It has long been an organizational axiom that a lack of AI-ready data is the primary obstacle moving projects from pilot to production. Yet, over the last few years, the AI field has hyper-evolved. With the rise of multimodality, RAG, reasoning models, and agentic workflows, commercial acceptance and infrastructure tooling have fundamentally shifted.
Can it really be true that despite massive data remediation efforts, data readiness remains the single greatest roadblock to success?
Not necessarily. Organizations hungry for AI risk-and-success metrics are often forced to rely on backward-looking surveys. Compounding this is lazy, sensationalist reporting that cherry-picks attention-grabbing headlines from highly nuanced reports.
Anatomy of a Flawed Headline: The MIT NANDA Study
Consider the widespread headlines surrounding the 2025 MIT NANDA survey, which shocked the industry by claiming that 95% of enterprise AI pilots fail to achieve measurable ROI and stall before generating revenue.
If you chase down the original methodology, the structural flaw becomes obvious: investigators stopped the clock on pilot projects after just six months. Declaring a complex enterprise AI pilot a failure after 180 days doesn't measure project viability; it measures corporate impatience.
When you dig into the actual details, a very different story emerges:
- The 95% failure rate applied exclusively to embedded, task-specific LLMs. General-purpose LLM initiatives actually enjoyed a 40% success rate.
- Enterprise timelines demand patience. True enterprise companies (>$100M annual revenue) averaged nine months or longer to move from pilot to implementation. Conversely, nimbler mid-market companies averaged 90 days.
By defining any project without positive ROI at the six-month mark as a failure, the study automatically disqualified the most complex, successful long-term enterprise implementations. Yet, this crucial constraint was entirely ignored in the media's scramble for clicks.
The Battle of the Benchmarks: Factoring in Definition
To see how easily these narratives can be manipulated, look no further than how quickly the MIT study was contradicted by Wharton’s Accountable Acceleration study.
Wharton flipped the script, claiming that 74% of enterprises see positive returns from generative AI, with over 35% reporting significantly positive ROI.
How could two prestigious institutions reach such wildly opposing conclusions? It all comes down to their definition of success. While MIT used a rigid, short-term financial clock, Wharton defined "return" much more broadly. They factored in indirect benefits like individual employee productivity, employee retention, and overall operational efficiency rather than just immediate bottom-line revenue.
Finding completely conflicting narratives in enterprise tech studies isn't just common, it's the norm. It all depends on what the researchers choose to measure, when they stop the clock, and what story they want to tell.
A Tale of Two Surveys: Tracking the Incentives
When we look at fresh information from late 2025 and 2026, this bottleneck picture becomes far more nuanced—and heavily dependent on who is asking the questions. Let’s look at two competing viewpoints recently released just weeks apart.
Perspective 1: Gartner (April 2026)
Gartner surveyed 782 infrastructure and operations (I&O) leaders. Released just ahead of their major May IT Infrastructure conference, the findings focused heavily on operational management:
- Only 28% of AI use cases fully met ROI expectations, while 20% failed outright.
- For the 57% who experienced failures, the primary culprit was misaligned expectations, expecting too much, too fast.
- While 38% blamed skill gaps and another 38% blamed poor data quality, the overwhelming drivers of success were organizational: executive buy-in and embedding AI directly into existing employee workflows.
Gartner's takeaway? Data is a factor, but the real enemy is a lack of up-front preparation and unrealistic business cases. Oh, and IT service management is the best place to invest AI dollars if you want a successful project.
Perspective 2: Dun & Bradstreet (May 2026)
Now look at Dun & Bradstreet’s AI Momentum Survey of 10,000 global businesses. While they found that 97% of organizations have active AI initiatives, their data metrics tell a starkly different story from Gartner’s:
- Only 5% of respondents say their data is fully ready for AI.
- 50% cite limited data access as their leading obstacle, alongside data quality (40%) and system integration issues (38%).
Why is D&B so laser-focused on data readiness? Follow the money. D&B is the creator of the D-U-N-S® Number and markets a proprietary D&B.AI offering designed to connect business identity across fragmented systems. They have a direct commercial interest in AI data readiness being a massive problem. They explicitly cap off their survey by warning that "every system an agent touches must agree on who the entity is. Our D-U-N-S® Number makes that possible. The companies that solve for verified business identity first will be the ones whose agents actually work."
Reading Between the Data Lines
What do we make of this corporate tug-of-war? Gartner frames AI deployment as a complex IT management and workflow integration challenge—a narrative that perfectly positions them to sell advisory services and event tickets to I&O leaders. Meanwhile, Dun & Bradstreet frames AI as an identity and data validation crisis, a narrative that conveniently requires a Dun & Bradstreet subscription to solve.
This doesn’t mean the data from these surveys is fabricated, but it does mean the survey questions were heavily steered by organizational self-interest.
As an enterprise leader navigating the AI landscape in 2026, your takeaway shouldn't be to panic over a "95% failure rate" or a "5% readiness rate". Instead, recognize that AI project success obstacles have evolved. Data readiness is no longer an unavoidable project killer; it is simply one factor alongside corporate patience, workflow integration, and realistic expectation setting.
The next time an alarming AI statistic crosses your desk, don't rush to change your business strategy. Instead, ask the definitive analytical question: What are they trying to sell me?