A first analytics capability does not have a price. It has two, and the distance between them is not a negotiation about day rates. It is a question about what the buyer is actually purchasing.
A European food and consumer goods manufacturer, looking to get started with analytics across marketing, sales and its supply chain, was given two costed ways in. The first put a small team of analytics specialists and data scientists on one agreed business question: 1.3 FTE, six to eight weeks, EUR 7,300 a week. The second ran the same weeks against two questions instead of one, at 2.1 FTE, and added a cloud analytics platform with its licensing: EUR 15,000 a week. Multiply the rates by the quoted duration and the first lands somewhere around EUR 44,000 to EUR 58,000, the second around EUR 90,000 to EUR 120,000. That arithmetic is ours. The proposal quoted a weekly rate and a duration and left the total for the reader to work out, which is itself a choice about how the thing gets signed.
Roughly double the run rate. Only 1.6 times the people. Everything left over is the platform.
The challenge
The question that arrives from a board is what a data capability costs, and the only honest answer is that it depends which of two different objects you are buying. The cheaper option buys an answer to a question you already have. The dearer one buys somewhere for answers to keep arriving after the team has gone.
The proposal was blunt about the selection bar. Nothing counted as worth running unless it returned five times the effort it consumed. That is a crude ratio and its crudeness is the point. Weighted scoring grids exist so that nothing gets rejected, which is how organisations end up with a portfolio of pilots and no result. A single ratio at five to one rejects most candidates in the room where they are proposed, and it forces the sponsor to say out loud what the answer would be worth if it were true.
The duration carried its own surcharge, stated in a footnote rather than in the headline. Work of this shape normally runs four to six weeks where an analytics environment is already established. Six to eight was quoted here because it was not. One to two extra weeks, priced at the same weekly rate, is EUR 7,300 to EUR 14,600 on the cheaper option and EUR 15,000 to EUR 30,000 on the dearer one. Data readiness is not an abstraction that gets discussed in a governance forum. It appears on the first invoice, before a single model has been fitted, as the cost of finding out what the source systems actually contain.
The approach
The shape of the work was priced in three blocks with the weeks written against them: one week at the front to agree the question and what an answer would be worth if it were true, four to six weeks in the middle to get the data into a state where it can be interrogated and to test what it says, one week at the end to stand up benefit tracking. Framing and closing take a week each. Everything that can go wrong lives in the middle block that consumes the rest, and the platform question sits inside it.
Both options bought that middle block. What separated them was where the work ran. Under the cheaper option the manufacturer supplied the environment, and the effort plan was explicit that choosing the analytics platform, and putting security and technical safeguards in place before any data moved, were ninety per cent the client's own labour. Under the dearer one the supplier brought a connected environment and its licensing, so the eight weeks could begin without waiting on an internal procurement cycle. The extra roughly EUR 7,700 a week is not buying more analysis. It is buying the removal of a dependency.
That inverts the usual reading of the two lines. The dearer option is not the ambitious one, it is the impatient one, and the arithmetic says so: adding a second business question and the whole environment it runs in costs marginally more per week than the entire first option did on its own. If you already have somewhere to put the data, you are paying a second time for something you own.
Which is where a costing exercise turns into something we build. Choosing a platform takes an afternoon. Establishing which source systems can be read on a schedule, which fields are trustworthy enough to join on, what the lineage looks like when a number gets challenged in a management meeting, and how a result is promoted out of a notebook into something that runs every Monday: that is the Platform work, and it is what the weekly rate is really paying down.
The proposal also named, on the same pages, the conditions that break work of this kind: absent sponsorship, no candidate clearing the return bar, data that cannot be reached across silos or that is held by people with reasons to hold it, business specialists and data scientists who are never in the same room, no habit of reporting benefits once the closing presentation is over, and no appetite for improving the underlying fact base once the first answer is in. Six failure modes, priced into neither option, and not one of them technical.
The outcome
Nothing described here is a delivered result. This is a priced proposal, and every benefit figure in the surrounding pages belongs to other clients or to an illustrative example rather than to this manufacturer. What the document produced was a decision put in front of a buyer with real numbers attached, which is more than most capability discussions manage.
The commercial design deserves one note. The scoping workshop that selects the first question was priced at zero, conditional on the client buying one of the two delivery options. A discovery session offered free if you subsequently buy is not generosity. It is a commitment device, and it works because the workshop produces the shortlist that makes the delivery contract signable.
Three things would be argued differently today. The first is the premium in the dearer option. Renting a managed analytics platform by the week is now ordinary, so paying a consultancy to bring one is a smaller advantage than it was. What cannot be rented is the deployment discipline around it: a feature store so the same definition of a customer or a batch is used twice, lineage so a challenged number can be traced, and a path from an experiment to a scheduled job that does not run through one person's laptop. If you are buying the dearer option now, buy that, not the licences.
The second is the readiness surcharge. Those one to two extra weeks were estimated from experience. They are measurable now. Process mining over the event logs in order management or production planning gives a defensible picture of how the process actually runs and where the records break, in days rather than in a survey of opinion.
The third is that the early and cautious language-model pilots now appearing in commercial functions change none of the arithmetic above. A model that reads a supplier contract or drafts a promotion brief still needs somewhere for its output to land, and where the pipeline is thin it lands in a person's inbox and the automation is a demonstration. Our Consult engagements start by measuring that landing point, because it separates an interesting pilot from a capability.
Two prices for the same six to eight weeks. The lower one tests a question. The higher one starts a function. Choosing between them is not a procurement exercise, it is the first honest statement an organisation makes about which of the two it intends to have.