Why your AI licences aren't paying back — and the numbers behind it
Most enterprise AI and SaaS spend never earns its keep: around half of all licences go unused, and Microsoft Copilot adoption has stalled near 5%. Here is the data — and what actually fixes it.
The uncomfortable answer: most enterprise AI and SaaS spend never gets used. Around half of all software licences sit idle, and flagship AI tools like Microsoft 365 Copilot have stalled at single-digit adoption. The money is already spent — whether it pays back depends on adoption, not on buying more.
How much is actually wasted
The waste is not a rounding error. It is one of the biggest uncontrolled line items in the IT budget.
- The average organisation wastes roughly $19.8 million a year on unused SaaS licences, according to Zylo's SaaS spend research.
- About 50% of all software licences go unused — the highest waste rate on record.
- Even disciplined mid-market companies typically find 20–30% of SaaS spend going to waste, and SaaS licence waste now ranks as a top IT spend challenge for finance leaders.
AI is the newest line item — and the least used
The pattern repeats with AI, only faster, because the licences are new and the habits do not exist yet.
- Microsoft 365 Copilot adoption has stalled around 4.5%, with weekly usage near 1%.
- Fewer than 4 in 10 employees with Copilot access actively use it.
- Only about 16% of Copilot pilots convert to production.
- Gartner found that just 6% of enterprises have moved generative-AI projects past the pilot stage into production.
Why buying more never fixes it
When adoption is low, the instinct is to add seats, add tools, or add another pilot. None of that changes the reason the last batch went unused. The bottleneck is not licence count — it is that people do not have an approved, obvious, faster way to do their real work with the tools you already bought. More licences just widen the gap between what you pay for and what gets used.
What actually moves the number
Adoption is a behaviour-change problem, not a procurement one. The sequence that works:
- Measure real utilisation. Get a hard number for what is actually used versus paid for, per tool and per team.
- Cut or redeploy the idle licences. The recovered spend funds everything downstream — no new budget.
- Enable people on their real workflows. Role-based sessions on the work they already do, not a generic AI course.
- Grow internal champions. Adoption sticks when it is owned by your own people, not a vendor.
- Report ROI in hours and euros. What gets measured gets defended at the next budget review.
Common questions
Is unused-licence waste really that high? Yes. Independent research consistently puts unused SaaS licences around half of all seats, and 20–30% of spend even in well-run mid-market firms. AI tools are currently worse, not better, because adoption habits are still forming.
Does buying more training fix low adoption? Generic training rarely does. Adoption improves when enablement is tied to a person's actual workflow and there is a clear, approved path to use AI on real tasks. Otherwise usage decays back to baseline within weeks.
What is the fastest way to see our own number? Start with a free, 2-minute [Tool Waste Self-Check](https://aibusinesspro.eu/tool-waste-check) for an estimate, then a paid audit for the defensible figure.
The money is already spent. The only question left is whether it pays back — and that is decided by adoption, not by the next purchase order.