Beyond the Buzz: Why Most AI Projects Still Struggle to Deliver

Yet, for all the investment pouring in, companies seeing successful returns have been hit and miss. According to a report by the MIT Media Lab and Project NANDA, a staggering 95 percent of AI projects are getting zero in return. But beyond the splashy headline number was a familiar problem for businesses: large-scale change is hard. The companies succeeding were those focusing on projects with clear objectives, often deployed toward improving operational efficiency rather than jumpstarting sales and marketing.

The MIT report focused on interviews with customers, but there is another place where we can listen to how businesses are faring in new investment: public companies’ Quarterly Earnings transcripts. In my recent report, I used every earnings transcript from Q1 2024 through Q2 2025 for the S&P 500 to understand how the most successful public companies in the world were deploying AI.

What Did the Data Actually Show?

• AI Initiatives Are Everywhere — But Most Are Still Failing to Deliver
  By Q2 2025, more than 60 percent of S&P 500 firms had announced an AI initiative. But companies only reported tangible returns 21 percent of the time.
• Operations Succeed Most Often
  While firms are deploying more projects in every area, returns are most likely to occur in projects aimed at improving procedural back-end processes using more mature AI techniques, such as Voice AI, document processing, and code generation.
• Sector Surprises
  Financial firms announced the most AI projects but had the lowest percentage of initiatives described with any impact (just 19 percent). By contrast, utilities — hardly the sector most people associate with cutting-edge tech — mentioned returns on 30 percent of their earnings.
• Experience Isn’t Making Companies Better
  Despite a 36 percent increase in launched AI projects from Q1 2024 to Q1 2025, the percentage of projects reporting returns hasn’t improved.

Success Is Found Along the Road Most Traveled

Overall Rates of Return Mentions changed little over the six quarters monitored. But within categories of AI deployment, there were three types of AI that showed higher rates of return in both the MIT study and the earnings reports:

• Voice AI (38.5% Rate of Return Mentions)
• Code Generation (35% Rate of Return Mentions)
• Document Processing (30% Rate of Return Mentions)

We can see the contrast in struggles between AI deployment targeting Customer Experience and Sales vs Backend Operations most clearly in the Finance Sector.

Despite having the most deployments of AI mentioned, it had the lowest Rate of Return Mentions, at 19 percent. However, when it deployed the three categories above, it saw markedly higher rates of return. For Code Generation projects, companies had a 40 percent Rate of Return Mentions, they made up just 5 percent of the projects. On AI initiatives outside the core three categories, it had a Rate of Return Mentions of under 10 percent.

Limitations & What’s Left Unsaid

This research only counts projects mentioned in earnings calls, so it’s possible that some successes (or failures) are left out. And while the MIT report highlights investment in marketing and sales, those projects may simply be more expensive and slower to roll out, while operational projects deliver ROI faster. But here’s the provocation: If only 21 percent of publicly discussed AI projects are described as having an impact, what does that say about the other 79 percent? Are we measuring the wrong things, or is the AI revolution still more talk than transformation?

Want to dig deeper? If you’d like to dive into all the details, charts, and sector breakdowns, check out the full report.

Photo Credit: Dim Gunger