Enterprise spending on artificial intelligence software is accelerating. According to IDC, worldwide AI software revenue is forecast to exceed $250bn by 2027, up from roughly $64bn in 2022. Yet a growing body of evidence suggests that a significant portion of that expenditure is wasted on licences that employees do not use or use only sparingly.
This is the procurement paradox: companies are buying AI tools at scale, often under pressure to demonstrate AI readiness, while internal adoption lags far behind. The result is a financial drag that undermines the business case for AI investment and distorts procurement strategy.
The Scale of the Waste
Industry benchmarks for traditional software utilisation are already sobering. Gartner has estimated that as much as 30 per cent of enterprise software spend is wasted on unused or underused licences. For AI-specific tools, the picture may be worse.
A 2023 survey by software asset management firm Flexera found that organisations typically use only 40 to 60 per cent of their SaaS licences. AI tools, which often require specialised training and workflow integration, tend to sit at the lower end of that range. Early adopters report that many employees simply revert to familiar manual processes rather than learning new AI interfaces.
One Fortune 500 technology company, speaking on condition of anonymity, told The FY Times that it had purchased 5,000 seats of a generative AI assistant but saw active usage from fewer than 1,200 employees after six months. The remaining 3,800 licences were effectively dormant, costing the company approximately $1.5m annually in unused capacity.
Why Enterprises Overbuy
Several structural factors drive this over-procurement. First, the competitive pressure to adopt AI is intense. Boards and investors expect visible AI strategies, and procurement teams are incentivised to secure enterprise-wide agreements that signal commitment. Volume discounts further encourage buying more seats than needed.
Second, AI tools are often procured by central IT or innovation teams without close consultation with end-user departments. The people who will actually use the software are rarely involved in the purchasing decision. This disconnect leads to tools that do not fit existing workflows, require steep learning curves, or duplicate functionality already available in other platforms.
Third, many AI vendors structure their pricing around per-seat models inherited from the SaaS era. These models reward broad deployment rather than deep adoption. Vendors have little incentive to encourage efficient usage, and some actively discourage seat reduction through contractual minimums and annual commitments.
The Commercial Impact
The financial consequences are material. For a mid-sized enterprise spending $10m annually on AI software, a 40 per cent utilisation rate implies $6m in waste. That is not merely a procurement inefficiency; it is capital that could have been directed toward training, integration, or custom development that would actually drive adoption.
There is also an opportunity cost. When budget is locked into underused licences, it reduces the organisation's ability to experiment with new AI tools or scale the ones that do work. The procurement paradox thus creates a self-reinforcing cycle: poor adoption leads to poor ROI, which leads to tighter budgets, which leads to even less investment in the conditions for adoption.
Why It Matters
For founders and operators, the procurement paradox signals a market inefficiency. Startups that offer usage-based pricing, outcome-based contracts, or adoption consulting alongside their software may gain a competitive advantage over incumbents that cling to per-seat models.
For investors, the gap between purchase and usage is a red flag. Companies that report rising AI software spend without corresponding productivity gains may be masking deeper operational problems. Due diligence should now include a review of software utilisation metrics, not just procurement totals.
For enterprise executives, the message is clear: buying AI is not the same as adopting AI. The business case for any AI investment should include a realistic adoption plan, not just a licence count.
Risks and Unknowns
Several uncertainties remain. It is not yet clear whether the current low utilisation rates are a temporary phenomenon as organisations learn to integrate AI, or a structural feature of the market. If the latter, vendors may be forced to shift pricing models, which could disrupt revenue forecasts and valuations.
There is also a risk that the data on utilisation is itself unreliable. Many enterprises do not track software usage granularly, and those that do may use inconsistent metrics. Active users, for example, can be defined in multiple ways, making cross-company comparisons difficult.
Finally, the rapid pace of AI development means that today's underused tool may become essential tomorrow. Procurement teams must balance the risk of overbuying against the risk of being caught short when demand suddenly rises.
FY Outlook
We expect the procurement paradox to become a more prominent topic in boardroom discussions over the next 12 to 18 months. As CFOs scrutinise AI spending more closely, utilisation data will become a standard part of procurement reviews.
Vendors that proactively offer usage-based or consumption-based pricing will likely gain market share, particularly among cost-conscious enterprises. We also anticipate the emergence of third-party tools that audit AI software usage and provide recommendations for renegotiation or consolidation.
For now, the most prudent course for enterprise buyers is to pilot AI tools with small, motivated user groups before committing to enterprise-wide agreements. The evidence suggests that broad deployment without deep adoption is a recipe for waste.
Conclusion
The procurement paradox is not a sign that AI is overhyped or that enterprise investment is misguided. It is a sign that the buying process is misaligned with the adoption process. Until procurement teams, end users, and vendors align their incentives around actual usage, the gap between purchase and utilisation will persist.
Enterprises that close this gap will not only save money; they will build the organisational muscle needed to deploy AI effectively at scale. Those that do not will find themselves paying for a future they have not yet learned to use.
