Every first analytics engagement I am asked to shape gets argued on technical ground. Which data is clean enough, which method, which platform, how long the pipeline takes to stand up. Then the work lands, the answer is right, and nothing happens afterwards. I have watched that ending often enough to go back through my own paperwork looking for where the mistake is actually made.
The clearest specimen I hold is a proposal I co-wrote and signed for a European food and consumer goods manufacturer that wanted its first serious analytics capability. It is a sales document. Nothing in it was delivered, and every week count, effort percentage and price in it is an offer. That is what makes it useful, because a proposal records what people believed before any work happened, without the tidying a closing report does.
The belief in this one is about order, and it sits in the cover letter rather than in the method pages.
The order is the argument
The letter asks the company to explore its challenges, then narrow to one to three areas where a first proof case with concrete results can be delivered in a relatively short time frame, with the choice made in a properly prepared workshop rather than by whoever argues hardest. Then the ordering is made explicit: building on the results delivered and the insights gained during that first case, explore the roadmap that would help build the right platform and the right skills.
Look at where the roadmap sits. After. Not later in the document, but after evidence exists. The same goes for the skills plan, the architecture and the hiring case, all of which a company is most eager to design on the first morning, when it knows least and has proved nothing.
Start small while thinking big is the phrase in the letter. Read as engineering advice it is unremarkable and slightly glib. Read as a claim about how a large organisation grants permission, it carries the whole method. A platform roadmap is an argument about money, headcount and whose systems get touched. Nobody wins that argument on a whiteboard. You win it holding something that already worked, in front of people who watched it work.
Enthusiasm was the stated purpose, in writing
Sales documents rarely admit what an artefact is for. This one does, before any method appears: concrete real-life cases are an instrument to show what is possible and to create enthusiasm with stakeholders. That sentence is why the sequence runs the way it does.
You can see the same logic in what the letter reaches for while the company has no result of its own. It leads with other people's outcomes, named as prior work with other clients and stated as such, including an online marketing reallocation that lifted conversion by 20 percent. Borrowed proof, used to buy attention until local proof exists. That is what the first delivered case is meant to replace: not a benchmark and not a technical validation, but a story the company can tell about itself.
Accept that and the acceptance test moves. A first case is not finished when the numbers are right. It is finished when somebody who did not build it repeats the result to somebody else.
Sponsorship sits in the risk register
The delivery cycle in the proposal carries a short list of challenges under each stage. At the front, before any data has been touched, two are named: that no case worth doing gets identified, and that senior sponsorship is absent.
That placement is the tell. Sponsorship has been written into the risk register rather than the preamble. The engagement does not assume a sponsor and then execute; it assumes sponsorship may be thin and treats producing it as work. Which is also why the window is short and the scope is one to three areas rather than a programme. A pilot that runs long enough for its sponsor to change jobs has failed whatever came out of it.
I would put it more bluntly than the proposal does. A company without a convert cannot act on a result, because acting requires a budget decision, and budget decisions get made by people repeating a claim to other people. If the engagement creates nobody willing to make that claim when the delivery team is out of the room, it has produced an artefact and no capacity to use it. It is why RealAI Consult now writes the intended convert into the scope of a first engagement beside the dataset, since the sponsor is the part of the design that fails without making a noise.
The decision the pilot still has to borrow
The seam where proof-first sequencing strains is visible in the proposal, though it goes uncommented.
The roadmap is deferred until after the result. But the pilot has to run somewhere, so the effort plan puts agreeing which analytics platform to use in the first week, at 90 percent client effort. The platform argument has been postponed as a strategic question and reintroduced as a logistical one, landing on the client at almost full weight in the week it has the least evidence and the least appetite for a fight.
Two ways out sit in the priced options. Run the proof case on something already approved for other purposes, so no new argument is required. Or take an environment from outside for the duration, which the second option does by bundling a supplier-provided analytics environment with its licensing, at 2.1 full-time equivalents and 15,000 euros a week against 1.3 and 7,300 euros for the single case, both quoted at roughly six to eight weeks. Prices offered, not costs incurred.
What you must not do is let the proof case become the occasion for the platform decision. The moment it does, the case carries an argument it has not earned the standing to win, and the instrument gets consumed settling procurement instead of creating a sponsor.
What carries a firm price
The options page says the same thing in commercial language, and it is the part I find most honest in hindsight. The workshop that selects the one to three areas is offered at no cost if the company proceeds to a paid option. The fee for the capability review is left open, marked as discussable if it is wanted at all. The delivered case is the only item carrying a firm weekly rate.
Read that as a belief about which artefact converts anyone. Assessments and roadmaps produce documents, and nobody has ever changed a budget because a document arrived. So everything ending in a document was given away or left unpriced, and the only thing ending in a repeatable result was charged for with a straight face. The pricing is not generosity. It is a ranking of instruments by their power to move a decision.
The deliverable is a person
The risks named at the far end of the cycle finish the argument. They are that the company lacks the discipline to run continuous benefit reporting, and lacks the habit of improving its own fact base. Neither is a technical condition. Both describe behaviour in the weeks after the delivery team leaves.
So the real deliverable of a first proof case is not the pipeline, the notebook or the dashboard. Those decay. It is a named person inside the company whose own scorecard moved, who understands the result well enough to defend it under challenge, and who is still in post when the roadmap conversation happens.
The acceptance test for a first case is not whether the result is correct. It is whether one person who did not build it will say it out loud in a room you are not in.
- 1 to 3
- Areas the letter asks the company to narrow to
- 4 to 6 wks
- Analytical middle of the offered band
- 7,300 euros/wk
- Single case at 1.3 FTE, offered
- 15,000 euros/wk
- Two cases plus a supplied environment at 2.1 FTE, offered
What I would write differently now
The method has aged well and I would still argue for it. The document around it is what I would change, in four places.
Name the intended convert on the first page, as a person rather than a function, beside the number on their scorecard the case is supposed to move. Define done as a repeat rather than a demonstration, so the closing week is scheduled around somebody else presenting the finding. Put the roadmap conversation in the plan as a dated event contingent on the result, so deferring it reads as sequencing rather than avoidance. And carry a second candidate case in the same document, unpriced, so the first result has somewhere to point.
The last matters more than it looks. Enthusiasm has a short half-life, and a convert with nothing to ask for next converts back.
This is more urgent now than when the proposal was written, and not because the politics got harder. The technical half keeps getting cheaper. Machine learning pipelines are assembly work in most stacks, deployment is close to routine, feature stores and lineage tooling have turned questions that used to consume a quarter into configuration, and the careful language model pilots crossing my desk stand up in days rather than months. The political half has not moved at all. Most first engagements are still budgeted, staffed and argued as though the ratio between the two were unchanged.
Drawn from a proposal for data innovation support at a European food and consumer goods manufacturer, which I co-wrote and signed: its cover letter, delivery cycle, example effort plan and priced options. It is a pre-sales document, so the durations, effort splits and prices are commercial terms offered rather than a record of delivered results, and the percentage outcomes it cites were achieved with other, unnamed clients. Reading the first delivered case as a political instrument, and the sequencing of the roadmap behind it as deliberate, is mine.
“The acceptance test for a first case is not whether the result is correct. It is whether one person who did not build it will say it out loud in a room you are not in.”
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.