Consulting engagements report progress every week and money every quarter. The distance between those two cadences is where the closing argument comes from, and the argument is always the same shape. The client believes it bought more than it received. The supplier believes it delivered more than it billed. Neither side has a shared record fine-grained enough to settle it, so the settlement is a negotiation about memory.
A large European public-sector organisation was preparing to put its outsourced IT service desk back out to tender, and commissioned a sourcing study to produce the specification. The study ran five working weeks: seven phases across twenty-five working days, four formal progress reviews at exactly seven-day intervals, and a final presentation sitting on the last working day of the plan with no buffer column behind it. The artefact worth publishing from it is not the tender specification. It is the reporting page.
The challenge
Twenty-five working days leaves no room for a phase to drift, and it leaves even less room for a conversation about money that arrives after the work is finished. The plan made that exposure worse before it made it better. Of the seven phases, six named a key output. The remaining one named none. Planning and data gathering carried the literal entry "-NA-" in the key outputs column, scheduled as work in its own right, with the stated objective of gathering the inputs needed to minimise project risk, and with nothing at the end of it that anyone could hold.
That is an honest way to plan and an awkward way to bill. A sponsor watching a week of paid effort produce no deliverable has two ways to interpret it. Either the phase is de-risking everything that follows, which is what it was for, or the engagement is drifting. Which of those the sponsor believes depends entirely on whether the spend is visible at the same moment as the work, or only afterwards.
The kick-off deck offered a second clue about how this team thought. Its agenda promised assumptions and risks. What it delivered was fifteen bulleted assumptions and no risk register at all. Assumptions covered where the team would sit, what converted into a change request, how many business users would be consulted, what the client itself had to supply, and what happened if a team member had to be replaced. Every line was a falsifiable statement someone could object to in the room. But a risk register presented once at a kick-off is a document, and it ages out almost immediately.
The approach
Both problems were solved by the same page, and the page was not new. The team carried over a reporting template it had already used with this client on earlier work, on the stated grounds that the client had been satisfied with it. Adopting an artefact the client has already accepted removes an entire week-one negotiation and buys credibility before any finding exists.
The report covered a seven-day period and ran to two pages. It had three titled sections. The first summarised progress against the plan, aspect by aspect, as current status plus next steps. The second was headed for actions, decisions, risks and issues, and its structure did the work: every issue row and every risk row was followed immediately by a paired agreed action. The skeleton carried four such pairs as placeholders, three issues and one risk. You could not record a risk in this instrument without recording what had been agreed about it.
The third section is the one that matters commercially. It was a resource table with five columns: name, role, days worked, daily rate in euro excluding expenses, and total charges for that period. The days worked column was itself split three ways, into two client locations and an offsite bucket. Below the table sat a total for the period, and below that a standing line giving the budget remaining after that week, qualified as sitting before any early-payment discount. The page closed with two signature blocks, one for each party.
The three location buckets are not administrative detail. The assumptions page had committed the team to visiting one site per week to hold travel expenses down, and to working from its own offices only with the client's approval. Those are exactly three states, and the timesheet had exactly three columns. A commitment with a field in the reporting artefact is a commitment. A commitment without one is a sentence in a slide.
The same logic now applies to every AI programme being scoped in Europe. Retrieval-augmented pilots, evaluation sets, vector stores and the data readiness work underneath them are consumption-priced and largely invisible to the sponsor paying for them. Weeks of lineage and pipeline work produce nothing anyone can hold, exactly like that "-NA-" phase, and the spend accrues the whole time. Our Consult engagements now set the reporting instrument before the technical plan, because the instrument decides whether the unglamorous phases survive their first budget review.
The outcome
Be clear about what the record supports. What we hold is the kick-off document, which means what is evidenced is a design, a supplier claim of prior acceptance by the same client, and a committed cadence, not a measured saving. Every day count, daily rate, charge total and budget figure in that template is a blank placeholder, so no number from this engagement's finances is published here or anywhere else. The structure is the product.
Two properties of that structure survive translation to any engagement. The first is that the people who wrote the charges were not the person who signed for them. The two consultants delivering the work prepared the progress, the risks and the resource table. The countersignature on the supplier side sat with the account-level team member who had produced none of it, and whose role label in the same table was quality assurance. That is a segregation-of-duties control hiding inside a word processor file. The supplier side of the signature block even shipped pre-filled and dated, with the client side left blank, which tells you precisely who the page was built to convince.
The second is that the confidentiality classification on every page was generated from a document property rather than typed by whoever happened to be writing that week. Classification that depends on an author remembering fails. Classification that is a template field does not. Anyone shipping model-drafted deliverables this year is about to relearn that, because a generated document inherits no such field unless the pipeline puts one there.
The wider point holds regardless of what the work is. Publishing consumption every seven days removes the end-of-engagement argument, because there is no moment at which the client discovers the number. A client that watches it fall the whole time, and signs beside each step of the fall, has no number left to be surprised by. The European AI rules agreed by the institutions, which are not yet in application, will push organisations toward exactly this habit: contemporaneous records of what was done, by whom, at what cost, rather than a reconstruction assembled once the money is gone. A weekly countersigned page is the cheapest possible start on that discipline, and it needed no regulation to justify it.