Eighty-six per cent of mid-market IT leaders say that managing AI complexity has actually increased their team's workload, according to new research from Freshworks’ survey of over 12,000 IT decision-makers. The finding cuts against one of the central promises of enterprise AI investment to reduce, not multiply, operational burden.
The report, titled ‘The Global Cost of Complexity: The Mid-Market AI Complexity Trap’, found the problem is particularly acute in the UK. Some 83% of British IT leaders reported that AI outputs regularly introduce noise, errors, or rework, also known as "AI slop." The cumulative cost of making AI functional, based on what Freshworks describes as “directional market-level estimates”, amounts to an estimated £11.7bn in wasted UK AI spend every year.
The complexity tax
On average, mid-market organisations lose 25% of their AI budget to what the report calls a "complexity tax", which includes integration troubleshooting, data wrangling, governance overhead, and rework that was never factored into original project plans. These are among the hidden costs that routinely go unaccounted for in AI programmes, and the Freshworks data puts a pound figure on their cumulative impact. A further 26% of IT teams' AI-related time is absorbed by managing that complexity rather than pursuing strategic outcomes.
The productivity paradox deepens when executive expectation is set against deployment reality. Almost three-quarters (72%) of mid-market executives expect AI investments to show ROI within eight months. Yet more than half of organisations surveyed report that deployment alone can take between six and twelve months before meaningful returns can even begin to be realised. In many cases, the ROI clock starts running before the systems are fully live.
How this compares with other research
The Freshworks findings sit in tension with a body of evidence pointing, selectively, to AI delivering genuine returns. Deloitte's 2025 survey of nearly 1,900 executives found that 85% of organisations had increased AI investment over the prior twelve months and that generative AI was already producing measurable productivity gains where programmes were structured effectively. The divergence between the two studies is not necessarily a contradiction. Deloitte's respondents skew toward larger enterprises, where AI budgets, integration capabilities, and governance maturity are typically more advanced.
McKinsey research, cited in the Freshworks report itself, adds a further dimension. Its 2024 global survey of AI adoption identifies a wide performance gap between organisations capturing significant value from AI and those that are not, with top-quartile companies generating 3.5 times more revenue impact than median performers. The differentiating factor, McKinsey uncovered, is not model selection or budget size but execution discipline relating to governed data, integrated systems, and a defined path from pilot to production.
The scale of that gap is reinforced by the IBM Institute for Business Value's 2025 CEO Study, which surveyed 2,000 CEOs globally and found that only 25% of AI initiatives had delivered expected ROI, with just 16% scaled enterprise-wide. Taken together, the three studies suggest that the gap between AI leaders and AI laggards is widening, and that mid-market organisations currently sit closer to the struggling end of that spectrum.
The governance gap
A consistent thread running through the Freshworks data is the absence of formal AI governance. Only 33% of mid-market organisations have a governance framework applied consistently across their AI programmes. A further 46% operate with a partial framework applied inconsistently. This enterprise category now runs an average of 4.2 AI tools, with 10% managing seven or more, yet without the structures needed to make that stack coherent or auditable. Separate Freshworks research revealed that more than seven in ten US mid-market IT leaders reported unapproved "shadow AI" use is common inside their organisations, compounding the governance challenge further.
"The tools are not the bottleneck"
Murali Swaminathan, Chief Technology Officer at Freshworks, views the issue as structural rather than technological: "For mid-market organisations, where budgets are already stretched, this represents meaningful lost capacity. Oftentimes governance gaps, not technology limitations, are cited as the primary barrier to going deeper. The tools are not the bottleneck. The absence of the structures needed to use them consistently is."
The data suggests a buying behaviour shift is already under way in response. Nine in ten mid-market leaders say they prefer AI solutions with built-in workflows over those requiring heavy configuration, and 54% are now purchasing AI capabilities from vendors rather than building in-house. Whether that approach is able to close the gap between executive expectation and operational reality remains a defining question for enterprise AI programmes.