The State of AI After the MIT Report on 95% Project Failure
There was a viral “95% failure” headline from the NIT NANDA State of AI in Business 2025 report, and it is being completely misread.
The study’s own data and limitations plus Smarteeva’s field experience across MedTech manufacturers point to a different conclusion altogether:
- AI projects fail when they are generic, consultant-specified, or internally custom-built without deep workflow embeddings and learning loops.
- Enterprises that partner with vertical product companies in this case, vendors that operate within regulated processes are scaling value twice as often as internal builds.
The MIT Report says: The “95% failure” refers to the pilot-to-production chasm for custom enterprise AI, only 5% reach durable, P&L-relevant production; generic chat tools see adoption but not workflow impact. The core barriers are:
- Learning Gap: Tools don’t retain feedback or adapt to context.
- Lack of Integration in Day-to-Day Workflow: Tools do not integrate with daily operations.
- External Partnerships Outperform Internal Builds: Strategic buy/partner approaches were twice as likely to deploy successfully as in-house development.
- Shadow AI Is Real: Employees get real value from personal LLMs while official initiatives stall. This shows that flexible, high-quality tools work when they fit the job.
- The authors recognize their methodological limits, making the “95% failure” claim directional rather than universal.
What Does That Mean in practice?
- Consultant-led projects often result in committeeware. Internal IT and large SIs can align stakeholders, but they often cannot ship performant, AI-native software. By contrast, Smarteeva’s deployments for top 10 MedTech companies deliver live, validated systems in as little as 12 weeks.
- “We will build it ourselves” is usually wishful thinking. Enterprises overestimate their product and AI engineering capacity. Several Smarteeva clients previously spent large amounts on internal NLP/AI-based classification tools that failed and could not pass validation. Smarteeva’s vertical knowledge and tools achieved 98% accuracy within the first production cycle.
- AI-washed add-ons or systems lose to native vertical products. Smarteeva’s success in replacing generative automation bots reflects a clear pattern: deep integration with quality systems and regulatory workflows drives adoption.
- Domain and vertical knowledge make a huge difference. Smarteeva’s feedback loops and domain-specific agents trained on MedTech compliance increase the chances of success.
- ROI hides in back-office/regulated processes. Smarteeva’s deployments show that enterprises can accomplish vastly more with vertical AI systems. AI is being pushed by management on the Sales and Marketing side. That is fine, but from our experience, most of the ROI comes from allowing the back office to operate much more smoothly. It becomes a major help for front-office operations.
Bottom line: Don’t generalize the “95% failure” into AI pessimism. Read it as a procurement warning. Generic tools, big-bang consultant builds, and unlearning/non-vertical systems fail. To get on the right side of the AI divide, buy from vertical product companies that embed into your regulated workflows, learn from your feedback, and deliver clear and improved operational outcomes.
About Smarteeva: Headquartered in Dallas, Texas, Smarteeva Software pioneers Post Market Surveillance for the medical device sector. We offer Complaint Handling, Adverse Event Reporting, Risk Management, and Recall Management solutions, designed for compliance and streamlined operations. Built on Salesforce.com with cutting-edge AI and Machine Learning, we are true partners to our users.
Reach us at: info@smarteeva.com | www.smarteeva.com
