Especially if you saw the headline in the MIT study that caught the headlines with the claim that 95% of AI initiatives were not showing impact.
Google’s new research, released in September '25, is vastly different.
What Was The Study?
Google's was a qualitative study of 3,466 senior leaders, focused on enterprises with $10m+ in revenue with a balanced industry spread (retail, finance, healthcare, manufacturing, etc.). Respondents were executives, mostly C-level or senior leaders.
What Did They FInd ?
74% of companies already see ROI from gen AI.
88% of early adopters (agentic AI) see ROI.
ROI is mainly in productivity, customer experience, sales/marketing
Why The Discrepancy?
-
Firstly, there were very different sample sizes: the MIT study was 52 company interviews of 153 people and 300 case reviews.
-
Google’s focus was on 3,400+ executives, mostly C level or senior leaders. Google’s study focused on enterprises with $10m+ in revenue, with a spread across industries.
-
MIT’s study was focused more on mid and small sized companies that are more resource constrained.
-
Lastly, the measures of productivity differ - Google counts any measurable productivity or CX gain as ROI where as the MIT study demanded P&L-level impact (revenue, profit, headcount savings)
What Should Leaders Take Away ?
Alarmist headlines on AI failing or sales-spun headlines on AI solving all company issues will continue. MIT’s study outlined a divide with AI - where only a few companies succeed. The Google study painted a far more optimistic picture showing a far greater wider impact of the ROI of AI.
The MIT study findings - that there are implementation and learning gaps - are very valid points for many organisations, large and small. But for leaders who have resources that can be applied, and can implement the change management required to take advantage of AI, there is good evidence in the Google study to suggest great upside from implementing AI. The framing that Google use of ‘agentic shift’ as the path to AI ROI is a clear sign that investing in understanding the world of agents and agentic AI is becoming essential. Proceeding with caution is advised however. As the MIT report suggested, deployments that deliver ROI need capabilities that may be lacking currently in many organisations- memory, learning, contextualisation etc.
Building out an agentic framework is made to sound easy by those companies selling solutions in this space. The truth is that the things below the surface (infrastructure, governance etc.) as well as leadership of the change is where the ROI of AI will be made and lost.
