The weekly KPI scorecard, done before the Monday coffee. It finds what's off target, asks the person who owns it, and waits for the Controller's sign-off.
Every company, person and number on this page is fictional.
| KPI | This week | Target |
|---|
A Controller's Monday, for every client
The same eight jobs, for each of five to seven clients, every week. Six of them are routine. Two need a person, so they stay with people.
Follow one week at Brightline Services
Week of Aug 3, 2026. Pick any step to see what actually happened.
Same week, other clients
Most weeks, most numbers are fine. Nobody gets an email unless something is off target.
Built so accountants can trust it
- The math is code, never AIEvery KPI is calculated by tested code with exact decimals.
- AI says what moved, never whyEvery sentence it writes is checked against the books. Reasons come from owners.
- Approved first, sent onceNothing reaches a client without the Controller, and never twice.
- Each client's books stay separateThe database itself keeps clients apart, and the AI runs locally, so client data stays in-house.
Checked against an answer key
Measured on synthetic data with planted problems whose answers are known. It shows the system does what it was designed to do, not how it would score on messy real books.
Why I built this
Fractional CFO firms run the same weekly KPI review for every client, and much of it is copy, paste and chase. I wanted to automate that routine with tools these firms already use, while keeping accountants in control of every judgment call.
I built it with Claude Code assistance. I designed the workflow, the safeguards, the tests and the evaluation, and checked the QuickBooks pull against QuickBooks' own report.