The fastest-growing line on your software budget is the least measured.
AI assistants were bought at pace across Asia-Pacific through 2025 and 2026, usually in bulk, usually before anyone defined what a user was supposed to do differently. The invoices arrived. The measurement did not.
Approximate share of purchased enterprise Copilot seats with an active user, per widely cited independent analysis. For every 1,000 licences, roughly 640 generate no return.
Share of purchased seats in weekly active use according to independent surveys — materially below vendor-reported monthly active user figures.
Effective cost per active user against list price, once idle seats are counted across the estate.
Figures are drawn from published third-party analyses of enterprise AI deployments and are indicative of the category, not a promise about your estate. Vendor-reported and independent figures diverge substantially — that divergence is itself the point. We measure your own numbers before making any claim about them.
Four things make this category worse than ordinary SaaS
Habit products decay
An AI assistant only earns its seat if someone returns to it. Week-one enthusiasm rarely survives to month four without a named task and manager follow-through.
Waste compounds
Most enterprise AI add-ons require a qualifying base licence underneath. An idle AI seat burns the full stack, not just the headline price.
Tools overlap
It is common to find teams paying for two or three assistants at once, bought by different departments, with no combined view of the cost.
Bought broad, used narrow
Rollouts are sized to headcount rather than to the cohort that can be proven to use it. The gap shows up at renewal, twelve months too late.
Pilots become baselines
A pilot seat count quietly becomes the renewal number because nobody produced evidence to argue it down.
Nobody owns the question
Procurement owns the contract, IT owns provisioning, the business owns adoption. Utilisation sits between all three.
India's IT majors already proved the gap is fixable.
Through the first half of 2026, the largest Indian IT services firms each scaled past 100,000 AI assistant seats and publicly reported active usage in the mid-80s to mid-90s percent — against a global average closer to a third.
The software was identical. The difference was implementation: named tasks, workflow integration, and follow-through after launch week. That gap between 35% and 90% is not a licensing problem. It is a management problem — and it is the one we are set up to solve.
- Sustained usage measured per cohort, not per licence purchased
- Dormant seats identified at 30 days and flagged before renewal, not after
- Overlapping assistants surfaced with combined cost attached
- Seat counts sized to the cohort you can prove, not the headcount you hope to convert
- Renewal entered with twelve months of evidence instead of last year's number
What we will not claim
This category attracts more overstatement than any other in software. We would rather be dull and accurate.
- We are not an AI company. We source and measure AI licences; the intelligence is the vendor's.
- We cannot make your teams adopt a tool. We can show you precisely who has and who has not, and what that costs.
- Not every AI product is available through regional distribution. Where we cannot source something, we will tell you and you should buy it direct.
- Published utilisation figures vary widely by source and methodology. Treat them as category context, not as a forecast for your business.
How many of your AI seats were opened last week?
Most businesses cannot answer that. The review answers it in about a week, at no cost.
Request a license review