Why AI Pilots Struggle to Scale: Debunking the 95% Failure Myth

November 12, 2025
By Rebecca Abrahams
7 minute read

Key takeaways:

MIT's widely cited claims that 95% of AI pilots fail is flawed research based on a sample of just 52 interviews
Actual failure rates sit between 40-80%: the challenge isn't that AI doesn't work, it's that organisations struggle to scale pilots into enterprise-wide initiatives
Four underlying patterns explain why promising pilots stall: inadequate infrastructure, overlooked change management, unclear ROI frameworks and restrictive governance

MIT’s State of AI in Business 2025 report claimed that 95% of AI pilots are failing, creating considerable uncertainty for organisations investing in artificial intelligence. The claim has been circulating widely, but the reality is more nuanced—and more hopeful—than headlines suggest.

On 7 October, 30+ senior leaders joined Sullivan & Stanley's AI Lead Archie Cobb for a virtual session exploring why promising AI pilots struggle to become enterprise-wide initiatives.

Debunking the 95% failure claim

The underlying research was based on just 52 executive interviews, with success defined narrowly as ROI within six months. This timeframe is far too short for meaningful enterprise transformation. The methodology admitted selection bias and inconsistent success definitions.

Broader industry research from Gartner and McKinsey paints a more accurate picture: somewhere between 40-80% of AI pilots stall before becoming scaled enterprise initiatives. Still concerning, but it shifts the conversation from "AI doesn't work" to "how do we scale AI effectively?"

Understanding these nuances helps organisations set realistic expectations for their transformation journey.

Where organisations are getting stuck

We asked participants to identify their single biggest blocker moving from pilot to scaled implementation. The responses clustered around several key themes.

Cost and perceived value challenges

Many organisations struggle to demonstrate clear ROI from AI pilots, particularly when they remain disconnected from core business processes. Without tangible value metrics, securing continued investment becomes difficult. One participant captured this: "We're running interesting experiments, but we can't connect them to business outcomes that matter to the board."

Fear of job displacement undermining adoption

When leadership messaging focuses solely on cost reduction and automation, employees have little incentive to engage. This creates a paradox: the people who understand processes best are least motivated to help improve them.

Lack of senior leadership support

Without executive buy-in and clear strategic direction, AI efforts remain fragmented. Different teams pursue conflicting approaches, duplicating effort and missing opportunities for coordinated scaling. As one leader noted: "We've got five different teams building chatbots. None of them are talking to each other."

Governance and security concerns

Organisations are grappling with questions around data privacy, ethical AI use and the risks of "shadow AI," where employees use unauthorised tools and inadvertently expose confidential information.

Data quality and infrastructure limitations

Whilst AI doesn't require perfect data to deliver value in a pilot, scaling to an enterprise-wide initiative does require robust data foundations and clear governance. Many organisations discover their infrastructure can't support organisation-wide deployment only after investing significantly in pilots.

Why these blockers exist: four underlying patterns

Beyond these immediate obstacles, the session revealed four deeper patterns that explain why AI struggles to move from pilot to production. These provide a diagnostic framework for organisations to assess their readiness for scaling.

1. Data and infrastructure aren't enterprise-ready

Pilots often live in sandboxes, isolated from enterprise systems and real-world complexity. One participant described rolling out an impressive AI tool within their IT function, only to discover business teams had no idea how to access the underlying data. The technology worked perfectly in the pilot. The integration required for enterprise implementation didn't exist.

Scaling requires enterprise-grade data governance addressing quality, access and privacy. It needs integration with existing processes, not parallel workflows that never connect. The key isn't waiting for perfect data but building foundations whilst pursuing pilots that demonstrate quick wins.

2. Change management and people considerations are overlooked

Pilots are typically run by enthusiasts within IT functions who love experimenting with new technology. Scaling to an enterprise-wide initiative needs company-wide adoption, which requires addressing different motivations, skill levels and concerns across the organisation.

One participant noted their IT function had been rolling out impressive AI tools for months. The technology was solid in pilot form. Business teams had no idea how to use them or why they mattered.

Successful scaling requires clear communication about how AI creates opportunities (not just eliminates costs), training across all levels, incentive structures that reward innovation and change management expertise embedded throughout the transformation.

3. ROI frameworks and strategic alignment are unclear

The "AI or die" mentality of 2023 led many organisations to launch pilots without clear business cases. When asked to justify scaling a pilot, teams struggle to articulate the value being created—not because value doesn't exist, but because nobody defined what success looked like before starting.

Value doesn't always mean immediate revenue or cost savings. It can include productivity gains that enable growth, improved decision-making quality and speed, enhanced customer experience, competitive positioning and risk reduction. But value must be defined, measured and communicated clearly from the start.

One of the session participants made a compelling point: "Stop looking down at your bottom line or you're going to trip over. Start looking up at your top line." The real opportunity is creating new business models, developing competitive advantages and enabling innovation that wasn't previously possible.

4. Governance structures hinder rather than enable

Without proper governance, organisations face growing risks as they attempt to scale pilots: shadow AI exposing confidential data, duplicated effort across teams, regulatory compliance gaps and inconsistent approaches to ethics and bias.

Effective governance connects top-down strategy with bottom-up experimentation, shares learning across teams, manages risk whilst enabling safe experimentation and maintains a "golden thread" linking pilot innovations to business value. One participant described implementing a tiered approach: low-risk pilots get approved in days, medium-risk in weeks, high-risk get full review. This keeps innovation moving whilst managing genuine risks.

How we help organisations scale AI successfully

At Sullivan & Stanley, we've developed frameworks specifically designed to bridge the gap between promising pilots and enterprise-wide AI initiatives. Our approach combines AI-powered insights through MissionHub.ai with Mission-Based Working methodology to close the gap between strategy and implementation.

A recent example: We worked with a UK facilities management organisation whose Finance function was grappling with fragmented data and manual processes, creating inefficiencies across reporting and controls. The challenge wasn't experimenting with AI, it was embedding it responsibly into their Finance operating model whilst demonstrating clear value.

We partnered with their Finance leadership to co-create a practical AI strategy. Through comprehensive discovery, we identified 63 opportunities and prioritised these into quick wins and 22 Copilot initiatives aligned with group governance. We built the business case, launched the first live initiative, and trained 21 Finance "AI Champions" to ensure sustainability.

The impact: £5m savings potential identified (10x initial expectation), three high-feasibility pilots launched, £0.5m savings and ~8k days capacity uplift delivered within the year, and a federated AI operating model established. Finance is now positioned as a blueprint for responsible AI adoption across the Group.

Diagnosing your readiness to scale

These patterns offer a diagnostic framework for your AI journey. Ask yourself:

Are our pilots genuinely connected to business strategy?
Can we articulate the value being created in terms that would justify enterprise-wide initiatives?
Do our people have incentives to help scale pilots, or reasons to resist?
Does our governance enable the journey from pilot to production, or add bureaucracy?

The answers will tell you where to focus your efforts.

The overarching insight is clear: AI works in pilots. Scaling those pilots into enterprise-wide initiatives requires treating it like major transformation, not a technology project. Organisations that take this approach—focusing on process before technology, proving value early, and building foundations whilst pursuing quick wins—will be best positioned to capture AI's full potential.

Ready to scale your AI pilots into enterprise initiatives?

Sullivan & Stanley helps organisations move from promising pilots to production-ready AI at scale. Our approach combines proven transformation methodology with AI-powered delivery tools.

Book a 30-minute briefing to explore how we can help your organisation bridge the gap from pilot to production: get in touch.

Or join our monthly Community Events where senior leaders explore transformation challenges together: https://www.sullivanstanley.com/communities/.