Why companies struggle to get value from AI

2025 was the year companies heavily invested in AI to solve internal business problems and accelerate growth. However the return on investment for most organizations is virtually non-existent. According to a report from MIT:

"Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P\&L impact."

The report dubbed this the "GenAI Divide," where a tiny fraction of companies are succeeding with AI while the rest struggle to move the needle.

A quick note: this challenge is specific to internal AI for business, or the systems companies build for their own operations, marketing, and sales. The failure rate reported by MIT does not include external AI-powered features companies build into their products.

On social media, influencers have pointed to the 95% of failed AI projects as proof that AI is not ready for business. But the MIT researchers spotlight a different culprit:

"This divide does not seem to be driven by model quality or regulation, but seems to be determined by approach."

In other words the AI models are currently smart enough. But the haphazard approach most companies take with AI is killing results.

AI is still failing companies

At Adapt we spoke to dozens of CEOs and founders of startups, and we keep hearing the same recurring challenge. AI is everywhere, and it promises more than ever, but still most people aren't seeing the results they'd expect.

When we dig into the cause of these issues we find that startup leaders are frustrated at a few common issues. Those who tried building internal solutions from scratch discovered the problem of unreliable agent orchestration. Others have been burned by ChatGPT and Claude's "enterprise" level packages, only to find that their data connectors are limited, hard to set up, and don't include their preferred tools.

We spoke to one customer who uses HubSpot purely for marketing and Salesforce for sales. Extracting customer insights from these systems is full of friction, with each product offering its own siloed AI solution. When it comes time to make decisions about what programs and campaigns to pursue the next quarter, it takes days - or even weeks - of manual work to get answers.

The data bottleneck to AI success

Deloitte's Q4 State of Generative AI report corroborates what we've seen. The firm quotes a software engineering manager at a leading technology company:

"Data emerged as the central factor for \[our GenAI] success. While the models and computing power existed, accessing the right data proved to be the biggest bottleneck."

And why is the data barrier so persistent? The root cause often lies in who is driving the adoption.

Another Deloitte interview quotes a director of product management at a semiconductor company:

"Within our organization, the demand for GenAI use cases and innovation primarily comes from middle management and employees, rather than being driven by the C-suite."

This illustrates the structural flaw in many current AI projects. Middle managers can purchase a tool, but they rarely have the authority to unify data silos across business functions.

To truly cross the divide, companies must move beyond isolated tools that only boost individual productivity. Real business growth requires systems that are capable of connecting the dots between sales, marketing, and operations.

MIT clarifies that the 5% of successful projects they surveyed all create "systems that integrate with existing processes." Successful AI deployments within a business help to accelerate how employees currently work by connecting to the tools they use and the data that's most important to them, even if it exists outside of their team.

Work AI: where to go from here

At Adapt, we are building the company agent platform: a system that deeply understands how you work and enables your entire company to become AI-native.

By providing a unified platform that everyone in the company can use, Adapt turns your fragmented data into a central source of truth that can reason and take action.

You can get a demo, or check out our guide, AI for Startup Leaders, for tips on evaluating AI solutions for work.