Most AI investments fail—here’s what the winners get right 

Yet despite billions in investment, most organizations still struggle to move from pilot to production to adoption. In fact, according to Gartner® research, “in 2024, 60% of GenAI POCs were abandoned upon completion¹.”  

The difference between AI experimentation and success isn’t about choosing the right large language model; it’s about much more.  

Through our work with partners and customers at various stages of their AI journey, we’ve observed consistent patterns that separate successful implementations from those that stall. Organizations that successfully move from pilot to production focus on four interconnected pillars—and critically, they recognize that technology is only one of them. 

Here’s what we at AWS see winners doing right.

1. Build Your Data Foundation Strategically 

Simply having data isn’t enough—how you organize, govern, and activate it makes all the difference. Leading organizations implement three specific practices: connect all your data together, label and organize it so it’s easy to find, and set controls to ensure only the right people (or agents) have access to sensitive data sets.  

Heavily regulated industries like financial services and healthcare often have an advantage here—their existing governance frameworks can accelerate AI initiatives. However, for organizations starting from scratch, rather than attempting to unify your entire data warehouse, start by working backwards from a specific use case. For instance, a telco operator might begin by connecting network performance data with customer service tickets and billing records for a single purpose: predicting service degradation before customers experience issues. Once that use case delivers value, you can determine which additional data connections matter most and scale from there. 

2. Build Trust Through Security and Verification 

In enterprise AI, trust isn’t just a nice-to-have—it’s the foundation that determines whether your investment moves from pilot to production. Organizations face a dual challenge: they need AI systems secure enough to protect sensitive data, yet accurate enough to make consequential decisions. 

Consider one healthcare provider with 700,000 members. Their customers call at their most vulnerable moments, needing either medical advice or information about their coverage. The opportunity AI could provide was enormous—supporting customers faster, 24/7, in any language. But a single hallucination in this context could cause real harm, eroding trust that takes years to build. 

Leading organizations are moving beyond “trust but verify” to “verify, then trust.” They’re implementing multiple layers of validation: checking inputs for malicious content, verifying outputs against known facts and policies, and continuously monitoring for drift or unexpected behavior. Emerging techniques like automated reasoning—a mathematical approach used for decades in chip design and security verification—can now check AI outputs against defined rules, in some cases reducing hallucinations by 99%. This verification-first approach accelerates innovation rather than slowing it down, empowering teams to experiment more boldly when they know guardrails will catch errors before they reach customers. 

3. Transform Culture, Not Just Technology 

The biggest inhibitor to AI adoption isn’t the technology—it’s change management. Organizations are structured around complex processes, with employees who manage those processes. Getting individuals to step back and reimagine those processes to be end-to-end automated or handled by agents requires intentional cultural transformation. 

Success requires both top-down commitment and bottom-up enablement. Leaders must demonstrate visible commitment beyond words, while employees need the space and support to reimagine their own workflows. BT Group exemplifies this approach: when they embarked on their AI journey in 2024 to accelerate productivity and elevate customer experiences, they didn’t just deploy technology. They built an enablement strategy that matched the technology’s capabilities. Today, nearly 4,000 employees use an AI coding assistant to write and maintain 4 million lines of code per year—but that achievement required investing in training, creating champions within teams, and giving people permission to experiment. 

The reality is nuanced: AI will automate many tasks while simultaneously creating new opportunities and elevating human potential in others. The most successful organizations are transparent about this transformation and invest in reskilling their workforce to thrive in an AI-augmented environment. 

4. Work with the Right Experts 

While some organizations have the resources and expertise to build generative AI capabilities entirely in-house, most find that strategic partnerships accelerate their journey from pilot to production. The question isn’t whether you can go it alone—it’s whether that’s the fastest path to realizing value. 

The right partners bring three critical advantages: technical expertise to navigate the rapidly evolving AI landscape, domain knowledge to apply AI to specific industry and regulatory environments, and change management experience to drive adoption at scale. 

The data bears this out: organizations working with partners possessing deep AI expertise and proven customer success moved their AI projects into production on average 25% faster than those working without specialized partners. In a landscape where speed to value often determines competitive advantage, that acceleration can be decisive. 

The Path Forward 

Successful organizations approach generative AI as a business transformation, not just a technology deployment. The organizations that will thrive aren’t those with the most advanced models, but those that recognize AI success requires equal investment in technology, people, and processes. 

¹ Gartner Report, Forecast Analysis: Artificial Intelligence Services, Worldwide, By Colleen Graham, Ben Fieselmann, etc., September 2025. GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.