Margin & Exception Intelligence
Find the issues hidden by aggregate reporting.
A distributor grew revenue to $38.40M and lost 1.59 points of gross margin over two years. The monthly package showed both facts and explained neither. What changed that should not have changed—and what deserves management attention?
The result
A large invoice ledger reduced to a 20-item management tier and an explained review queue.
Rules first, then statistical departure and trend breaks, then attribution—each row carrying the baseline it departed from, by how much, an estimated size where supported, and the next action or question.
- Blended margin decline
- 1.59 pts0.79 of it from revenue mix
- Invoice lines analysed
- 278,169142 accounts, 36 months
- Issues for management
- 20promoted from 629 detections
- Seeded problems found
- 7 of 7all 7 surfaced in the full queue; 6 reached the management tier
Demonstration project using synthetic data designed to simulate a realistic operating company.
The finding
Revenue mix explains roughly half the margin decline.
A blended margin is an average of two businesses. A report that shows only the blend cannot tell a line getting worse from a shift in which line the revenue came from.
What the two lines earned
0.81 points, and not in one direction: Distribution gave up 1.30 points while Installed Service gained 0.60.
Which line the revenue came from
0.79 points. Installed Service, earning 41.8% against Distribution’s 22.1%, fell from 26% of revenue to 22%. Pricing changes alone would not reverse that half.
How it holds up
The queue's accuracy is measured, including where it is weak.
Problems were seeded into the ledger and recorded before the engine ran, so every accuracy figure is scored against a known answer.
Only one tier is presented as work
20 of 122 rows reach management, at 100% issue precision in sample. The other 102 are review signals, explained and not presented as something to act on.
Size estimates carry their error
Estimated impact came in at a median 2.08x the seeded magnitude. The estimates overlap across grains and are never summed; the full analysis shows each one against the true amount.
The evidence package ships a standard-library verifier. 95 of 95 checks pass on this build across the reconciliation and exception-intelligence outputs, including the headline metrics shown here. Run it yourself: python verify.py.
Start here
Want this done on your data?
The first conversation is about the question, the data that exists, and whether the work is worth doing at all.