Every proposal has one page where capability stops being the subject and money starts. Clients read it first. Consultants write it last. It is usually the least examined thing in the document.
I went back through an old proposal written for a European food and consumer goods manufacturer and found that page in it twice. Two versions, near duplicates, both still sitting in the file as it was left. Somebody meant to delete one and did not. What survived is a before and after of a commercial offer, which is not something you normally get to read, because the earlier draft is normally gone.
The offer had three rungs. A short session to align stakeholders and surface the candidate cases worth working on. One working use case, priced on a blended 1.3 FTE over roughly eight weeks at EUR 10,000 per week. The same use case bundled with a portable data platform that would connect to whatever data sources and warehouses the client provided, priced on 2.1 FTE over roughly eight weeks at EUR 12,500 per week, with the platform and its licensing supplied free for the initial trial.
Between the two versions, neither paid rung moves. Same rates, same team sizes, same duration, same footnote saying that a use case typically runs four to eight weeks and only where the data analytics environment is already established. The bundled rung gets renamed and its outcome line gets split in two, which is copywriting. The prices are untouched.
The one material edit is on the rung nobody was ever going to pay for.
The only thing that changed was the thing nobody pays for
Read the two versions side by side and the second one is not a better offer. It is the same offer, described by someone who had noticed something.
The economics are identical. Both versions print a weekly rate and a duration and never a total, which is its own small piece of commercial framing and survives the edit unchanged. The platform stays free during the trial in both. Going from 1.3 to 2.1 FTE, about sixty percent more people, moves the weekly price by a quarter, so the buyer taking the bigger rung is being paid to accept the infrastructure. That asymmetry sits in both drafts untouched, which tells me it was deliberate rather than an oversight.
Only the session moves. On the first pass it is described and given away. On the second pass it is described, given a number, and given away.
One honest caveat about the pair. The waiver condition names option numbers, and the numbering does not agree between the versions, though the options themselves are the same three. I read nothing into the numerals except that these are two drafts of one page and the renumbering was never chased through.
What a number does that a description cannot
A free thing with no price attached has no stated worth, and a buyer supplies one. Usually it is zero.
That is the mechanical problem. A session described as complimentary is read as a sales call, because that is the only unpriced meeting most executives take. A session described as EUR 5,000 and then waived is read as EUR 5,000 that somebody decided not to charge. Same meeting, same agenda, same preparation behind it, and the second framing survives contact with a procurement team while the first does not.
The second thing the number does is aimed inward. Naming a fee forces the firm to work out what the session costs it to run, and a qualifying session is the most preparation-heavy thing in an engagement: somebody has to read the annual report, work out where the data actually sits, arrive with candidate cases specific enough to argue about, and hold a room of executives who disagree with each other. An organisation that cannot cost that cannot tell whether it is doing too much of it.
The third thing is a claim about worth, and it is the one I care about. A price is what the work is worth to a buyer who has no reason to be generous. Free discovery is a subsidy paid out of the expectation of paid work downstream. That expectation is a market condition, not a fact about the work.
The free session is the input everything else runs on
The proposal is unusually blunt about how much rests on that session. Its stated output is stakeholders aligned and candidate cases identified, and the same document elsewhere sets a hard floor for whether a case is worth starting at all: a case should return about five times the effort it consumes, and a programme with no case clearing that bar is listed alongside absent senior sponsorship as a reason this kind of work dies. That bar can only be applied to candidates the session produced.
The delivery model leans on it just as hard. Fourteen activities are split out across the eight weeks as a percentage of effort between the firm and the client, and the pattern is that the firm takes the grind while the client keeps everything requiring authority. Crunching data into a usable state and transforming it to test hypotheses sit at ninety percent on the firm's side. Choosing the analytics platform sits at ten percent on the firm's side and ninety on the client's, and so does providing security clearance before any data is gathered.
Every one of those client-heavy rows is a commitment somebody senior has to make. The session is where that person is either persuaded or not. So the cheapest item in the document produces the input that the most expensive items depend on, and it was the only item in the document without a price.
- No figure
- The qualifying session as printed on the first version of the commercial page
- EUR 5,000
- The same session on the second version, one time fee, excluding travel and stay
- EUR 10,000/wk
- Use case only, 1.3 FTE, about eight weeks
- EUR 12,500/wk
- Use case plus platform, 2.1 FTE, about eight weeks, licensing free for the trial
Free is a subsidy, and subsidies are cyclical
I am writing this in a market that is giving almost everything away. Trial seats on copilots, credits against a first workload, a retrieval-augmented prototype on your own documents built at no charge in a fortnight, a vector store and an evaluation set thrown in to prove the point. Every one of those is the same move as the free session, at a larger scale and with the same reasoning behind it: the discovery work is worth less to us than the engagement it unlocks, so we will absorb it.
That reasoning is sound while it holds and it does not hold forever. When the pipeline thins, the unpriced item is the first one an organisation quietly stops doing properly, because nobody can defend the hours. It was never a line item, so it cannot be cut, which means it degrades instead. The preparation thins. The candidate cases arrive generic. And the paid work that follows is scoped on a weaker input, which is where the real cost shows up, a long way from the thing that was free.
Denominating it prevents that. Not by charging, necessarily, but by making the giveaway visible enough that someone has to approve it. A waiver that has to be granted is a decision with an owner. A courtesy has neither.
There is a live reason to fix this now rather than later. The obligations that follow from the European AI Act have reached political agreement, and the practical effect on early work is that discovery has to cover more ground: where the data came from, what its lineage is, what the evaluation set will look like, what documentation the eventual system will need. Discovery is getting more expensive to do properly at exactly the moment the market has trained everyone to expect it for nothing.
An unpriced giveaway is not generous, it is undeclared. Nobody defends a line item that was never a line item, and the first thing an organisation stops doing well is the thing it never learned to cost.
What I take from the second version
The edit between those two pages is small enough to miss and it is the most useful thing in the file. Somebody looked at a page that gave one thing away and decided the giveaway should be legible.
I would go further than that draft did. Price the session, print the waiver condition beside it, and price the capability review too rather than leaving its fee as something that can be discussed, because negotiable is just unpriced with a longer sentence in front of it. Then hold the number when the client asks what happens if they do not proceed. That answer is the whole point of having a number: it says the session had value independent of what it sold.
For a buyer the same reading works in reverse. When a vendor offers you something for nothing, ask what it costs them, and take the answer as their estimate of how much work understanding your business actually is. A vendor who cannot produce that figure has either not costed it or is not planning to do much.
None of this is unique to advisory work. It applies to the internal platform team giving away onboarding, to the data function absorbing every ad hoc extract, and to any RealAI Platform engagement where the first weeks of data readiness work get folded into a delivery budget and then disappear from the record. Cost it, publish it, waive it deliberately. Give the same thing away, and know what you gave.
Figures are as printed on the two versions of the commercial page in a proposal prepared for a European food and consumer goods manufacturer: three options, their rates, their team-size basis and their duration. This is a proposal, not a delivery record. The rates, the durations and the effort splits are what was offered, not what was achieved, and no client outcome is claimed here. The observation that the two versions differ only in the price of the thing being given away is ours.
“An unpriced giveaway is not generous, it is undeclared. Nobody defends a line item that was never a line item, and the first thing an organisation stops doing well is the thing it never learned to cost.”
Get in touch
Put RealAI’s applied-AI team on your hardest data problem.
We help enterprises move from pilots to production: sovereign models, governed data, and agents you can audit. Start with a value-first assessment.