Published on the 04/03/2026 | Written by Heather Wright

AI overage and zombies driving budget burn…

Enterprises are confronting a dual financial pressure in their SaaS software stacks, with a surge in applications exceeding negotiated contract ceilings and an equally costly problem of over-purchased licenses going unused.

Both trends reflect a deeper shift underway in enterprise IT. Despite predictions that AI would streamline or consolidate technology portfolios, the 2026 SaaS Benchmark Report from Torii shows the opposite has occurred with AI adoption accelerating software sprawl, expanding shadow IT and pushing usage far beyond what governance and procurement teams can monitor or control.

“The applications most likely to exceed their contracted spend limits are disproportionately AI-powered.”

The report found the average enterprise now runs more than 830 applications, with over 61 percent operating outside formal IT oversight, creating the perfect environment for overspend to flourish and highlighting rather than scaling neatly, stacks compound – especially in the long tail of apps that never touch procurement.

“Most benchmarks undercount apps because they only look at what procurement touched. The majority of apps show up elsewhere – browser activity, direct signups and usage than never becomes a purchase order,” the report says.

 More than half of the top shadow IT apps are pure-play AI tools, with nearly 700 new AI applications entering the enterprise environments in the past year.

“These tools didn’t replace legacy software, they layered on top of it, accelerating sprawl while quietly redefining productivity.”

And while AI didn’t create shadow IT, it has dramatically increased its speed and blast radius, Torii says. “These tools connect deeply, gain broad access instantly and often persist long after teams stop using them.”

The report’s data on overage frequency reveals that the applications most likely to exceed their contracted spend limits are disproportionately AI-powered. ChatGPT tops the list, exceeding its contracted spend limits more than 35 percent of the time. It is followed closely by productivity and project management platform ClickUp, Keeper Security and workflow automation tool n8n, all with overage 33 percent of the time.

Zoom, long embedded in enterprise environments, exceeds its contract 30 percent of the time, while data movement platform Fivetran and Claude AI also generate overages at notably high rates. Snowflake, a data platform with consumption-based billing, exceeds contracted amounts 22 percent of the time, while tools such as OpenAI, Cursor AI, Anthropic, ‘AI-powered’ unified IT and security platform Iru round out the list.

The overage intensity data shows a sharp divergence in how severely different applications exceed their contracted financial limits, underscoring how a small subset of – largely AI – tools can disproportionately damage enterprise budgets.

While most create only modest overruns – such as n8n at 0.3 percent, Snowflake at 1.7 percent, Claude AI at 2.8 percent and Iru at 3.5 percent – others are leaving significant scorch marks on the budget. Cursor AI is the biggest offender, with a median overage of 20.5 percent. ChatGPT follows at 13 percent, with Anthropic at nine percent.

The report notes much of the variance which leads to applications consistently exceeding contracted ceilings is directly attributable to consumption-based pricing.

The central theme connecting these applications becomes clear in the context of Torii’s broader findings. The report notes AI-first tools have become a ‘dominant source of individual software use’, adopted rapidly and often without formal approval as employees search for productivity advantages. Because many of these tools rely on usage-based billing tied to tokens, compute cycles, queries or automated workflows, their consumption can scale dramatically with even modest increases in activity. At the same time, AI tools frequently rely on OAuth-based permissions and instant integrations that connect directly to corporate data.

As AI tools generate more data and trigger automated processes, downstream platforms are also experiencing heightened load from rapidly expanding data flows. Meanwhile, even established tools like Zoom and ClickUp exhibit a mix of entitlement drift, auto-provisioned features and inconsistent de-provisioning practices that lead to unexpected consumption growth.

Torii’s analysis shows as organisations run ever-larger portfolios, traditional controls break down. Employees interact with an average of 40 applications each day – productivity when managed, risk and waste when it isn’t.

But there’s also a parallel problem, with widespread overbuying, as shown by license utilisation data showing where unused licenses and stale access are creating a SaaS landscape littered with inactive accounts, unused licences and lingering access – with spend and risk quietly piling up.

Salesforce sits at the top with as the most overbought app with a staggering 55 percent non-utilisation rate, meaning more than half of purchased licenses are going unused. PagerDuty follows at 45 percent with Monday.com at 40 percent and Iru at 39 percent.

Well established enterprise tools including Zendesk (35 percent), 1Password (33 percent), Zoom (32 percent), Adobe (30 percent), Atlassian (28 percent), DocuSign (27 percent), Calendly (25 percent) and Slack (24 percent) all show double-digital non-utilisation.

The findings underscore a chronic imbalance in enterprise purchasing as the desire to ensure employees have access to tools they might need leads organisations to over-license, even as actual usage patterns remain uneven and unpredictable.

Meanwhile, 2.5 percent of license seats on average are assigned to offboarded users in ‘identity decay’, with the zombie accounts providing a significant security surface area and wasted budget.

The presence of Zoom on both lists – exceeding contracted ceilings 30 percent of the time while simultaneously leaving 32 percent of its purchased licenses unused, exemplifies how chaotic enterprise software management has become.

“Software adoption no longer follows centralised or predictable paths,” Torii says. “Our data shows that governance gaps aren’t the result of missing policies, but of speed. Applications, especially AI tools, are being adopted faster than traditional procurement and identity controls were designed to handle.”

The answer, Torii says, is governance models built for continuous discovery, not annual reviews.