Complete IT visibility fell to 36% in 2026, down from 43% in 2025, according to Flexera's 2026 State of ITAM Report. The usual explanation is that there is simply more software to track. That is true and it is not the interesting part. The stack is changing shape from two directions at once, and neither runs through the system that holds the inventory. On the buy side, business units now control 81% of SaaS spend. On the build side, 32% of organizations have declined to buy software they decided they could build themselves.
Two things are true at once. Software asset management has genuinely improved as a discipline, and the organizations practicing it can see a smaller share of their own estate than a year ago. Both come from the same report, and they stop being in tension once you notice the improvement and the decline are on different axes.
Why is software stack visibility falling?
Flexera's 2026 State of ITAM Report, published June 24 2026 from a global survey of 512 technology professionals, is unusually direct about the mechanism. It names three root causes for why SaaS is harder to manage than expected: a growing number of applications, decentralized procurement, and increasingly complex software use rights.
Only the first is a volume problem, and better discovery tooling solves volume problems. The other two are structural. Decentralized procurement means the purchase decision happens outside the function that maintains the record. Complex use rights mean that even a complete list of what you own does not tell you what you are permitted to do with it. A discovery tool pointed at IT-managed infrastructure keeps getting better at seeing IT-managed infrastructure, and that is not where the estate is growing.
The rest of the picture is consistent with a visibility problem rather than a spending problem. 78% of ITAM teams name an accurate software inventory as their top priority, which is a strong signal that the inventory is not currently accurate. 31% report visibility into AI software specifically, and only 29% measure the value of AI software at all. Fewer than a third of organizations are checking whether the software category they are expanding fastest returns anything.
What is changing on the buy side?
Zylo's 2026 SaaS Management Index, published January 29 2026 and built on analysis of more than 40 million SaaS licenses and $75 billion in spend under management, puts a number on where purchasing authority has gone. Business units now control 81% of SaaS spend. IT directly manages 15%.
Those two figures do not sum to 100%, and Zylo does not break out the remainder, so the defensible reading is a clear business-unit majority against a small IT minority rather than a precise partition.
Decentralized procurement is now the default, not the exception
The mechanism is visible in the same report. Expense-based SaaS spend increased 267% year over year, with ChatGPT now the most expensed application. Software is entering on an employee credit card and surfacing weeks later as a line in an expense report, if it surfaces at all. Within large enterprises, organizations add an average of 21 applications per month even while total application counts look flat year over year. A stable headline number sitting on top of constant churn is the shape of a metric that has stopped being informative.
The report's license figure is the one most often quoted and most often misread. Organizations leave an average of 36% of SaaS licenses unused, and Zylo states plainly that this is measured against industry-recommended utilization levels. That is a benchmark comparison, not a count of seats nobody has ever logged into. Read loosely, the number argues for cutting seats. Read accurately, it argues for finding out which seats do work.
This 36% and Flexera's 36% are unrelated: Flexera's is a share of organizations reporting complete visibility, Zylo's a share of licenses inside one. Zylo also sells SaaS management software, which is why it matters that Flexera, surveying a different population, independently reports SaaS wasted spend continuing to rise.
What is changing on the build side?
The second movement is newer and almost nobody is counting it yet.
McKinsey's State of AI survey, published August 25 2026, found that 32% of respondents say their organizations have decided against purchasing at least one software product or feature because they were able to build the functionality in-house using agentic coding tools. Among the respondents McKinsey classifies as AI high performers, the share is nearly half. McKinsey's own framing is careful: this "could be a sign that AI is beginning to reshape how technology budgets are allocated."
The scope limit on this figure
This is a survey of 1,719 respondents in 97 nations, fielded May 4 to June 8 2026, reporting decisions their organizations made. It is not an audit of software budgets, and "at least one product or feature" is a low bar. Anyone reading a collapse in enterprise software demand into it is overreading, and McKinsey's own EBIT data argues against that reading anyway: 37% attribute any EBIT impact to AI, essentially unchanged from a year ago.
What survives the caveats is the direction. A product that is never purchased generates no contract, no renewal date and no license record, and the internal component built in its place generates none either. Both are invisible to every system built to track software, for the same reason: those systems track transactions, and no transaction occurred.
Is this a procurement problem or a measurement problem?
It is diagnosed as the first and behaves like the second. Nobody in this picture is wrong and, in most organizations, nobody is even arguing. Finance, IT and the business units read the same estate through instruments each built to answer one question well, and no system holds the readings together.
| Instrument | What it sees accurately | What it cannot see |
|---|---|---|
| Finance ledger and expense system | Every dollar that left, and which budget it left from | Whether the seat that dollar bought does any work |
| ITAM and license inventory | Sanctioned, contracted, entitled software | The majority of spend that did not come through IT |
| An application's own admin console | Logins and seat assignment inside that one tool | Whether the same capability is already paid for elsewhere |
| Observed work in the tools where work happens | Which people and teams actually use which capability | Whether the cost and contract data it is joined to is complete or current |
Each row is authoritative for its own question and for nothing else. Declaring one the single source of truth does not reconcile the others. It picks a winner and suppresses the disagreement, the same failure pattern that produces two defensible and different headcount numbers from finance and HR.
This is not abstract. A CFO at a private-equity-backed IT services holding company described three days of manual reconciliation for every cash flow forecast, because each acquired agency kept its own tools and no system held the combined view.
What would it take to see the stack whole?
Four things, and only one of them is a tool.
- A written definition before the number. Decide what counts as an active license, in writing, before two functions produce different totals. "Unused" against a vendor benchmark and "unused" against zero logins are different metrics, and reconciling them afterward is a negotiation rather than an analysis.
- An independent read on usage. The application's own console reports on itself. A read across tools is the only way to find the same capability paid for twice by two cost centers. That is the job of SaaS Stack Intelligence, which surfaces utilization, duplicates and abandoned licenses across every connected tool, by team and role.
- An explicit join between usage and cost, and an owner for it. Levos measures work through its connectors and reports cost per active user against the spend data it is given. It does not hold contracts, purchase orders or the accounts payable ledger, so the completeness of that spend data is the organization's responsibility. A cost-per-active-user figure computed on a stale contract value is precise and wrong.
- A disclosed confidence level. A reconciled figure built on partial coverage is an estimate, and it survives a board conversation better when it says so. That is the principle behind our measurement methodology: controlled cohort analysis with confidence scoring rather than attribution we cannot support.
The organizations that get this right will not be the ones that bought better discovery tooling. They will be the ones that stopped treating the transaction record as the system of record for something that is no longer, mostly, a transaction at all.
Frequently asked questions
Why is software stack visibility falling when SaaS management tooling has improved?
The tooling improved on one axis while the stack moved on two others. Flexera found complete IT visibility dropped to 36% from 43%, and names decentralized procurement and complex use rights among the causes. Neither responds to better discovery.
Who actually controls SaaS spend inside a typical organization?
Business units, by a wide margin. Zylo reports 81% of SaaS spend controlled by business units against 15% directly managed by IT. The two figures do not sum to 100% and Zylo does not break out the remainder.
Does 36% of SaaS licenses being unused mean 36% of seats are never logged into?
No. Zylo measures against industry-recommended utilization levels, a benchmark comparison rather than a count of dormant seats. Read loosely it argues for cutting seats. Read accurately it argues for finding out which seats do work.
What does agentic coding have to do with SaaS visibility?
It changes the stack from the other end. McKinsey found 32% of respondents reporting their organization declined to buy at least one product or feature because it could be built in-house. A tool never bought produces no contract, no renewal and no license record, and neither does the internal component built instead.
Can a workforce intelligence layer fix this on its own?
No. Levos supplies an independent read on observed work and cross-tool utilization, and reports cost per active user against the spend data it is given. It does not hold contracts, purchase orders or the AP ledger, so keeping that spend data complete and current is the organization's work.
See what your stack is actually being used for. Levos sits above the tools your teams already run and reports utilization, duplicates and abandoned licenses by team and role. Request a Demo to see it against your own connected tools.
Levos is accepting design partner applications from US organizations of 150 employees or more with an active AI rollout. Start your 90-Day AI Impact Audit.
