Interested Definition
Where the party that benefits from a measurement also controls what it counts, so the definition and the incentive point the same way.
When not to use it
- Where an industry standard fixes the definition, which removes the discretion the concept describes.
- As an accusation. The defining party usually has the only information sufficient to write a definition at all.
- Where the alternatives do not move the number, in which case the discretion exists and does not matter.
Reach for something else instead
- Definition in the contract — moves the load-bearing half of a deal into the document that binds both parties.
- Event-level disclosure — publishing the breakdown beneath the total, which lets an outside party recompute under an alternative definition.
Read more on the blog
- Same GPUs, same month, opposite depreciationTwo companies bought the same hardware, both were audited, and they reached opposite conclusions about how long it lasts. The difference flows straight into reported profit.
- Same company, same year, revenue figures 63% apartThe most quoted numbers in AI are run-rates from private companies, reported by outlets triangulating from leaks, and sources covering the same period differ by multiples.
- AI support: 90% deflected, 40% actually resolvedA support system can report 90% deflection on a 40% resolution rate, because a customer who gets an unhelpful answer and gives up counts as deflected. The most-cited deployment reversed.
- Twenty-four hours of silence is a billable resolutionCustomer service AI has moved from per-seat to per-resolution pricing. The corpus already established that deflection and resolution count different events, and now that distinction sets an invoice.
Further reading
- Raji et al. (2021), AI and the Everything in the Whole Wide World Benchmark — operationalisation choices shaping what a measurement can support.
- Zheng et al. (2023), Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena — model evaluators carrying position, verbosity and self-preference biases, which matters when a model verdict becomes a billable event.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Disputing the arithmetic when the disagreement is definitional, which loses an argument that was never about the sum.
- Negotiating a rate without negotiating the unit, which settles half a contract.
- Treating a stated definition as a neutral one because it was disclosed.
At a glance
Where this sits
A starting point. Nothing needs to come before it.
Computed from the prerequisite graph, not assigned. How this works