Why project profit is wrong when scope and billing live in different systems

Take the common agency arrangement. The quote was raised in one tool, by a salesperson, as a set of line items with rates. The time was recorded in another, by the people who did the work, days later and approximately. The invoice was raised from a third, and the hour count behind it came from whoever was assembling the invoice that week. The cost per hour is a single blended number someone typed into a report.

None of these is a lie. Each answers a different question well. The failure is that they are being asked to answer a fifth question — what did this project actually make — and they have no shared referent. There is no key that ties a line on the quote to a block of hours to a line on an invoice, so nobody can ask the only question that matters: is the work I billed for the work I agreed to do?

Divergence is not rare and it is not usually dramatic. It enters at four specific moments, and once you know them you can see them in almost any agency's numbers.

Every figure below is invented, for illustration. Nothing here is drawn from a real engagement, a real agency, or real people. What it demonstrates is the shape of the error, which is what makes it useful.

An agency quotes a build project at a flat hourly rate, and the quote is raised properly as a structured record with items, rates, tax treatment, and terms. That is the agreed baseline: 328 hours at ₹2,400 an hour, a quoted value of ₹7,87,200. The agency's fully loaded internal cost rate is ₹1,450 an hour, which includes salary, overhead, tools, and the management time that is easy to leave out. At those two numbers the plan is 39.6% gross margin, and the project is priced correctly.

Now the three systems produce three different accounts of the same twelve weeks. The timesheet app holds 372 hours — 44 more than were agreed, and 31 of those 44 traceable to work the quote never described. The invoice register holds ₹7,47,840, because the agency invoiced the 328 agreed hours and applied a 5% goodwill discount to one part of the invoice without asking anyone who could approve it. The cost report multiplies 372 by ₹1,450 and calls the result the project cost.

Now the two margins. The first is the one an agency actually reports. It takes the invoice total, subtracts the cost the timesheet app implies, and announces 27.9% — ₹2,08,440 on ₹7,47,840 of billed revenue. That is the number that goes in the deck, and on its own terms it is entirely defensible arithmetic applied to two unrelated datasets. It is also wrong in two separate ways, and neither wrongness is visible in the result.

The second margin is built on the agreement rather than on the invoice. Agreed value was ₹7,87,200 for 328 hours. Actual cost was ₹5,39,400 for 372 hours. So the engagement genuinely earned ₹2,47,800, or 31.5% of what was agreed. Then the two silent losses get counted. The 44 hours of work that were never re-quoted and never invoiced removed ₹1,05,600 of value at the rate it should have been sold at. The 5% discount removed a further ₹39,360. Realised margin on the engagement is ₹1,02,840 — 15.9% of what actually came in, against a plan of 39.6%.

The most useful line in that example is the last one. Had the 44 hours been re-quoted as a new quote attached to the same project and accepted, and had the discount not been given away, the revenue would have been 372 hours at ₹2,400 — ₹8,92,800 — and the margin ₹3,53,400, which is 39.6%. Exactly the planned margin. The rate was never wrong. The estimating was not wrong. The delivery team did not run over on skill. The entire gap between 39.6% and 15.9% is a records failure, and every rupee of it is recoverable on the next project of the same kind.

The decomposition is the whole value. "37% margin" tells a team nothing about what to change next quarter. "Six per-cent of margin went to hours that were worked and never invoiced" tells them exactly which habit to install, and it is an installable habit rather than a talent problem. This is the same argument our piece on scope drift makes from the delivery side: the gap between agreed and invoiced is only useful if both ends of it are structured records you can put side by side.

The instinct when work was added is to go back and add it to the original quote so the total looks right. Do not. The original is what the client accepted and very often the copy they still hold, and editing it means your two versions differ — which is a harder conversation than the one you were avoiding. It also destroys the only evidence that you under-quoted that type of work, and a revised total tells the delivery team nothing about which part of the job is now being paid for.

Raise a new quote for the additional scope, with its own items, tax treatment, and terms. Get it accepted. Attach it to the same project so the project's economics show both pieces. That single move is what converts the 44 hours in the example above from a loss into revenue, because the addition becomes an agreement rather than an act of charity. The mechanics of the comparison are in our article on why agreed scope and invoiced scope diverge; the reason the second quote is a new record rather than an amendment is a data-integrity point, and it is worth being blunt about what NoxOrigin does and does not have.

This is not universal, and presenting it as universal would be its own kind of dishonesty. Two cases genuinely fall outside it.

Fixed-fee work with no time recording. If you price a ₹40,000 build at a flat fee, deliberately accept the risk, and do not attempt to measure effort at all, then there is no hour count to reconcile and the entire problem above dissolves. What you have chosen is a different business: you are selling an outcome and absorbing variance as a matter of policy. That is a legitimate way to work, and the argument in this article simply does not apply to it. Be clear with yourself about which one you are in, because the failure mode is drifting into fee-for-time while believing you are still selling a fixed outcome.

Very small jobs. A two-hour support visit or a single-day studio booking carries less total variance than the cost of measuring it. Instrumentation has a fixed overhead, and below some size the measurement costs more than the error it would catch. Below that threshold, judge the work by whether you would quote it the same way again, and skip the margin report entirely.

There is also a case where the argument cannot rescue you: a cost base you have never written down. If nobody knows what an hour of senior time costs all-in, no record structure will help, because there is nothing to compare the quote against. Fixing that is a piece of arithmetic, not a software decision, and it is a precondition for everything above.

To be explicit about the limits of this article: every figure in it is invented for illustration. We have no measured drift rate, no measured distribution of billed against quoted, no measured cost base, no data on how often agencies apply unapproved discounts, and no customer data whatsoever. Nothing here reports what agencies actually do or how much margin they lose. The example is arithmetic constructed to make a structural argument visible, not a measurement of a behaviour.

If a number about margin leakage is offered to you as a benchmark, ask which records produced it, over what period, how cost was defined, and whether the cost side was complete. A margin statistic built on an unrecorded cost base is a guess with a decimal point, and it will sound more authoritative than anything in this article.