Sort the engagement archive by date and the oldest thing in it is not a proposal, a term sheet, or an email. It is a folder of eight raw FMI strip renders, and the folder is dated March 2020. The signed service agreement that started the paid programme is dated November 2021. The kickoff is December 2021. So the first picture anyone drew of this operator's borehole predates the contract that paid for the work by roughly 21 months, and that gap is not an accident of filing. It is the whole story of how the engagement was won.
The oldest file in the archive is a picture of the client's rock
The March-2020 folder holds strip renders labelled plainly, FMI1 through FMI8. Open the first one and it is an eight-pad Formation Micro-Imager image with the depth axis running from about 8,400 to 9,900. The number that gives it away is the unit: the axis is in feet, where every later plot in the programme is metric. It is an orange colormap, bedding banding visible across the pads, the characteristic wedge gaps between pads where the tool measures nothing. It is, in other words, a competent render of the operator's own downhole image, produced from the operator's own DLIS files, a year and a half before there was a contract to produce it under.
Nobody was paid to make those renders. There was no statement of work, no scope schedule, no milestone tranche attached to them. They exist because the vendor took a DLIS drop, wrote the reader that maps the tool's sentinel values into a clean array, and rendered the pad images to see whether the carbonate had the structure worth building a detector for. That work kept going through Covid: parsing files, rendering strips, learning the operator's data quirks long before any commercial conversation had a number attached.
The instrument above is the argument in one frame. Two teal anchors, March 2020 and December 2021, and between them an orange band that is the only thing in the picture doing any arguing: the 21-month pre-contract gap. Drag the pale marker to wherever you think the first serious proposal meeting happened. It does not matter where you put it. The render already existed. There is no month you can choose inside that gap where the vendor would have walked into the room with a generic demo, because the demo was already built from the client's own log. That is the mechanism. When the proposal meeting finally happened, the vendor did not pitch a capability. They showed the operator a picture of the operator's rock and said, we already did this with your data.
Why the demo being their own data changes the sale
Enterprise AI sales in subsurface has a credibility problem that no slide deck solves. Every vendor claims their model generalises. Every vendor shows results on a public benchmark, an open seismic block, somebody else's well. The operator's honest internal response is always the same: our carbonate is not your benchmark, our tools write values yours does not, our data will break your pipeline in ways your demo never showed. The gap between a polished demo and a working system on the client's actual files is exactly where these programmes die.
Rendering the client's own DLIS before the proposal collapses that gap. The FMI strip in the March-2020 folder is not a claim that the pipeline will work on their data. It is proof that the ingestion already ran on their data, that the sentinel mapping already handled their tool's encoding, that the pad geometry and the depth units and the file structure were already understood well enough to draw a clean image. The operator does not have to trust that the vendor can handle the specifics. The specifics are on the screen, in feet, in their formation.
This is a different activity from marketing. Marketing produces assets about the vendor. This produced an asset about the client, at the vendor's own cost, before any commitment existed in either direction. The renders were speculative, and most would have been thrown away if the operator had never engaged. That is what makes them persuasive: they are the sunk cost that proves the vendor was serious before there was a reason to be.
The gap was long, and that was the point
Twenty-one months is not a quick proof of concept. It is closer to a courtship than a pitch. The vendor was not rendering one strip and firing off a proposal the next week. The March-2020 date is the earliest surviving artifact, which means the technical work of understanding this operator's data ran for the better part of two years as a slow, unpaid parallel track to whatever commercial conversation was happening in the foreground.
That length matters because it is what a real relationship looks like from the data side. Subsurface AI is not a transaction you close on a first call. The operator has to develop confidence that the vendor can be trusted with confidential well data, that the models are grounded in the physics of their reservoir, that the team will still be there in two years. None of that is built in a sales cycle. It is built by showing up, repeatedly, with work that is already about them. The rendered FMI strips are the receipts of that showing up: each one is a concrete demonstration that the vendor was spending its own effort on the operator's problem while the contract was still hypothetical.
What the contract looked like when it finally arrived
The commercial terms, when they landed, were shaped by that long runway rather than despite it. The signed service agreement carried a 2-year term with 60-day either-party termination notice, and it was written as an effort obligation rather than a result obligation. Read against the pre-sales history, those terms make sense. A 2-year term is not what you sign with a vendor you just met. The 60-day notice is a normal off-ramp, but the length of the commitment reflects a relationship that had already been running informally for far longer than the contract would.
The effort-not-result framing is the other tell. A vendor selling on a polished demo gets pushed toward result guarantees, because the demo implies the outcome is known. A vendor whose demo is the client's own raw data, pad gaps included, has already been honest about the difficulty. The contract could be written as a research effort because the pre-sales work had never oversold. The March-2020 renders showed the operator exactly what the raw material looked like, so the agreement did not have to pretend the answer was already in hand.
The transferable lesson
The reusable asset here is not a rendering script. It is a posture toward enterprise AI business development: do the technical work on the prospect's real data before there is a contract, at your own cost, and let the first demo be their data rather than yours. The economics are uncomfortable, because most of that speculative work is written off. But the pieces that land do something a proposal cannot. They move the conversation from can you do this to you already did this with our files, which is the only version of the question an operator actually believes.
For a subsurface AI vendor this is concrete. Ask for a sample DLIS early, before any commercial ask. Build the reader, map the sentinels, render the image, run the QC. If the carbonate is worth it, you now have a demo made of the client's own borehole and a head start on the ingestion the paid programme will need anyway. If it is not, you learned that cheaply. The March-2020 folder is what that discipline looks like when it works: a picture of the client's rock, dated 21 months before anyone signed anything, that turned out to be the most persuasive thing in the room.
Limitations
This is a single engagement, and the dates are read from folder timestamps and document dates in one archive, not from a reconstructed sales log. The March-2020 render is the earliest surviving artifact, which bounds the pre-contract technical work at roughly 21 months but does not prove the work started exactly then, nor that the renders alone won the deal; commercial relationships turn on people and timing this account does not capture. The draggable meeting marker in the instrument is a reading aid, not a recorded meeting date. The contract terms cited, a 2-year term and 60-day notice on an effort obligation, are from a draft of the signed agreement in the archive. Treat the argument as one well-documented instance of pre-sales technical courtship paying off, not as evidence that unpaid speculative work is the right business-development strategy in general.
References
-
Engagement archive, earliest data folder: eight raw FMI strip renders folder-dated March 2020, one covering roughly 8,400 to 9,900 feet on a depth axis in feet, rendered from the operator's DLIS files, 21 months before the December 2021 contract kickoff. Internal materials, withheld under operator confidentiality.
-
Signed service agreement (draft in archive): 2-year term, 60-day either-party termination notice, structured as an effort obligation rather than a result obligation. Internal materials, withheld under operator confidentiality.
