Enterprises are spending billions on AI and operationalising almost none of it. Kudo Advisory is betting the fix was never technological

At least half of enterprise generative AI projects are abandoned after the proof-of-concept stage, according to Gartner, and MIT research published in 2025 put the share of pilots delivering no measurable return as high as 95%, a failure rate that has held even as global spending on AI-centric systems is forecast to pass $300 billion in 2026. Into that gap, Vijay Jaswal, Founder and CEO of Kudo Advisory, has launched a UAE-based firm on a deliberately unfashionable claim: that the reason so few pilots survive contact with the wider organisation has little to do with the models themselves.

Jaswal spent years on the vendor side at IFS and Software AG before concluding that the enterprise AI market did not need more technology. The diagnosis was built from watching well-funded initiatives stall in identical ways across telecommunications, manufacturing, energy, and aviation clients in multiple regions. “What struck me most was not a lack of ambition or technology, but the lack of structure around adoption, governance, enablement, and execution,” he said, describing organisations that left employees to self-learn with minimal guidance, likening it to “handing someone who has never driven before a high-performance car without proper training, rules of the road, or a clear destination.”

The most expensive failures look like successes on paper

The pattern that convinced him the firm was needed was not the visibly broken project but the one that photographed well. He described an energy generation utility that had invested heavily into AI proofs-of-concept that looked innovative until closer inspection revealed teams using different tools, sensitive operational data copied into external platforms without governance, and no agreed ownership for scaling, while employees manually copied outputs into spreadsheets and emailed them between departments because the underlying systems stayed disconnected. “The market does not need more AI hype. It needs practical guidance, governance, prioritisation, and disciplined execution that helps organisations turn AI investment into measurable outcomes rather than disconnected experimentation and dead-end pilots,” he said.

That diagnosis aligns with the wider research. RAND Corporation found that more than 80% of enterprise AI projects fail to deliver promised business value, the cause lying not in model performance but in missing data infrastructure, absent change management, and governance never built. McKinsey has reported that roughly 62% of organisations are experimenting with AI agents while only around 23% have scaled them into production, a distance that maps onto the breakdowns Jaswal described.

Having sold the platforms, Jaswal is direct about their limits. “Many organisations begin to view AI as a silver bullet that will somehow solve deep operational or process problems on its own,” he said, arguing that most already have capable platforms and that the harder work is understanding the root causes of operational breakdowns first. The warning sign he watches for is when spreadsheets become the glue holding a process together and are constantly emailed around an organisation, which signals disconnected systems, manual workarounds, and dependency on individuals rather than structured processes.

Why the pilots die between the demo and the organisation

On why pilots never move beyond proof-of-concept, Jaswal returned to the sequence. “Organisations often start with the technology before fully understanding the operational problem they are trying to solve,” he said, adding that many are launched out of excitement or pressure to show quick progress, without clear ownership or a roadmap for scaling. He recalled a telecommunications client whose customer-service pilot demonstrated impressively but sat on data fragmented across disconnected systems, with frontline employees never involved in designing it, so the process was already broken before any AI was introduced, and the pilot never scaled despite significant investment.

These are operational transformation initiatives misclassified as technology exercises, he insists, requiring process alignment, governance, data readiness, adoption, and executive sponsorship to succeed. Asked whether the UAE's problem is fundamentally about technology or change management, he was unequivocal that it is far more the latter.

However, who is responsible for fragmented AI initiatives, Jaswal declined to blame a single department, locating the cause in AI adoption moving faster than governance and leadership alignment can keep pace with. He compared it to the early days of cloud software, when departments signed up for their own tools on company credit cards faster than procurement could move, except that AI is evolving the same way, far faster.

The remedy he argues for is earlier intervention than most organisations accept. “It is much easier and cheaper, and makes more sense, to involve an architect before a building is constructed than trying to redesign the structure once multiple floors have already been added in different directions,” he said, careful not to call those now seeking help too late.

A defined-outcome pitch into a market built on billable hours

What makes the timing interesting is who Jaswal is choosing to compete against. The UAE's artificial intelligence market was valued at around $7.4 billion in 2025 by some estimates and is forecast to grow above 45% annually through the next decade, and the advisory layer above it is dominated by firms moving into the same territory. The Big Four and the major strategy houses have committed more than $10 billion to AI initiatives since 2023, with Accenture alone reporting around 77,000 AI professionals and tripling its generative AI revenue to $2.7 billion in fiscal 2025, running engagements at $250 to $500 an hour over six to 18 months.

Against that scale, his positioning rests on an argument about incentives, contrasting a model in which longer engagements are more profitable with a defined-outcome approach where success criteria are agreed before any proof-of-concept begins.

“A successful AI outcome should never be defined as simply deploying AI,” he said, describing outcomes that vary by organisation, from reducing manual processing time by 40% to improving forecasting accuracy or reducing revenue leakage, each set against a baseline before work starts. He resists treating AI as a boost over unresolved problems, arguing it “should not be viewed as a magic engine upgrade bolted onto a car with underlying mechanical problems.”

Whether a boutique firm can hold that discipline against competitors with managed-services revenue in the billions is the open question the market has not answered, given how few engagements have crossed from pilot to scaled production.