Engagement Change Control That Actually Protects Margin
Firms do not lose margin to the change everybody argued about. They lose it to the fourteen nobody wrote down.
Ask a delivery lead where the margin went on a bad engagement and you rarely get a single answer. It went in small pieces: an extra workshop, a second round of review, a report reformatted for a board that was not in scope, a data extract that turned into a data cleanse. Each was too small to raise a change request over, and together they were the entire contingency.
The threshold problem
Every firm sets a threshold below which a change is absorbed rather than raised. The threshold is sensible and the arithmetic around it is not: a change is measured against the contract value, so a small one always looks small, and nothing measures them together.
The fix is cumulative. Track the sum of absorbed changes against the original baseline and put that number on the status report. A single one percent change is noise. Fifteen of them is the conversation somebody should have had in week three.
A change request has to move three things at once
An approved change moves the approved value, the end date and the baseline. Systems that move one without the others produce a plan that no longer matches the commercial position, and the discrepancy is found at closure by whoever is trying to raise the final invoice.
The signed contract value is never edited. It is the thing everything else is measured against, and a system that lets it be overwritten has destroyed its own reference point.
Who approves what, decided in advance
The approval path should be a property of the engagement rather than a decision taken per change, because a change under time pressure is exactly when somebody will find the shortest path to a yes. Value bands, a named approver per band, and the client-side counterpart where the contract requires one.
What Atlas does here
Change requests in Atlas move the approved value, the end date and the baseline together, never touch the signed contract value, and record cumulative drift against the original baseline so ten approved changes cannot quietly move an engagement without anybody seeing the total. The commercials surface holds the rate card, the fee model and the revenue recognition trigger beside them, so the change and its effect on the number are one record rather than two.