Applied project
Turn a large data set into a short list of material findings.
A business grew 9.1% over two years and earned 2.9% more gross profit. This review is the answer to why the second number was not the first one — and an example of the work product an analytical engagement produces.
The deliverable
One integrated view across commercial performance, financial position, and receivables control.
Eight findings and six exhibits, written for a reader who has to decide something and supported by an evidence package that recomputes the commercial analysis or traces imported figures to their verified source builds.
- Gross margin decline
- 1.594 pts28.03% to 26.43%
- Gross-profit counterfactual
- $612Kclosing revenue at the opening gross margin
- Invoice lines analysed
- 278,16936 months, 142 accounts
- Findings
- 82 critical, 6 sized
Sample deliverable using synthetic data.
The finding
Revenue mix explains roughly half the margin decline.
The decline splits almost evenly between changes in the two lines' own margins and a shift toward the lower-margin line. The first calls for pricing and cost-recovery analysis; the second calls for a different operating response.
What the two lines earned
0.806 points, and not in one direction: one line deteriorated while the other improved. Treating the decline as one problem would have combined two different mechanisms into one response.
Which line the revenue came from
0.788 points. Holding closing line margins fixed, returning the revenue mix to its opening split would mechanically improve gross profit by about $303K at current revenue. The higher-margin line fell from 26% of revenue to 22%. This is a counterfactual, not a forecast or a recoverable opportunity.
The two effects sum to the observed margin movement with no residual.
What it contains
Every finding carries supporting evidence and a proposed next action or question.
| Finding | Severity | Amount | Basis |
|---|---|---|---|
| Revenue grew; gross profit did not follow | Critical | $612K | gross-profit counterfactual |
| Revenue mix explains roughly half the margin decline | Critical | $303K | gross-profit counterfactual |
| The within-line margin decline is entirely in Distribution | Material | $389K | gross-profit equivalent |
| 2 accounts explain half the margin decline | Material | not sized | — |
| Growth is masking contraction in the largest accounts | Material | $1.17M | revenue |
| Service revenue fell on realised rate, not on hours | Material | $1.07M | revenue |
| Discount off list rose in every customer segment | Material | $133K | revenue |
| The year swings 1.66x from trough to peak | Watch | not sized | — |
The amounts are stated in different units and must not be ranked against one another or added — a revenue movement and a gross-profit counterfactual are different quantities. The same underlying economics can appear at more than one analytical grain, so the finding amounts are not additive. The exact, additive view is the decomposition in the report.
How it holds up
The method states its own limits.
The report exposes its arithmetic, provenance, and limitations alongside the conclusions.
Every decomposition ties
The margin bridge, the retention bridge and the price/volume/mix split each ship with a residual column, and the build refuses to publish a run where any residual is non-zero.
One source for prose and exhibits
Finding text and exhibits are generated from the same published artifacts, reducing the risk of copy drift between narrative and numbers.
Units are respected
This book bills in five units of measure, so price and volume are decomposed within each one. An average price across hours and boxes is not a price.
It states where it came from
The report names its own build, the upstream analysis builds it reads, and the SHA-256 of the transaction ledger used for the commercial analysis. A reader can check those identifiers against the source state in the evidence package.
What the data cannot answer is said
Cohort and logo-retention analysis are absent because the account roster is fixed: 142 accounts throughout, none acquired and none lost. A retention rate there would be 100% by construction.
This report is an integration layer. Its evidence package 53 of 53 checks pass on this build. Of the figures the report publishes, 31 are recomputed from the transaction ledger and 19 are checked against their referenced source builds; 1 is read rather than derived (management_issues). Run it yourself: python verify_executive_report.py.
Figures on this page are read from the published artifacts, generated 2026-09-14. Closing revenue $38.40M.
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.