Spatium Flow — case studies

Case studies

Proof you can re-run.

Every finding here is reached by rebuilding the numbers and reconciling the systems against one another, with the method shown in full. Nothing is asserted that can't be reproduced — run the code, reach the same number. This is the discipline of revenue integrity: not opinion, evidence.

B2B SaaS · revenue reconciliation

Flow Audit

Flow Audit reconciles a B2B SaaS revenue stack — CRM, contracts, billing, and commissions — against one another, and against a known ground-truth ledger. Because the correct answer is established in advance, every leak it surfaces is proven exact, not merely plausible: silent churn, duplicate revenue, commission overpayment, under-billing, and more.

See it run → The build on GitHub →

€4.6M

Leakage surfaced · demo run

100% / 100%

Precision & recall, calibrated

267

Leaks across six types

The pain

A deal closes in the CRM, a contract is signed, invoices go out, commission is paid. Four systems that should agree — and quietly don't. Revenue that's counted but never collected; commission paid above plan; the same revenue booked twice.

Why it persists

Each system is correct on its own terms. No single tool sees across all four, so the disagreements between them are invisible — and what no one can see, no one fixes.

The method

Reconcile the systems against each other, classify every unexplained variance by type, and price it. A calibration layer with planted, known leaks proves the detection catches everything and invents nothing.

The proof

Against the ground-truth ledger, detection reaches 100% precision and 100% recall — zero misses, zero false alarms — surfacing €4.6M across 267 leaks in six categories. The full pipeline is on GitHub.

Retail · currency consolidation

The $252K in the exchange rate

A retailer's order book: $54.7M of revenue, 46.6% of it booked in foreign currencies. Reconciling every order to its transaction-date exchange rate — against the common shortcut of one year-end rate — surfaces a $252,477 revenue misstatement hiding entirely in rate timing.

The engine on GitHub →

$252K

Revenue misstated · rate timing

46.6%

Revenue in foreign currency

$54.7M

Revenue reconciled

The pain

Almost half of revenue is earned in EUR, GBP, CAD, and AUD. Before it reaches a report it must be converted to one currency — and the rate chosen for that conversion quietly changes the number.

Why it persists

Currency conversion looks like plumbing, so it's treated like plumbing. Applying a single year-end rate to a whole year of transactions feels harmless; no system flags the gap, because every system did as it was told.

The method

A documented DuckDB engine aligns the raw files, rebuilds line-level revenue, joins each order to its transaction-date spot rate, and reconciles that against the year-end shortcut — pricing the difference.

Reproduce it

Run the published engine against the same data and you reach $252,477. The finding isn't asserted — it's falsifiable, every figure locked by tests, the full method published for anyone to check.

D2C e-commerce · margin, returns & attribution

The margin you keep isn't the margin you think

A direct-to-consumer fashion brand's full stack — catalog, orders, line items, ad spend, and the web funnel. Rebuilding margin from unit cost and reconciling the systems against one another surfaces a realised margin well below the full-price figure, crores tied up in returns, and an ad dashboard that disagrees with the order book.

The engine on GitHub →

52.1%

Realised margin · vs 62.4% full-price

₹2.69Cr

Revenue returned or RTO'd

36 / 50

Campaigns off vs the order book

The pain

The brand believes it runs on 55–65% margins, watches a healthy ROAS climb, and treats returns as just part of fashion. Three teams, three dashboards, each saying things are fine — while the cash reaching the bank is thinner than any one screen explains.

Why it persists

The loss isn't on any one screen — it's in the seams between them. The catalog's margin versus the discounts that actually shipped; the order header versus its line items; the ad platform versus the order book. No system is wrong; no one owns the space between them.

The method

Rebuild margin from selling price and unit cost, sum line items against the order header, reconcile reported ad conversions to actual orders, and separate revenue kept from revenue returned — then price each gap.

The finding

Realised margin is 52.1% against a 62.4% full-price margin, with ₹5.37 lakh lost selling below cost. ₹2.69 crore sits in returns and RTO — the controllable share driven by size and image mismatches. And 36 of 50 ad campaigns diverge from the order book by over 20%. The full engine and tests are on GitHub.

More case studies are added as new datasets are analyzed — each a real reconciliation, each reproducible. Bring a dataset →