Workflows · Finance systems · Tier 1 · first automations
The data-change audit log, summarized weekly
Thousands of rows a month of who changed what: date, form, action, user, old value, new value. The trail is captured and nobody reads it. Summarizing it is low risk and high credibility.
By hand: Weekly, one to two hours, usually skipped · Anyone with a planning or GL audit trail.
The card, as it runs by hand
- Input
- The audit export from the planning or GL system for the week.
- What I do to it
- Filter and threshold. Find large amounts landing in a 'no org' bucket, off-cycle edits, unusual users. Summarize.
- Output
- A one-page summary with the exceptions listed.
- Who checks
- The budget or finance-systems lead reviews the exceptions.
- How often
- Weekly.
- How long
- One to two hours by hand, when it happens at all.
- How long now
- Measured in the course after the first governed run, per card. Not estimated here.
As a governed workflow
- Stays in code
- The filters and thresholds.
- The AI does
- The summary and the anomaly flags in plain language.
- A person still
- The lead reads the exceptions. Nothing is changed by the agent.
- The log
- Rows read, thresholds applied, exceptions raised.
AI does the preparation. A named person approves it. Nothing posts to the books without that approval, and every run is recorded.
Where it goes wrong
Thresholds set once and never revisited. Put the threshold on the card and review it quarterly.
Why it lands in a room
Directly supports the governance story: the control already exists, now someone reads it.