MIT’s NANDA report ("State of AI in Business Report 2025") finds that 95% of enterprise AI initiatives have delivered zero ROI, and that “buy” strategies (specialized AI tools and vendor partnerships) outperform in-house builds. For those who have read the report:  Do the findings match your experience, and have you seen more success with buy vs. build?

Interesting read, but I’d challenge the 95% failure stat—it feels like a harsh verdict based on a narrow slice of survey respondents. In my experience leading enterprise-scale Data & AI transformations, I’ve seen closer to a 1-in-3 success rate. Still not perfect, but far from catastrophic.

As for buy vs. build: buying specialized GenAI tools can accelerate time-to-value, especially when governance and integration are baked in. But building in-house isn’t dead—it just needs a clear business case, strong engineering, and executive patience (a rare trifecta!).

I shared more thoughts on this in a recent post, where I quoted the same report, but focus more on the pitfalls and the best practices to avoid becoming another number in that stat: https://www.linkedin.com/posts/ali-asgar-juzer_genai-enterprise-ai-roi-activity-1234567890123456789

I suspect "buy" solutions are finding more success because companies have more controls around financial investments and they're neglecting the rollout aspect of "build" solutions. There's the temptation to believe that folks will "figure it out" rather than putting time into upskilling business teams, helping SMEs identify  high value opportunities, and getting folks the data and support they need to develop agents achieving true ROI.

Both have their merits. A properly built internal AI tool (be it LLM/GenAI or other - for example a sales omnichannel engine) has the advantage that you can enter company confidential information for continuous learning and improvement; it will speak 'your language'. Bought tools also have their merit, but you need to be cautious with your company data.