For the chief financial officer
We have not seen your volumes, your cost base or your exception rate, so any figure we produced would be arithmetic performed on assumptions your own sponsor supplied to a model we built. You have approved one of those before. What is here instead: what you are committed to, what varies, what leaving costs, and where a saving would have to show up for anyone to believe it.
By the time it reaches you the sponsor has run a pilot they liked, the supplier has supplied a model, and the executive team has heard the headline. Your job is described as approving the funding and is actually the last opportunity anybody has to ask whether the arithmetic survives contact with a real cost centre.
And the models are always favourable, because of who builds them and from whose inputs. The hours saved come from an estimate. The rate applied to those hours is fully loaded when it flatters and marginal when it does not. The benefit lands in year one and the disruption is not modelled at all. None of that is dishonest exactly; it is a document produced by somebody whose interest is a signature.
Then the savings do not appear. Not because the tooling failed, but because nothing structural changed underneath it: the same team is on the same establishment, the freed hours were absorbed by work that was already being deferred, and no line in any budget went down. The benefit was real and it was never bankable, which is a distinction almost nothing in the purchasing process is built to make.
Meanwhile the commitment compounds quietly. An annual subscription renews, a seat count that only ratchets upward, an implementation charge nobody attributed to the case, and a contract whose exit terms were never assessed because assessing them felt like negativity during an approval. Three years later that spend is a fixture and the original justification has left with its author.
Almost every case for this category of purchase converts an estimate of hours into money, but hours only become money when an establishment, a contract or an invoice actually changes — and the case rarely says which one.
That is the question worth forcing, and it is worth forcing before signature rather than at the first renewal: if this works exactly as described, which specific line falls, by how much, in which period, and who owns making that happen? A case that cannot answer it is describing capacity relief. Capacity relief is genuinely valuable and it is not a saving, and conflating them is how a portfolio of successful projects coexists with a cost base that has not moved.
Three shapes bank cleanly, and they are worth separating from the rest. A tool actually decommissioned, which shows up as a cancelled subscription with a date. A contractor or agency engagement not renewed, which shows up as an invoice that stops. A seasonal or overflow arrangement not entered into, which shows up as a cost never incurred. Each of those is checkable by your own team against your own ledger without anybody trusting a model.
The rest — faster handling, lower backlog, work that finally happens — is real, valuable, and only becomes money if something downstream changes as a consequence. Say so in the case as capacity relief, and it survives your scrutiny. Dress it as savings, and it does not survive the first review that goes looking for the line.
On our own side: the free tier is genuinely free and stays that way, which means a pilot can establish whether the mechanism works before you have committed anything. That is not a discount strategy. It exists so the first commitment can be made against something observed instead of something projected.
Whether a named budget line actually falls — measured by the specific line, in the specific period, against the baseline taken beforehand.
Whether a subscription is genuinely cancelled — measured by cancellation date and final invoice for each tool named as a retirement candidate.
Whether contractor or agency spend stops — measured by invoices that cease, compared against the run rate for the two prior quarters.
Whether a seasonal arrangement is avoided — measured by the overflow or temporary cost incurred this peak against the same peak last year.
How much of the commitment is fixed against variable — measured by the contracted floor, the volume-linked component and the stated ceiling, read from the agreement.
What ending it would actually cost — measured by notice period, data-return timeline, and your own estimate of internal reconstitution effort.
Whether the case survives its author leaving — measured by whether the banking route is written down and owned by a named person rather than remembered.
a return figure, a payback period or a percentage of anything. Not modesty and not evasion — we have not measured your operation, and a number produced without measuring it is a sales artefact that your own review process exists to catch.
Implementation effort is the line most often missing from a business case and it is not zero. A first bounded scope needs somebody who owns the process, somebody who can grant authority, and access to the systems holding the relevant records. It does not need a migration, and where a supplier’s case assumes one without pricing it, that is worth finding before approval rather than after.
It joins your existing estate rather than replacing your systems of record, so the finance ledger, the records platform and the case-management system stay and keep their existing costs. That is a smaller change than a replacement and it is also a smaller saving, and both halves of that belong in the case.
Where a connection to something you depend on does not exist yet, it is named during scoping with the work it would take. A cost identified at scoping is a line in the case. The same cost identified during implementation is a variance you have to explain.
The commercial questions a finance function should press on are not the same as the security ones, and they are usually answered with less precision because nobody is auditing the answer.
What is below is the commercial position. Where something is a design intention rather than an attested fact it says so, because a commercial claim that turns out to have been aspirational is discovered at exactly the moment it costs the most.
Ask for these while the sponsor still needs your signature. Everything below is cheap for us to provide before the agreement and awkward to obtain afterwards, which is exactly the asymmetry to use.
The exit and data-return terms are the highest-value request on the list, and the reason is structural: they are worth the most to you at precisely the moment they cost us the most to make specific.
One bounded function with a cost you can state today — small enough that stopping costs a scope, specific enough that a named budget line could plausibly move because of it.
Measure the current cost of the function before anything starts. Taken beforehand it settles the question afterwards. Reconstructed afterwards it becomes a negotiation between people who each remember a different starting point, and the reconstruction always favours whoever is more motivated.
Then make somebody own the banking route by name. Not the benefit — the route. Which line, whose budget, which period, and who is accountable for the cancellation or the non-renewal actually happening. A benefit with no owner is a benefit that does not occur, and that is true of well-run organisations too.
And read the exit terms while you are still the party with leverage. A finance function that has assessed the exit before signature has priced the whole relationship. One that has not has priced the first year.
We could produce one in an afternoon and it would be worth nothing to you, because the inputs would come from your sponsor and the model would come from us. You have approved that document before and watched the saving not appear in a cost centre. What is genuinely useful is narrower and checkable: name which budget line should fall, take the baseline before starting, and check the line afterwards. If it moved, you have a measurement. If we had given you a figure instead, you would have an assertion and an argument about whose assumptions were wrong.
With a bounded first scope, a baseline taken beforehand, and a named line that should move — presented as a test rather than as a projection. Boards approve tests that cost a scope far more readily than they approve projections that cost a programme, and the number you bring to the second conversation is then a measurement from your own ledger rather than a supplier’s arithmetic. That is a stronger paper than the one you would have written from a return model, and it is stronger precisely because it is smaller.
Ask for the commercial structure in writing before signature: the fixed floor, the volume-linked component, the ceiling, and whether any count can ratchet upward on renewal without a new signature. That last clause is where multi-year exposure usually hides, because it converts a decision you made once into a cost that grows without anybody deciding again. Where a bring-your-own model provider is used, that spend sits on your own account at your own rates with no margin added by us, so it is visible to you directly rather than through our invoice.
A pilot proves the mechanism functions. It does not establish that anything bankable follows, and those are different findings that the same document usually reports as one. The useful question for the sponsor is which specific line they expect to fall as a consequence and who owns making that happen. If the answer is capacity relief rather than a line, that is a legitimate and valuable answer — and it should be written in the case as capacity relief, because a case that calls it savings will not survive the review that goes looking for the line.
Three things, and only the first is contractual: the notice period, the data-return timeline, and your own internal effort to stand the function up elsewhere — which is usually the largest of the three and is never in a supplier’s exit clause. Ask for the export format and the reconstitution procedure before signature; the answer tells you more about a supplier than anything else you can request, precisely because it is the one question they gain nothing by answering well.