your Weekly Business Review Without Inventing a Single Number
By Ehab Al Dissi ·
Watch the Weekly business review walkthrough
Read the video transcript
Finance says ninety-two thousand. The sales deck says eighty-eight. This should get resolved before the Monday meeting.
Define where each metric comes from, what period it covers, and who owns it.
When the sources disagree, show both numbers and their records. Give finance a decision to resolve.
In our fictional example, a delivery percentage had no source export. The workflow held it for operations to verify.
Pick the one metric people argue about most. Automate the reconciliation and see what your team can decide faster next Monday.
Finance opens Monday’s meeting with $92,000 in recognized revenue. The sales deck says $88,000. Operations has a delivery percentage and no export anyone can check. The meeting spends its first half debating the numbers. An AI workflow can do that reconciliation before the meeting and bring the unresolved decisions to the people who own them.
How to automate a weekly business review
Start with three metrics and one business unit. Assign a source of record, a definition, and an owner to each metric. Pull those records on a schedule, compare them against the slides, and flag missing or incompatible evidence. An AI agent can summarize the change and draft questions for the owner. Calculations and the final decision about what to include should follow explicit rules.
- Write a metric contract with its definition, currency or unit, time zone, period, source, freshness window, and accountable person. Bookings, recognized revenue, and cash collected each need their own contract.
- Pull the closed-period records and keep their snapshot IDs. If a source is late, carry its last verified value with the date or show it as unavailable. Keep late data visible to the owner.
- Compare each slide with the source using the same filters and period. If there is a discrepancy, show both numbers and their sources. Ask the owner to resolve it before promoting either as the final figure.
- Produce a brief with one row per decision. Include the change, supporting records, an owner, the decision required, and its deadline. Link the previous decision if one already exists.
- Have a designated reviewer release the brief. Record what changed after review and carry open decisions into the next week.
Reproduce the Monday briefing
Run python replay.py review --output review.csv. The script uses three fictional metrics and keeps unsupported figures out of the verified brief.
| Metric | Evidence | Decision |
|---|---|---|
| Bookings | CRM and slides agree on $124,000 | Include with its source. |
| Recognized revenue | Closed ledger shows $92,000. Slide shows $88,000. | Finance checks the reporting period and definition. |
| On-time delivery | Slide says 96%. Source export is missing. | Hold until Operations supplies the export. |
The fixture tests a matched value, a disputed value, and a missing source. In a real deployment, add stale exports, changes in currency, and a revised prior period. Have the model explain the reason for a mismatch only when it has evidence for that explanation.
Decide if the pilot worked
Record time spent reconciling numbers before and during the meeting, the share of metrics with inspectable source records, and the number of decisions that leave the meeting without owners. Review which discrepancies the workflow surfaced early enough to resolve. Measure whether work moved, using the time from the first detected exception to an approved decision.
For the next meeting, pick the one metric people argue about most. Ask both teams to show the source, period, and definition. That gives you the first case to automate and a real outcome to check the following Monday.
Download the Python replay · Download the example CSV