Margin Architecture
How a signal diagnostic actually runs.
Margin Architecture does not begin with a dashboard request. It begins with a decision that feels exposed and the number currently carrying it. The work from there is sequential, and each step has to hold before the next one is worth doing.
Start with the decision
Identify what is about to be decided — a renewal, a price, an investment, an exit — and which report, metric, or model is carrying that decision today.
Test the signal
Trace definitions, source logic, and cost attribution until the point where the number stops being true. Uniformity, round assumptions, and unallocated pools are where it usually breaks.
Reattribute the cost
Put each cost against the contract, product, or customer that actually caused it. No new dollars enter the picture. The same dollars are placed honestly.
Rebuild and hand over
Express the corrected signal in a reporting layer the internal team owns and can maintain, with definitions documented so the signal does not drift back.
Built to be taken apart.
Client margin files are confidential, so these are not client files. Each example is a complete synthetic company — invented name, invented figures, real method — built so the reasoning can be inspected end to end rather than summarised. The interactive views run on the same data the narrative describes.
Why synthetic rather than anonymised. Anonymising a real engagement still discloses a client’s cost structure, and it asks a reader to trust figures they cannot check. A synthetic example discloses nothing, and every number in it can be reconciled on the page.
Start with the report or number leadership no longer fully trusts.
If a margin, profitability, pricing, cost, or revenue-unit signal is important enough to act on but not trusted enough to act on confidently, that is the conversation.
The first conversation is not a scoping call for a dashboard. It is a decision-signal conversation: what decision is exposed, what number is being trusted, and why confidence in that number is uncertain.