Before the purchase order
The Three Questions to Ask Before You Approve the Next AI or Data Tool Purchase
The price can be right while the decision is still incomplete. These three questions close the gap before you sign, not after.
Where this starts
A finance team approved $50,000 in AI platform credits. Tokens, in the vendor's own pricing. Clear unit price, a plausible use case, a vendor everyone has heard of. The three questions below were never asked before signing.
What you could see from inside
The price was real. It was not the issue.
A number that clean is easy to approve. What it replaces, what it costs to keep alive, and the risk underneath it: none of that shows up next to the price.
“The token is the billing unit, not the value unit.”
What was invisible from where they were standing
Three questions nobody had answered.
01 · Value
What can this replace or improve, compared with what the team already does today?
Tokens bought are not value extracted. Write down what the team does today, how long it takes and what it costs. That is your before number. Without it, there is nothing to measure the tool against.
Every standard return method starts there. Total Economic Impact needs the cost displaced. NPV needs the cash-flow comparison. A productivity multiplier needs the output from before the tool arrived. No before number means $50,000 stays a line item, not a return.
02 · Long-term run cost
What happens as usage scales and the tool needs support, retraining or repair?
The token price is the entry fee. It is not the cost of keeping the capability useful.
One benchmark, drawn from more than 100 enterprise projects, puts change management at 20% to 30% of the AI project budget. A separate cost analysis puts ongoing maintenance and operations at 15% to 25% of the original implementation cost each year, with data quality, compliance and technical debt among the costs most often missed.
Sources: Alice Labs, AI cost-benefit analysis · Pertama Partners, AI implementation cost breakdown
03 · What you don't know yet
Who is accountable, and is the data feeding this structured enough to trust?
Governance and data readiness appear nowhere on the invoice. The Risk Checklist is where you check them: eight questions you score yourself on, covering accountability and whether the data behind the tool can carry its answers.
In EY's 2026 survey of 202 senior AI executives at large US companies, 69% were concerned that their organization lacked the internal expertise to keep evolving its AI governance controls. Nearly half of organizations using AI agents had not updated governance for agentic risks.
Source: EY, AI governance gap survey
The pattern
The vendor's unit is easier to approve than your outcome.
It happens whenever a purchase is priced in the vendor's unit: tokens, seats or credits, instead of the buyer's outcome.
The unit price is concrete. The value, the real running cost and the risk underneath are not. None of the three appears on the invoice. Catching that before you sign is what makes you the sharpest reader in the room.
What changes
Buy it. With the answers first.
This is not an argument against the purchase. It is a case for signing it with all three questions answered, so the number you approved is the return you can show. Asking first costs a short conversation. Asking after the spend is committed costs far more.
Before the next purchase order