84 Percent of Tech Leaders Say Legacy Systems Have Already Killed an AI Pilot at Their Company
Digital Transformation

84 Percent of Tech Leaders Say Legacy Systems Have Already Killed an AI Pilot at Their Company

A new survey of 945 CIOs and CTOs at companies with at least 500 million dollars in revenue shows legacy infrastructure, not model quality, is the most common reason enterprise AI pilots get canceled. The research also surfaces a less comfortable finding: widespread worry that some companies are using AI as cover for workforce cuts aimed at the stock price.

PublishedNovember 22, 2026
Read time6 min read
Share

The number that should reframe how you plan your next AI pilot

GFT Technologies, working with Wakefield Research, surveyed 945 CIOs and CTOs across 19 countries at companies with at least 500 million dollars in annual revenue, and the headline finding cuts against a common assumption about why enterprise AI pilots fail. Rather than model quality, data availability, or talent gaps topping the list, 84 percent of respondents say legacy system limitations have already caused their organization to cancel an AI pilot, a figure large enough to describe the median experience rather than an edge case.

That statistic matters because it reframes where due diligence should happen before an AI pilot even launches. An organization scoping a new pilot around model selection and training data without first auditing whether the surrounding infrastructure can actually support production deployment is, on this evidence, more likely than not heading toward the same cancellation outcome the large majority of this survey's respondents have already experienced firsthand, often only after sinking real budget and months of work into the pilot first.

The anxiety sitting just beneath the investment numbers

Beyond the legacy infrastructure finding, the survey captures a broader unease about whether AI spending is outrunning its own payoff. 89 percent of tech leaders globally worry that AI investment is growing faster than the business value it can realistically deliver, and that concern is not evenly distributed: 92.1 percent of US executives share it, compared with 80.6 percent in EMEA, a gap worth noting for any global organization trying to set a single AI governance standard across regions with meaningfully different risk appetites.

GFT USA CEO Rishi Chohan distilled the structural problem behind both findings in a single observation worth repeating directly to your own leadership team: every enterprise he talks to has an AI roadmap, but fewer have an infrastructure roadmap built to match it. That gap between ambition and the underlying plumbing required to support it is precisely what shows up later as a canceled pilot or an unrealized return, and it is a gap most organizations only discover once real money and real deadlines are already on the line.

The security angle that connects infrastructure debt to a different budget line

93 percent of respondents agree that running AI applications on outdated systems introduces security risks, a near-consensus finding that should move legacy modernization conversations out of the pure cost-center framing many organizations still apply to them. If your security team is separately flagging AI-related risk while your infrastructure team treats legacy system replacement as a deferred nice-to-have, this survey suggests those two conversations are actually describing the same underlying problem from different vantage points.

That connection gives CIOs a genuinely useful argument for budget conversations that have historically been difficult to win on infrastructure modernization alone. Framing legacy system replacement explicitly as AI security risk reduction, rather than as generic technical debt paydown, aligns the business case with where boards and CISOs are already paying close attention, and this survey's near-unanimous agreement among peer organizations adds real weight to that framing the next time a modernization budget request reaches committee review.

The uncomfortable finding about AI as cover for workforce decisions

The survey surfaces a finding that goes beyond infrastructure entirely: 93.3 percent of US respondents believe some public companies use AI as cover for workforce changes intended primarily to boost their share price, rather than changes driven by genuine productivity gains from AI deployment. Jason Droege, interim CEO of Scale AI, echoed this concern directly, describing companies using the technology as an excuse to remove headcount rather than as the actual cause of the reduction.

That skepticism has a direct bearing on workforce trust and morale inside any organization currently attributing layoffs to AI efficiency gains. If the large majority of tech leaders themselves suspect this framing is sometimes a pretext, employees inside those organizations are likely applying the same skepticism, which makes transparent, specific communication about what AI actually changed in a given role considerably more important than a generic AI-efficiency talking point in a layoff announcement.

The personal exposure angle tech leaders are quietly worried about

The survey also found that 47.0 percent of US respondents are very or extremely concerned that a wrong workforce decision made while scaling AI could put their own job at risk, compared with 42.9 percent globally. That is a meaningful share of the technology leaders responsible for making these calls expressing direct personal anxiety about the consequences of getting an AI-driven workforce decision wrong, which is a different and more consequential signal than general skepticism about AI value.

That personal stake is worth factoring into how your own organization structures AI-related workforce decisions. A decision-maker who is personally worried about career exposure from a wrong call has an incentive to over-document justification, seek broader sign-off, or delay a decision longer than the business case alone would suggest, dynamics that can quietly slow legitimate AI-driven restructuring even when the underlying business case is sound and the data genuinely supports moving faster than caution allows.

What this means for your next AI pilot and your next workforce decision

Before greenlighting your next AI pilot, run an honest infrastructure audit specifically against the question this survey raises: does the system this pilot will eventually need to run on in production actually support it, or is this pilot likely to join the 84 percent that get canceled over exactly that gap. GFT's Chohan put it directly, AI cannot simply be layered onto legacy systems, and that is a planning constraint worth confirming before investing in model selection and pilot design rather than discovering after a successful proof of concept stalls at the production stage.

On the workforce side, if your organization is contemplating AI-attributed headcount changes, build your internal and external communication around specific, verifiable productivity changes tied to the role in question rather than a general AI efficiency narrative. Given how widespread the skepticism this survey documents already is among your own peers, a vague AI justification for a workforce reduction is more likely to be read as cover than as candor, regardless of whether it happens to be true in your specific case.

Tagged#news#digital-transformation#enterprise#cio#erp#strategy#governance#gft-technologies#wakefield-research#legacy-systems#ai-pilot-cancellation#ai-workforce-trust#infrastructure-roadmap