AI Spend Management: What the Manual License Audit Misses

Picture an analyst who has run the annual license audit for six years. This is someone who knows which team over-orders design seats, which reseller quietly adds a support uplift, and which renewal always lands in the same busy week of March. The work is thorough and the instincts are sharp. Even so, the audit will miss the fact that the company is paying well above market for a commodity tool, because that number doesn’t sit in any contract on the review list. It sits in the comparison, and a benchmark finds it in seconds.

That gap is where AI spend management earns its keep. It also exposes a flaw in how most finance teams judge their audit process. Leaders count findings: tickets raised, licenses reviewed, renewals touched. Count isn’t the right way to measure an audit method. Average value per finding is.

The stakes have grown quickly. Median SaaS spend has climbed to roughly $9,455 per employee a year, close to double the level of a year earlier, and around 36% of licenses sit unused against recommended utilization levels. Software is now the second-largest operating line for many companies after payroll. An audit method that spends its hours on the wrong items is no longer a minor inefficiency. It’s a visible drag on EBITDA.

How manual “Optimise” tickets pile up, one seat at a time

A manual SaaS license audit works the way most people read a stack of contracts: top to bottom, one vendor at a time. The analyst opens the file, checks purchased licenses against assigned ones, notes the renewal date and raises a ticket. Then the next file. Then the next.

Each ticket is reasonable on its own. Together they produce a long tracker of small items, and the justification fields tell the story. Across manually raised “Optimise” tickets in Saasrooms’ European instance, which spans agritech software, animal nutrition manufacturing, digital services and professional firms, the most common notes are a single word: “Optimize,” “renewal,” “license review.” Nothing is wrong with those tickets. They’re just low resolution.

The savings attached to them are low resolution too. Of 60 manual Optimise tickets in that European data set, 19 carry an estimated saving of one unit of currency or less. Some are placeholders for meeting notes. Others are genuine reviews of a design tool, a stock footage subscription or a PDF editor, where the best realistic outcome is a two-figure saving.

There’s a structural reason this happens. Business units now control around 81% of SaaS spend directly, while IT manages roughly 15%. The analyst works from the contract list that’s visible to IT, and that list is mostly the small, centrally bought tools. The large, decentralized commitments sit elsewhere.

Sixty open optimization items on a board slide reads like rigor. It signals activity, and activity is easy to mistake for SaaS spend visibility. Meanwhile, the handful of findings that would actually move EBITDA wait in the same queue as a design tool line worth single digits.

Manual review is built to find what’s inside a contract. The expensive problems usually live between contracts.

Fewer findings, far more value per finding

To keep the comparison honest, the figures below exclude placeholder items worth one unit of currency or less, and they cover more than two dozen companies across Saasrooms’ US and European instances.

The sharpest case is a US healthcare revenue cycle services company. Its team raised 28 manual Optimise tickets worth about $67,000 in total, roughly $2,400 each. Over the same period, the Saasrooms insight engine flagged 12 findings worth about $692,000, an average close to $58,000. One cloud commitment worth $400,000 sits inside that total. Take it out and the remaining 11 findings still average more than $26,500 each, around eleven times the manual figure.

Across the full US instance, which also includes a pharmaceutical contract manufacturer, an online print and marketing products business and a multi-location restaurant group, the pattern holds. Sixty-one manual tickets average about $8,200. Fifty-nine AI-flagged findings average about $22,500, close to three times as much per item. At the median, an AI finding is worth $4,700 against $3,000 for a manual one.

Europe shows the gap from the other direction. A Scottish animal nutrition manufacturer logged six manual tickets worth about £33,000 combined. Two AI findings came to £30,400, nearly the same total from a third of the items. A Spanish agritech software company raised six manual tickets averaging about €2,400 each, while its single AI finding was worth €10,000. Across the European instance, 23 AI findings identified about 91% of the value of 41 manual tickets, at an average roughly 1.6 times higher.

Put the regions together and the result is consistent: fewer items, and between 1.5 and 3 times the value per item.

Identified is not the same as banked

Any CFO will ask the next question, and it’s the right one. Do these findings actually get acted on?

The pipeline data offers a clear early signal. In the US instance, 83% of the value attached to AI-flagged findings has already moved past the backlog into scoping, sourcing, contracting or completion. For manual tickets, the figure is 40%. In Europe, the gap is narrower but still meaningful: 87% against 69%.

Completed items tell a more nuanced story, and it’s worth stating plainly. In Europe, seven completed AI findings carry about 92,000 in savings, more than the 26 completed manual tickets combined, 18 of which were worth one unit of currency or less. In the US, manual tickets have closed more value so far. The reason is size. The largest AI findings involve cloud commitments, CRM renewals and multi-year agreements, which take quarters to negotiate rather than weeks. They’re further along than the manual backlog, but they aren’t finished yet.

That’s the honest picture: higher value per finding, faster movement into active work, and a longer cycle to close the biggest items.

Why anomaly detection sees what contract-by-contract review can’t

The difference isn’t effort or expertise. It’s vantage point.

A person reviewing contracts one at a time judges each agreement against itself. Is the seat count right? Is the renewal date logged? Did the price go up? Anomaly detection judges each line against everything else: the rest of the estate, the company’s own history and market benchmarks drawn from comparable buyers.

Consider the online print and marketing products business, running 22 active internet service providers across its production and office sites. Every one of those contracts looked unremarkable in isolation. The anomaly was the number 22. The same company’s managed cloud services agreement, worth more than $2.7 million a year, surfaced as its single largest opportunity once it was compared against market rates.

The multi-location restaurant group offers a second example. It was paying separately for an identity platform while its existing Microsoft 365 tier already included equivalent identity capability, which put more than half of that tool’s annual cost on the table. In the same estate, standalone event booking and marketing tools duplicated functions the group’s CRM already covered, with savings estimated at between 46% and 82% of those lines.

Acquisitions create their own version of the problem. At the healthcare revenue cycle company, the insight engine flagged multiple HR and finance system contracts inherited from separate acquisitions. Each was correct on paper. None had been consolidated. For PE-backed businesses running buy-and-build strategies, this is one of the most common and most expensive blind spots in post-deal integration.

Pricing drift is the anomaly that compounds

SaaS inflation reached about 16% in mid-2026, more than five times general consumer inflation, and nearly four in five IT leaders saw a price increase at their most recent renewal. Many of those increases are quiet: a CPI clause, a withdrawn volume discount, a legacy tier retired in favor of a pricier AI-inclusive one.

At a UK multi-site business with eight locations, the insight engine noted that a print management vendor had raised its per-user price the previous year on a tool with only around 17 live users. The saving is modest. The pattern isn’t. A price rise accepted at one renewal becomes the baseline for the next.

This is where timing becomes a financial variable. Saasrooms flags the pricing anomaly the month it appears, not the quarter the renewal notice lands, leaving room to negotiate before an auto-renewal locks the new rate in. The payoff is well documented: buyers who walk into renewals with benchmark data consistently land well below their contractual price caps, while buyers who don’t often end up above them. A manual audit typically meets the same line item on its annual cycle, frequently after the renegotiation window has closed.

Where the manual audit still earns its place

None of this makes the experienced analyst redundant. It changes where the analyst’s week goes.

AI-flagged findings still need human judgment before they become recovered savings. At the UK multi-site business, one finding pointed to an IT support contract with 130 committed seats against 116 the year before, then noted that removing seats mid-term would trigger a minimum-commitment clause. The anomaly was real. Whether to act now or at term needed someone who understood the relationship, the contract and the business plan.

Other findings carry similar instructions: validate the count, confirm the figure in writing, check the next invoice for a residual charge. That’s the manual audit’s real strength. People know which vendor is strategic, which department is mid-migration and which renewal is tied to a product launch. Software doesn’t have that context.

The practical model is a division of labor. The insight engine does the wide scan and ranks what it finds by value. The analyst takes the top of that list and does what people do best: pressure-testing the finding, negotiating with the vendor and making the call.

Consider the cost side. If each low-value ticket takes even two hours to scope, chase and close, the 19 near-zero European tickets alone represent close to a full working week. That week could have gone to a single finding worth more than all of them combined.

What to change on Monday

For finance and IT leaders who want to act on this without a major program, four moves make the difference.

First, re-rank the existing optimization backlog by estimated value, not by renewal date or the order tickets were raised. Most teams find that a small share of items carries most of the money.

Second, set a materiality threshold. Items below it still get handled, but in a batch, on a schedule, rather than one at a time at the front of the queue.

Third, benchmark the 20 largest renewals due in the next 12 months before any vendor conversation starts. With double-digit price increases now routine, preparation time is worth more than negotiating skill.

Fourth, report identified and completed savings side by side to the board. A tracker that shows only ticket counts rewards volume. One that shows value per finding and conversion rate rewards the right work.

Volume fills the tracker

A long list of findings is easy to produce, and it looks good in a board pack. It isn’t what reaches EBITDA.

Measure the audit by what each finding is worth.