McKinsey Claims 80% Companies Fail to Generate AI ROI. They’re Wrong.
Sometimes, you see a headline and just have to shake your head. Sometimes, you see a bunch of headlines and need to scream into a pillow. This week’s headlines on AI ROI were the latter:
- Companies are Pouring Billions Into A.I. It Has Yet to Pay Off – NYT
- MIT report: 95% of generative AI pilots at companies are failing – Forbes
- Nearly 8 in 10 companies report using gen AI – yet just as many report no significant bottom-line impact – McKinsey
AI has slipped into what Gartner calls the Trough of Disillusionment. But, for people working on pilots, it might as well be the Pit of Despair because executives are beginning to declare AI a fad and deny ever having fallen victim to its siren song.
Because they’re listening to the NYT, Forbes, and McKinsey.
And they’re wrong.
ROI Reality Check
In 20205, private investment in generative AI is expected to increase 94% to an estimated $62 billion. When you’re throwing that kind of money around, it’s natural to expect ROI ASAP.
But is it realistic?
Let’s assume Gen AI “started” (became sufficiently available to set buyer expectations and warrant allocating resources to) in late 2022/early 2023. That means that we’re expecting ROI within 2 years.
That’s not realistic. It’s delusional.
ERP systems “started” in the early 1990s, yet providers like SAP still recommend five-year ROI timeframes. Cloud Computing“started” in the early 2000s, and yet, in 2025, “48% of CEOs lack confidence in their ability to measure cloud ROI.” CRM systems’ claims of 1-3 years to ROI must be considered in the context of their 50-70% implementation failure rate.
That’s not to say we shouldn’t expect rapid results. We just need to set realistic expectations around results and timing.
Measure ROI by Speed and Magnitude of Learning
In the early days of any new technology or initiative, we don’t know what we don’t know. It takes time to experiment and learn our way to meaningful and sustainable financial ROI. And the learnings are coming fast and furious:
Trust, not tech, is your biggest challenge: MIT research across 9,000+ workers shows automation success depends more on whether your team feels valued and believes you’re invested in their growth than which AI platform you choose.
Workers who experience AI’s benefits first-hand are more likely to champion automation than those told, “trust us, you’ll love it.” Job satisfaction emerged as the second strongest indicator of technology acceptance, followed by feeling valued. If you don’t invest in earning your people’s trust, don’t invest in shiny new tech.
More users don’t lead to more impact: Companies assume that making AI available to everyone guarantees ROI. Yet of the 70% of Fortune 500 companies deploying Microsoft 365 Copilot and similar “horizontal” tools (enterprise-wide copilots and chatbots), none have seen any financial impact.
The opposite approach of deploying “vertical” function-specific tools doesn’t fare much better. In fact, less than 10% make it past the pilot stage, despite having higher potential for economic impact.
Better results require reinvention, not optimization: McKinsey found that call centers that gave agents access to passive AI tools for finding articles, summarizing tickets, and drafting emails resulted in only a 5-10% call time reduction. Centers using AI tools to automate tasks without agent initiation reduced call time by 20-40%.
Centers reinventing processes around AI agents? 60-90% reduction in call time, with 80% automatically resolved.
How to Climb Out of the Pit
Make no mistake, despite these learnings, we are in the pit of AI despair. 42% of companies are abandoning their AI initiatives. That’s up from 17% just a year ago.
But we can escape if we set the right expectations and measure ROI on learning speed and quality.
Because the real concern isn’t AI’s lack of ROI today. It’s whether you’re willing to invest in the learning process long enough to be successful tomorrow.
