Service Delivery: Why an 80% Cheaper Build Is Not an 80% Cheaper Quote
A delivery lead gets the number for the first time. The connector that used to take 8 days took an afternoon.
The first instinct is to cut the quote by 80%. It is the obvious move and the one the client will expect, and it prices the work below what it costs you, because the 8 days were never 8 days of building.
What the 8 days actually contained
Nobody ever quoted 8 days of typing. They quoted 8 days and the typing was inside it, with everything else.
Understanding what the source really was, which takes longer than anyone admits and is where the surprises live. Building. Reviewing what was built, which on a good team is a senior engineer's afternoon and on a rushed one is nothing. Iterating on what the review found. Handing it over to whoever runs it, with enough documentation that they can.
Build was perhaps half of that. Generation collapses the half. It leaves the other half untouched, and the other half was always the expensive part, because it is done by the people you have fewest of.
So an 80% cut prices the deliverable as if the whole thing had been the build. You have quoted below your real cost, on work you never separately priced, and you will find out in the margin review.
The 2 things you are actually selling now
Strip the build out and what remains is 2 activities, both senior, both previously hidden inside a number that looked like engineering time.
Qualification
Knowing what the source is before you commit to a price. The client says the export is clean, and the client has never opened it. The sample they send is the cleanest data they have. The change indicator they describe does not exist. All of this was discovered in week 3 of the build, at your expense, and it was the largest single cause of overrun. It is the qualification phase of an estimate, and it was always the risky work. It is now the visible work.
Acceptance
Somebody has to sign that the deliverable is right, and that somebody has to know what a production-ready pipeline actually contains. When the code was written by a person, acceptance leaned on the author being able to explain it. When nobody wrote it, acceptance rests entirely on the reviewer's judgement and on the checks they run. That is a deliverable in its own right, and it was never on the invoice.
Both of these were under-priced for the same reason: they were bundled with junior build time, and the bundle was priced at a blended rate.
Why this is good news for a fixed price
Fixed price on data work has always been dangerous for one reason. The build had variance, and the variance came from things you could not see when you signed: which engineer got assigned, how much schedule pressure they were under, and what they found in the source once they opened it.
Generation removes the first 2. Given the same input and conventions, the artifact comes out identical, whoever pressed the button and whatever the deadline. What remains variable is the source itself.
And the source can be qualified before you commit. That is the shift. For the first time, the part of the work that carries the risk is the part you can do first, on a small fixed engagement, and the part you then price is the part with no variance left in it. Qualify, then commit against the qualified source rather than the described one. The commercial decision that follows from that is a separate conversation, but the conditions for it did not exist 2 years ago and they exist now.
The uncomfortable part for the daily rate
If you bill days and the build takes an afternoon, there are 2 ways to write the invoice and both are bad.
Bill the afternoon, and you have been honest and given away the value of everything you know. Bill the 8 days the afternoon replaced, and you have a quote the client will refuse the moment they understand what happened, which will be soon.
The way out is to stop billing the build as time at all. It is small and predictable, and your value lies elsewhere. The daily-rate model was already under pressure for this reason; generation makes the pressure specific and puts a number on it. Data projects are one front of the wider AI disruption in consulting.
What to change in your next quote
3 lines instead of a single one, and a worked example.
- A qualification line: Fixed scope, fixed price, sold first. It establishes the source, its variants and the tier of effort. The output is a short document the client can read, and it is the thing that makes the next line quotable.
- A build line: Now small, and fixed against the qualified source. This is the line that used to be the whole quote.
- An acceptance line: Review against a written bar, the checks that were run, and a named person who signs. Priced as a deliverable, because it is one.
Before: 10 days at a blended rate, with the risk hidden in the middle and discovered in week 3.
After: 2 days of qualification, half a day of build, 2 days of acceptance. 4.5 days instead of 10, and every one of them is senior. The revenue line is smaller. The margin line is not the same question, because the variance that used to eat it has moved to a phase you now do first and get paid for.
That last point is worth sitting with. The old quote made money when the source turned out clean and lost it when the source turned out otherwise. The new one is paid to find out.
What it means
Most of what a client paid for was never the code. It was knowing what the source really contained, and checking that the result was right. The code was the visible part, and it hid the other 2.
Now that the code is cheap, those 2 are what is left on the invoice. Price them.
PrettyWhale.ai is an Engineering AI for ingestion and integration code. Given a sample of the source, it returns the complete data pipeline with its tests, checks, schema, configuration and documentation. The effect on a delivery business is the one described above: the build stops being the variable, and qualification and acceptance become the work you price.
You have any questions on PrettyWhale.ai ?