AI usage assessment: how to connect usage, licenses and spending

Restitution d’un diagnostic des usages de l’IA en entreprise

AI usage assessment: how to connect usage, licenses and spending

You know the price of your main AI subscriptions. But do you also know which licenses are still in use, which profiles lack capacity and how much spending escapes your contracts? No single source can tell you. An assessment brings these sources together so you can make decisions.

Start from a decision to be made

An assessment is all the more useful when it answers a specific question: preparing a renewal, explaining a budget increase, reviewing a team’s capacity, bringing unapproved tools under control. That question determines the data to gather and the period to study.

To adjust licenses, you need to match assigned seats, activity and contractual commitments. To understand a rise in API consumption, you need to look at volumes, models and the applications involved. Scoping also states what the assessment will not cover: out-of-scope entities, non-enrolled computers, unavailable sources.

Gather complementary sources

No single source describes the situation on its own. Each one provides part of the answer:

SourceWhat it providesWatch out for
Invoices and contractsAmounts, plans, commitments and renewal datesA payment says nothing about whether a license is useful.
Corporate cards and expense reportsPurchases made outside central contractsThe same purchase may appear in several sources.
Vendor consolesLicenses and activity of managed accountsCoverage varies by plan, and no conversational usage.
Workstation measurementActual activity by tool and by plan, in the office and when working remotely, with indicators on chat usageDepends on enrolled computers and supported tools.
Network logs (proxy)AI services reached through the corporate networkDon’t cover computers off the network.
DLP tool recordsSensitive data exposed in promptsOnly if a DLP tool is already in place.
HR directory and approved toolsAssignments and usage rulesKeep up to date, limited to what is necessary.

Cross-checking these sources is what produces a single, reliable view of usage. Reconciliation requires rigor: same periods, same suppliers, same populations, same rules for handling amounts.

Turn each gap into a check, then into an action

A license with no observed activity isn’t necessarily useless. Its holder may be on leave, have a seasonal need or use a feature the measurement doesn’t cover. Two subscriptions for the same person may complement each other, and heavy consumption may justify a higher-capacity plan. Each finding therefore goes through a check:

FindingCheckPossible action
A seat appears inactiveAbsence, occasional need, measurement coverageReassign, or adjust at renewal
Several subscriptions for one personComplementarity and business needsKeep justified plans, remove a confirmed duplicate
A user hits their limitsFrequency, tasks, cost of optionsIncrease capacity or change plans
An unapproved tool is observedNeed, account used, applicable rulesAssess, approve under conditions or offer an alternative

This is what keeps an activity measurement from turning into an automatic decision.

Distinguish actual spending, API equivalent and savings

The API-equivalent cost values a volume of tokens at API rates. It measures how intensive usage is, but it doesn’t match the bill for a subscription.

Likewise, an identified saving is not a realized saving. Cutting ten licenses only has an effect if the contract allows it, often at the next renewal. Good tracking distinguishes three states: the potential saving, the action decided and the reduction observed on the invoice, without forgetting the capacity to add elsewhere.

Estimate the carbon footprint without overpromising

Environmental indicators help compare usage and models, provided they are presented for what they are: estimates expressed as ranges, based on external frameworks such as those of ADEME (the French Agency for Ecological Transition) and Boavizta. Their value is first and foremost educational: making teams aware of the impact of their choices. A change of model is also judged on the quality achieved, turnaround time and rework needed. A lower unit cost isn’t enough.

Prepare the measurement with the teams involved

Scoping sets the purposes, the data required, the people authorized to access it and the retention period. Employees must be informed in advance: in the United States, several states require employers to give notice before monitoring electronic activity, and in Europe employee representatives must also be consulted. In all cases, such measures must be justified and proportionate. Passive measurement, where no conversation content ever leaves the computer and data collection is limited to what is strictly necessary, makes this dialogue much easier.

What a good findings presentation should enable

A useful findings presentation sets out the sources and their coverage, the reconciled spending, the situations to examine and the proposed actions. For each recommendation, it specifies an owner, a deadline, implementation conditions and financial assumptions. The assessment then becomes a decision-making tool shared by IT, Finance, Procurement, Security and the business.

To go further: how to reconstruct your AI spending and how to detect shadow AI.

How Saaswedo Can Help

With sobrAIvo, Saaswedo runs this assessment in about six weeks. Our AI Trace software passively measures actual AI usage on employee computers, in the office and when working remotely, then our consultants cross-check it with your other data in the portal to build a costed action plan with you, which also covers risky usage and the estimated carbon footprint.

1600 900 SAASWEDO | Telecom Expense Management & Managed Mobility Services