A UAE infrastructure firm gets its AI business case approved, selects a provider, builds a phased implementation roadmap, and launches a pilot. Six months later, the pilot is still a pilot. The predictive maintenance dashboard sits open in one browser tab while engineers keep working from the spreadsheet they have used for years. The project did not fail because the model was wrong. It failed because nobody managed the change that came with it.
That gap, between an AI deployment that works technically and a workforce that actually uses it day to day, is one of the most common and most avoidable reasons UAE infrastructure firms do not see a return on AI investment. This piece looks at why AI implementation projects stall after approval, what UAE specific data says about the scale of that resistance, what the UAE's own large scale AI rollout gets right about managing it, and a practical change management framework a construction, energy, or utilities firm can use to move a pilot into daily use.
Why AI Implementation Projects Stall After Approval
The UAE is not unusual here. Globally, roughly 95 percent of enterprise generative AI pilots deliver no measurable profit and loss impact, according to MIT NANDA's State of AI in Business 2025 study, which drew on a review of 300 public AI deployments, structured interviews with 52 organizations, and a survey of 153 senior leaders conducted between January and June 2025. Only about 5 percent of pilots produced a fast, measurable return. The report's central finding was not that the underlying models underperform. MIT calls the real issue a learning gap: the inability of an organization to fold an AI tool into its actual workflows, feedback loops, and decision structures so it keeps improving rather than sitting idle after launch.
Deloitte's most recent enterprise survey, covering 3,235 business and IT leaders across 24 countries between August and September 2025, points at the same gap from a workforce angle. Leaders surveyed named insufficient worker skills as the single biggest barrier to integrating AI into existing workflows, ahead of budget, data quality, or vendor selection, according to Deloitte's State of AI in the Enterprise 2026 report. In response, 53 percent of organizations are now investing in broad workforce AI literacy and 48 percent in structured upskilling, but far fewer, just 33 percent, are actually redesigning roles and career paths around the new way of working, which is usually the step that determines whether a tool sticks.
AI Resistance Is Now the Top Workforce Risk for UAE Firms
Inside the UAE specifically, employee resistance to AI has overtaken cybersecurity skills shortages, labour shortages, and regulatory change to become the leading workforce risk organizations expect to face over the next one to two years, according to Marsh's 2026 People Risks report, which surveyed 103 HR and risk professionals in the UAE as part of a wider poll of 4,500 professionals across 26 markets. Only 40 percent of UAE firms reported full collaboration between HR and risk teams on managing that risk, with another 40 percent describing it as partial and 20 percent as minimal.
Construction and infrastructure feel this more acutely than most sectors. The UAE government's own agentic AI push gives a sense of scale: the government plans to shift 50 percent of federal sectors to agentic AI within two years and is training 80,000 federal employees to make that transition, according to a July 2026 Gulf News analysis of construction's shift to agentic AI. The same piece notes that many construction professionals, particularly veteran staff who have spent decades on traditional methods, lack the digital skills to use new tools effectively, and argues the industry's AI transformation will be won or lost on the human side of the rollout, not the technology.
What the UAE's Own AI Rollout Gets Right
For a private infrastructure firm wondering how a country moves an entire government onto AI without the same failure rate research shows for private enterprise pilots, INSEAD research published in May 2026 on the UAE's public sector AI adoption offers a useful, if unglamorous, answer. Comparing the UAE against the UK, Singapore, the US, the EU, and China, the researchers found that similar technology and budgets produced markedly different outcomes depending on three institutional choices: sustained, concentrated leadership commitment over years rather than a single mandate, domain level redesign of how processes actually work rather than a portfolio of disconnected pilots, and treating procurement and partnerships as strategic tools rather than a back office function.
The UAE's TAMM platform, which now hosts more than 1,000 government services with AI capabilities built in, and Dubai's initiative that narrowed 183 candidate AI use cases down to 15 high impact deployments across mobility, healthcare, logistics, and urban infrastructure, are both products of that approach. The INSEAD researchers found execution failures came from data fragmentation across agencies, talent gaps at the intersection of policy and technology, and governance frameworks that lagged behind what was actually being deployed, not from weak model performance. Their core recommendation, treating AI as public infrastructure rather than an accumulation of pilots that never scale, applies just as directly to a private construction or utilities firm trying to move past its first pilot.
A Change Management Framework for Infrastructure Firms
1. Assign an Executive Sponsor Who Stays Past Kickoff
Most infrastructure firms assign a sponsor to get an AI project funded, then let that sponsor move on once the pilot launches. MIT's research on the learning gap and the UAE's own institutional data both point the same way: it is sustained leadership attention, not a single approval decision, that determines whether a deployment scales. The sponsor's job does not end at launch. It continues through the first messy months when adoption is lowest and easiest to abandon.
2. Redesign Roles at the Task Level, Not the Department Level
Blanket department wide rollouts create the most resistance because they ask everyone to change everything at once. A better approach, examined in more detail in our piece on the UAE's AI talent gap, is to look at each role task by task and decide what gets automated, what gets AI assisted, and what stays entirely human. The World Economic Forum's Future of Jobs Report 2025 names analytical thinking, resilience, leadership, and collaboration as the skills that stay most valuable through 2030, which is a useful filter for deciding what not to hand to a model.
3. Train on the Job, Not in a Single Workshop
A one time onboarding session at launch is the most common change management shortcut, and it is also the least effective one. Training that works is role specific and embedded in live, low risk project work rather than delivered as a generic course before anyone has touched the tool. Engineers, planners, and site staff retain far more when they learn a new workflow on an actual project than when they sit through a slide deck about it in a conference room.
4. Build Internal Champions and a Real Feedback Loop
Peer influence moves adoption faster than a top down mandate. Identifying two or three early, credible users inside the working team, and giving them time to coach colleagues, does more for adoption than another email from leadership. Our change management guide for UAE SMBs covers champion selection and role specific training tactics in more depth for smaller teams, and the same principles scale up directly to a larger infrastructure firm rolling out AI function by function.
5. Make Governance and Adoption a Cross-Functional Job
Leaving AI adoption entirely to IT is one of the more reliable ways to stall it. The firms that manage this well pull together engineers or operations staff, HR, safety or legal, and data or IT specialists into one working group rather than routing every decision through a single department. Cross functional ownership also creates the feedback channel needed to catch problems, like a workflow that technically works but nobody trusts, before they harden into permanent resistance.
A 90-Day Rollout Plan
- Weeks 1 to 2: Confirm executive sponsorship, name a working team of 8 to 12 people from the function the AI tool touches first, and set two or three measurable adoption targets, not just usage targets.
- Weeks 3 to 6: Run role specific, on the job training inside a live but low risk project, not a generic workshop, and open a standing feedback channel to the implementation team.
- Weeks 7 to 10: Expand from the working team to the full function, using early champions to train peers, and start reporting adoption alongside the original business case metrics.
- Weeks 11 to 13: Review what actually changed in the workflow, not just login counts, fix or retire whatever is not being used, and decide whether to scale to the next function.
Common Change Management Mistakes to Avoid
- Treating training as a single onboarding session instead of ongoing, role specific coaching.
- Rolling out to an entire department at once instead of proving the workflow with one working team first.
- Leaving change management entirely to IT, with no operations, HR, or frontline representation.
- Measuring success by logins or licenses issued instead of whether the old manual workflow actually stopped.
- Losing executive sponsorship after the pilot phase, exactly when adoption pressure is highest.
The 95 percent failure rate MIT documented globally, and the fact that AI resistance is now UAE firms' top workforce risk, both point to the same conclusion: getting an AI business case approved and a vendor selected is the easier half of implementation. The UAE's own public sector rollout shows the harder half, sustained leadership, redesigned workflows, and real training, is what actually determines whether a pilot becomes how the organization works. Infrastructure firms that treat change management as a line item alongside the technology budget, rather than an afterthought once the system goes live, are the ones that end up in the 5 percent instead of the 95.