If the only AI quotes you have ever seen came from a large consultancy, you were probably handed a single number with a logo above it. Seven figures, a slide of case studies, and no way to check what any of it pays for. So it is reasonable to assume every AI engagement is priced that way: an opaque lump you accept on trust or walk away from. The proposal we wrote to a major operator in Oman in August 2020 was built to be read the other way. It quoted EUR 85,945 for two phases over ten months, and almost all of that number breaks into lines you can name, price, and challenge one at a time. This post reads the proposal as a buyer would, from the top of the ledger down, to make one point: a subsurface-AI pilot is a knowable, budgetable object.
The number on the cover, and how it splits
Start with the headline. EUR 85,945 covers Phase 1 and Phase 2, sized at roughly four and six months, ten months in total. That is not a retainer and not a day-rate open cheque. It is a fixed fee against a scoped body of work, and the first thing the proposal does with it is carve off a fifth.
Twenty percent of the fee, EUR 17,189, is invoiced at the very start, before a model is trained, to clean and prepare the infrastructure the work will run on.
A mobilisation invoice is where an honest pilot and an opaque one part company. It says setup is real work with a real cost, it happens before delivery, and it has a stated purpose. A buyer can push back on the percentage or the timing, but they cannot pretend it does not exist. In a lump quote the setup cost is folded into the total and never named, so the client meets it only if the engagement stalls in week one.
Travel is a line, not an assumption
The next line most lump quotes bury is travel. Ours does not. The proposal budgets EUR 12,000, and it shows the arithmetic: six trips at EUR 2,000 each, covering roughly five weeks on site over the ten months. That is a number a procurement lead can sanity-check against their own flight and hotel costs in an afternoon.
The value of writing it as six trips at EUR 2,000 rather than a round EUR 12,000 is that the unit is exposed. If the client wants four trips instead of six, the line moves in a way both sides can see. Travel priced as a visible unit is negotiable; travel folded into an overhead multiplier is not.
The recurring lines: compute is rent, and rent is legible
Under the one-off lines sit the monthly infrastructure costs, and this is where an AI pilot differs most from a consulting engagement. The work needs machines, and the proposal rents them by the month in plain view:
- Cloud at EUR 3,000 a month for two dedicated EU servers, each specified down to the hardware: 1TB flash storage, 64GB RAM, and 8 to 24GB of GPU memory, with a burst arrangement behind them.
- Code and data repositories at EUR 300 a month, itemised further as GitHub at EUR 250 and Dropbox at EUR 50.
- Operating-system and ML tooling at EUR 700 a month for the IDEs and monitoring the team runs on.
That is EUR 4,000 a month of recurring infrastructure, every euro of it attached to a named service. A buyer who suspects the compute line is padding can price two EU servers of that spec themselves and check. The repos line is small enough to feel almost pedantic to itemise, and that is the point: a proposal that troubles to name a EUR 50 Dropbox subscription is not hiding the EUR 3,000 one next to it.
Because these lines recur, they accrue with time rather than landing all at once, and over ten months that EUR 4,000 a month becomes the largest single block of the fee. The instrument below reads the whole ledger this way. Drag the months lever and watch the itemised total climb as the recurring infrastructure accrues, with the orange mobilisation invoice at the head of the fee bar as the one line that most plainly says this is a scheduled invoice, not a lump.
What the named lines add up to
Add the fixed lines to ten months of recurring infrastructure and you have accounted for about four fifths of the fee. The mobilisation invoice, the travel budget, and the monthly cloud, repos, and tooling lines together sit near EUR 69,000 of the EUR 85,945. The remainder is the balance of the priced work, the human time and margin any engagement carries, and it is the one part of the ledger that reads as a professional-services number rather than a receipt.
That ratio is the whole argument. In a seven-figure lump quote, the fraction you can trace to a named, checkable line is close to zero; here it is roughly 80 percent. A buyer does not have to trust the total. They can rebuild most of it from the parts and see where the trust is actually being asked for, which is the smaller remainder, not the whole.
The clause that prices predictability
One line in the assumptions section is not a cost at all, and it may be the most valuable thing in the proposal for a buyer: a 96-hour client approval SLA. The team commits to a four-day turnaround on approvals, and in exchange asks the client to hold to the same. That single sentence prices predictability. It tells a procurement lead that the schedule has a defined cadence rather than drifting on whenever a stakeholder gets around to signing.
An SLA in a pilot proposal is a tell that the vendor has run pilots before and knows where they stall, which is almost never the modelling and almost always the waiting. Read alongside the phase structure, roughly four months for Phase 1 and six for Phase 2, it gives a buyer a timeline they can defend to their own management before a single invoice arrives.
Why the anatomy matters more than the total
None of these lines is large, and that is the reassuring part. A pilot priced this way is a decision a mid-sized operator can make inside a normal capital cycle, not an act of faith reserved for the majors. The reason to read it line by line is not to shave a few thousand euros off the compute; it is to see there is nothing to be afraid of in the number. Everything that can be named is named, and the part taken on trust is small and clearly bounded.
Pricing a pilot this way is the same discipline that later let the programme absorb a macro shock as a bounded draw rather than an emergency, which we cover in the contingency-clause piece and will not re-derive here. A legible budget is a negotiable budget, and a negotiable budget is one a buyer can actually sign.
If you have only ever been quoted a lump, the useful thing to take from a real proposal is the shape, not the specific euros. Ask any vendor to decompose their pilot price into mobilisation, travel, named monthly infrastructure, an approval cadence, and a stated remainder. A vendor who can is telling you the pilot is a knowable object. A vendor who cannot is telling you something too.
Limitations
The figures here come from a single proposal dated August 2020, to one operator in one market at one moment, and they should be read as the anatomy of that pilot rather than a going rate. Prices for cloud, tooling, and travel from 2020 do not transfer directly to a quote today, and the server specifications reflect what a subsurface computer-vision workload needed then. The four-fifths traceable share is arithmetic on the named lines against the total; the exact split between the recurring infrastructure and the human-time remainder depends on how you attribute the mobilisation invoice, which the proposal frames as infrastructure preparation. The month lever in the instrument is a reading aid over the sourced ten-month span, not a claim that costs were billed on that exact monthly cadence. Finally, this proposal scoped a geomechanics study; the programme that ran was technically different, so the commercial shape survives here but the specific deliverables it was attached to did not.
References
[1] The Contingency Clause You Hope Never to Use: Budgeting for Requirements That Will Change. Earthscan insights, 2025-11-13. https://earthscan.io/insights/contingency-clause-budgeting-for-changing-requirements
