- A 2026 survey found 48% of executives considered enterprise AI adoption a “massive disappointment,” up from 34% the prior year, with integration costs and data quality the most cited causes.
- A June 2026 report found 89% of enterprise AI agents never reached production, with reliability, governance and audit trail requirements as the primary blockers.
- Custom multi-system AI platforms with fine-tuned models cost $500,000 to over $1 million and take 12-18 months to build, making the “plug-and-play” premise of 2023 vendor pitches largely fiction.
A July 2026 report found that despite nearly eight in ten companies using generative AI, just as many reported no measurable bottom-line impact from their initiatives. That finding, combined with a wave of Q2 and Q3 2026 retrospectives, paints a consistent picture: the enterprise AI bets of 2023 were expensive, slower than promised and harder to govern than vendors suggested.
The ROI Gap
Enterprise generative AI software spending was projected to climb from $1.7 billion in 2023 to $37 billion in 2025, and budgets moved accordingly. The returns did not. A 2026 survey found 48% of executives called AI adoption a “massive disappointment,” up from 34% the prior year.
The disappointment was structural. Initial business cases routinely excluded the cost drivers that dominate actual deployments: inference at scale, governance frameworks, continuous evaluation and change management. An IBM report found that computing costs were expected to rise 89% between 2023 and 2025, with 70% of surveyed executives attributing the increase to generative AI workloads. Every executive in that survey reported cancelling or postponing at least one generative AI project on cost grounds. A May 2026 analysis from Nomtek found that roughly 95% of AI pilots show no profit and loss impact and only 39% of firms report any EBIT contribution from AI. For those tracking why enterprise GenAI pilots fail to deliver ROI the cost picture is the most consistent thread.
Agents Stuck in Pilots
A June 2026 report titled “The 11 Percent: Why 89% of Enterprise AI Agents Never Reach Production” found that while 79% of enterprises had adopted AI agents in some form, only 11% ran them in production. The 68-percentage-point gap between adoption and production reflects the same friction points that blocked deployment in 2023: reliability failures in edge cases, unpredictable behaviour, and the audit trail requirements that governance teams demanded before sign-off.
A Caylent survey from August 2026 reports that 59.5% of senior enterprise leaders now run AI agents autonomously in production, though with strict governance conditions attached. That uptick, if accurate, suggests progress, but it also confirms that the infrastructure and oversight absent in 2023 are now table stakes rather than optional. The agent framework choices enterprises make in 2026 reflect hard lessons from that earlier failure rate.
No Easy Button
The “plug-and-play” pitch was the dominant vendor positioning in 2023. It did not hold. A July 2026 report found that custom, multi-system platforms with fine-tuned models and agentic workflows cost $500,000 to over $1 million and take 12-18 months to build. Mid-complexity integrations for mid-sized enterprises typically ran $40,000 to $250,000, with timelines of six to 18 months.
Beyond the headline figures, budget overruns were routine. Cleaning unstructured data, retrofitting legacy APIs and training staff frequently added 30% to 50% on top of initial vendor quotes. Data governance was the deeper problem: without consistent, accessible data, models produced outputs teams could not trust, which stalled adoption further. The gap between a vendor demo and a working enterprise deployment turned out to be months of internal engineering work that few 2023 business cases had budgeted for. Contract terms in SaaS AI deals compounded the exposure for teams that moved fast.
Structural Drag Persists
Legacy infrastructure was not a solvable 2023 problem. Many enterprises ran on systems not designed for real-time data or AI workloads, and that mismatch converted quick pilots into months of rework. Worldwide AI spending is projected to reach roughly $2.59 trillion in 2026, up about 47% year over year, but forecasting that spend at the programme level remains difficult given its multi-source, usage-driven structure.
The talent gap compounded the infrastructure problem. A March 2026 analysis found insufficient workforce skills a major barrier to integrating AI into workflows, with only a minority of organisations making significant changes to talent strategy. The consistent picture across 2023 deployments is that the blockers were organisational as much as technical: change resistance, unclear ownership and governance gaps that no model release could fix. Board-level pressure for proof of ROI is now forcing the strategic discipline that 2023’s hype cycle skipped.
