Zylo counts 78% of IT leaders hit by unexpected AI and consumption charges

Take the two figures in the index that barely moved and divide them. Average annual SaaS spend of $55.7M ^[6] across an average portfolio of 305 applications ^[7] is about $182,600 per application a year ^[15]. Back out the 8% increase and last year's $51.6M ^[16] across a portfolio 0.07% larger ^[7] comes to roughly $169,000 per app. The same number of tools costs about $13,600 more each ^[17].

The 108% and the 393% are not describing that increase. They describe spend on AI-native applications, which Zylo defines as apps where AI is core to the product ^[2], with large enterprise spend on them up 393% in a single year ^[3]. Someone signed for those. A new AI-native vendor arrives with a contract, an internal champion and a budget line, and the purchase is visible while it happens. Zylo did not publish the base dollar figures behind either percentage, or say how it defines a large enterprise.

The mid-contract part sits in the survey numbers instead. Over the past year, 78% of IT leaders reported unexpected charges tied to AI features or consumption-based pricing ^[8], and 61% cut projects because of unplanned SaaS cost increases ^[9]. Zylo says AI add-ons and usage-based tiers are reshaping cost structures mid-contract, making spend harder to predict and harder to control ^[10], and that AI shows up inside existing tools long before governance teams identify it ^[11]. A charge like that is a different problem from buying a new tool, and it lands on a budget somebody already defended.

The supply side explains the timing. In High Alpha's 2025 SaaS Benchmarks Report, 64% of SaaS companies embed AI as a supporting feature and 36% say AI is core to the product ^[5], two shares that together account for the whole sample ^[18]. High Alpha also puts the share that have launched or plan to launch AI features at 92% ^[5].

The index is Zylo's own, and the fix it names is the category in its title. "Without a disciplined SaaS Management program, organizations will fall behind," the report says ^[12]. It also reports ITAM and FinOps teams moving closer together, both working from usage data and a shared system of record ^[13].

For the twenty contracts carrying most of that $55.7M, two properties decide how much trouble each one can cause. The first is whether the bill can increase without a new signature. The second is whether anyone on the buyer's side can see the consumption before it is billed. Variable and metered is forecastable, and you can set a threshold. Fixed and unmetered is quiet until renewal, when the AI tier is presented as the new baseline. Variable and unmetered is the quadrant that produced the 78%, and it is worth knowing which of your vendors are in it before the next true-up rather than after.