Methodology
How we calculate what you're wasting, and what we deliberately don't claim to know.
Wasted licence spend
Every licence lands in one of four states, based on the last recorded activity from the vendor. Each state counts differently.
| State | Definition | Counted as waste |
|---|---|---|
| Never used | assigned, with no activity ever recorded | 100% |
| Inactive | no recorded activity in the last 45 days (configurable) | 100% |
| Low use | active, but under 10 interactions in 90 days (configurable) | 50% |
| Active | used regularly — not waste | 0% |
Low use counts at 50% on purpose: someone using a tool a little may still need it, so we don't claim the full amount as recoverable. Two low-use criteria, not one: an absolute threshold you set, and a relative one — the bottom decile of the person's own department, only for groups of 10 or more. The relative one exists because a fixed number judges a team that uses the tool 200 times a week the same as one that uses it 5. Neither ever counts the same licence twice.
What counts as a billable AI user
- A person holding at least one licence of an AI tool that is active in your catalogue. One person with four AI tools counts once.
- Employees without an AI licence do not count, even if they are in your directory. Syncing your whole company costs you nothing.
- API consumption without a licence — a service account calling an API, for instance — is measured and reported, but does not count towards billing.
- The count is taken at the moment of invoicing, not as an average: it goes down as soon as you retire licences.
Which price we use
- Your contracted price, if you've entered it — this is the accurate one.
- Otherwise the vendor's list price, as a reasonable estimate.
- If there is no price at all, we do NOT invent one: the tool appears without a figure and we ask you to set it.
Overlapping tools
We flag a person holding two or more tools in the same category (for example two coding assistants). Holding both is not waste by itself — we only count savings when at least one of them is unused, and only for that unused licence.
The decision stays yours: a developer may legitimately need two. Wattprint surfaces the case with the numbers; it doesn't rule on it.
Confidence levels
Every action carries a confidence level, so you know which ones you can act on immediately and which need a look first.
- high your contracted price and activity data from the vendor's API
- medium list price, or activity from a CSV export
- low missing price or incomplete data — we tell you what to fill in
Environmental footprint
Only for API usage where we have real token counts from the provider. We never estimate the footprint of licences we can't measure.
- Primary engine: EcoLogits (GenAI Impact) life-cycle methodology, where the model is characterised.
- Fallback for uncharacterised models: documented static engine — energy per model class, PUE 1.2, embodied-hardware factor 1.25, WUE 1.8 L/kWh.
- Emission factor: annual average grid intensity for the zone your organization configures (sources: Ember / Electricity Maps).
- Every record stores which engine calculated it, so any figure is traceable back to its method.
What we don't claim to know
- Whether a person truly needs a tool. We measure recorded activity, not value delivered.
- Usage a vendor doesn't report. If a provider exposes no activity data, we say so instead of guessing.
- Personal AI subscriptions paid on someone's own card. That is not visible to any admin API, so we don't pretend to detect it.
- Whether your directory is complete. Our reports are only as good as the data your directory holds.
Per-tool detail is in Integrations