The Undefined Rung

A habit worth keeping. When a framework arrives with its boxes already drawn, count the boxes, then count the ones the same document bothers to explain. The gap between those two numbers is usually where the argument is hiding.

I went back this month to a sourcing review I worked on for a large European public-sector organisation that was preparing to re-tender its IT service desk. One module of that review was a scan of how the provider market prices this kind of work. It carries three exhibits. A ladder of commercial models arranged against the value at stake. A comparison grid marking a dozen providers against those models. Then a run of pages summarising what each provider said about its own charging, in its own words.

Read the ladder on its own and it is a competent page. Read all three exhibits against each other and one rung falls through the floor.

Eight labels, seven definitions

The eight labels run from staff augmentation and time and materials at the low-value end, through fixed price and unit or consumption pricing, up through gain share and outcome-based pricing to revenue and profit sharing at the top. The eighth label is collaboration.

Below the ladder sit the definition blocks. There are seven of them. Each defined rung gets a short paragraph saying what the provider is actually paid for and, in most cases, what has to be true before the mechanic works at all. Collaboration gets nothing.

A smaller thing, about how the page was assembled. The low rung is titled staff augmentation on the ladder, the definition block beneath it is titled staffing, and the grid column on the next page is titled staffing too. Nowhere does the document say those are the same thing. Even inside the seven, the labelling drifts. The eighth is not drift. It is absence.

What the seven had that the eighth did not

Take the definitions one at a time and a pattern shows up immediately. Each names the act that causes money to move, and four of the seven also name the condition the mechanic depends on.

Revenue and profit sharing pays when the client's revenue rises and the rise can be traced back to the provider, and the block concedes in the same breath that the tracing is the hard part. Unit pricing pays per ticket, per user or per device, and holds up only where demand can be forecast closely enough. Fixed price pays an agreed sum against an agreed scope, on the condition that the scope, the delivery method and the requirements all stay still. Outcome-based pricing pays on a pre-agreed result of high business impact, and the two worked examples given are a fall in incident volume and a sustained rise in first-contact resolution above target. Gain share pays a share of savings the provider finds outside what the contract already covers. Staffing pays per person supplied, for the case where the skills are not in the building. Time and materials pays against what was asked for, how deep it goes, and how much of it there is.

Every one of them tells a buyer what generates an invoice line, and four go on to tell them what would have to be true for that line to survive an audit. Either half separates a pricing model from a sentiment. Collaboration carries neither, because it does not describe a mechanic at all. It describes a mood.

The column that was never cut

The next exhibit is a grid. Twelve providers down the side, seven models across the top. Seven, not eight. The undefined rung was dropped between one page and the next, and nothing acknowledges that the taxonomy changed.

The grid is worth reading closely. Twelve providers against seven models gives eighty-four cells, and thirty-seven of them carry a mark. That is a fill rate of forty-four percent, which means the average provider is marked against a little over three of the seven models rather than most of them. The market clusters hard in the middle of the ladder: eleven of the twelve are marked for unit or consumption pricing, eight for fixed price, eight for outcome-based. At the ends it thins out fast. Revenue sharing is marked for two providers. Staffing is marked for one.

Then come the detail pages, where thirteen providers describe their own charging in their own words, one more than the grid has rows. Read all thirteen and the word collaboration does not appear. Not as a model, not as a heading, not in passing. Across the entire module it occurs once, as a label on the ladder, and never again.

That is the finding. A rung with no definition on its own page, no column in the exhibit beside it, and no provider offering it by name across thirteen submissions was not a model anyone could buy. It was a word placed between two mechanics to make the distance between them feel like a step.

What I cannot tell you about where it sat

One honest limit. The ladder is a graphic, and text extraction returns label order rather than plotted coordinates. The document runs the label between gain share and unit pricing in its text, and I will not turn a text order into a claim about where the bubble was drawn. That position is not recoverable, and I am not going to invent it.

It matters less than it sounds. What is verifiable is enough: it was on the ladder, it had no definition, it had no column, and nobody sold it. A model that cannot be located and cannot be defined is not rescued by knowing where somebody drew it.

Three questions that take about a minute

The transferable part is a test, and it works on any framework with boxes.

Does this box carry a definition in the same document, and does that definition name what triggers a payment? Does this box have a column in the comparison exhibit next to it? Does anyone in the supplier set offer it under that name?

A box that fails all three is not a commercial option. It is connective language, doing work in the room that the contract never will.

8

Commercial models named on the ladder

7

Models carrying a written definition, and columns in the grid beside it

37 of 84

Provider-model cells marked, a 44% fill rate across a dozen providers

1

Times the undefined rung appears anywhere in the module

The same page, redrawn for something else

The proposals crossing my desk this year are not for service desks. They are for copilots: retrieval-augmented generation over an organisation's own documents, a vector store behind it, a review queue in front of it, and in the more careful cases an early, narrowly scoped autonomous experiment on one low-risk process. The commercial pages of those proposals are built exactly like that ladder.

Some rungs on them generate an invoice line. Seats per month. Documents indexed. Queries served. A platform fee with a per-resolution charge above it. Some rungs do not. Partnership. Co-innovation. A shared roadmap. A joint centre of excellence. Apply the three questions and the second set fails all three, proposal after proposal, from firms that are otherwise perfectly serious.

The undefined rungs are not lies. They are the shape a proposal takes when a seller wants a progression to look continuous and has no product for the middle of it.

The stakes are higher here than they were for a service desk, because more of what is being sold is genuinely hard to define. An outcome-priced copilot needs the same three things gain share always needed: a baseline agreed before deployment, an attribution rule, and a measurement window. Those need data readiness, lineage that survives a question about where a figure came from, and evaluation sets somebody maintains. Whether those three exist is the first thing our Platform work establishes, because where they do not, the honest offers are a seat fee or a defined unit price. What gets offered instead, often, is a partnership rung. European AI regulation is heading towards documentation duties of exactly this shape: how a system was evaluated, and what it was trained and grounded on. A rung with no definition cannot be documented, which makes it a governance problem as well as a commercial one.

Seven rungs told a buyer what generates an invoice line. The eighth told them how the relationship would feel. Only one of those can be signed.

Define it or delete it

Two options, and they are the only two.

Define it. Say what triggers the payment, what has to be true before the mechanic works, and how it will be measured. If that can be written down, it belongs on the ladder and it belongs in the grid.

Or delete it, and let the ladder show seven rungs a buyer can actually price. What is left is real information, including how thin the top of it is. Two providers out of twelve willing to share revenue tells a buyer more about that market at that moment than a comforting label does.

Leaving it is the one bad option, because an undefined rung does work in the room. It gives everyone somewhere to point when the conversation gets difficult. Nobody has to say the top of the ladder is out of reach for an organisation that cannot yet baseline its own incident volumes. They can say we will start collaboratively and move up. Then the contract is signed on the rung below, and the ladder is never opened again.

Counting the boxes and counting the definitions is where our Consult work on a shortlist starts, and it has never taken longer than an afternoon.

Figures are as recorded in a market scan of service-desk commercial models, produced inside a sourcing review for a large European public-sector organisation ahead of a re-tender: its ladder of pricing constructs, its provider comparison grid, and the provider-by-provider charging summaries behind them. That review produced findings and recommendations ahead of a sourcing decision, not delivered results. The counts are ours, recovered by reading the three exhibits against each other. The ladder's plotted positions do not survive extraction, and none is claimed here.