An AI readiness assessment scores an organization across pillars like strategy, data, governance, operating model, and culture, then hands back a maturity band and a prioritized roadmap for closing the gaps it found.¹ A gap analysis, run against one committed roadmap item, names the specific blockers and the people accountable for closing them. Both produce a list of things to fix.
The difference lies in scope: your organization, versus the program or model you are actually trying to ship. Author Charlene Li argues the first will reliably tell you what you already suspect (you are not ready for AI!), and that only the second produces real progress.
Li, co-author of Winning with AI with co-author Dr. Katia Walsh, made the case to host Satyen Sangani on a recent episode of AI Radicals, saying that organization-wide AI-readiness tests don’t give you the guidance you need:²
"Spending any time on feasibility studies or readiness strategies and evaluation assessments, we think, is just going to tell you it's not feasible and you're not ready. So how does that help you?"
Leaders would do better, Li argues, to narrow the scope of their program:
"My point here is that if it's strategic, then it becomes strategic and a must-do. You wouldn't look at this and say, 'Well, can we do it?' And that's how we decide how to do things. No, because you don't know how to do any of it, frankly."
Readiness is not a standing property of your company writ large. It is a property of the workload you are trying to ship. In this blog, we'll cover why enterprise-wide AI readiness assessments reliably confirm what you already know, what the 2026 failure data actually points to, and how to replace the assessment with two things you can act on: a strategic outcome committed to a dated roadmap, and a gap analysis run against the workload itself.
Why AI projects fail: Skills gaps, data quality, and poor scoping
Gartner surveyed 782 infrastructure and operations leaders in November and December 2025 and found that only 28% of AI use cases in infrastructure and operations fully succeed and meet ROI expectations, while 20% fail outright.³ Among the 57% of leaders who reported at least one failure, the failures were largely driven by initiatives that were overly ambitious or poorly scoped⁴ — people expecting too much, too fast.
Gartner names two causes that a maturity model would catch. Among leaders who faced setbacks, 38% cited persistent skills gaps and 38% cited poor data quality or limited data availability⁵ as direct causes of failure. Those are standing organizational attributes, and every readiness framework scores them.
Today, organizations are already scoring themselves on data readiness…and failing anyway, because a pillar score of "developing" on data quality tells you nothing you can act on. It does not tell you which field is wrong, which of three revenue tables finance actually uses, or whose definition of "active customer" the model inherited.
Gartner's own read points the same way: returns depend less on model sophistication than on how well the technology is integrated, governed, and aligned with real operational needs⁶, three things that only exist once something real is being built.
The assessment quarter is not free, either. Forrester's 2026 predictions report that only 15% of AI decision-makers saw an EBITDA lift from AI in the previous 12 months⁷ and that fewer than a third can tie AI value to their organization's financial growth. Forrester expects finance leaders to apply real rigor to the business case and enterprises to defer a quarter of planned AI spend into 2027⁸ as a result. Budget is getting harder to hold onto for programs that cannot show value, fast.
Nobody can answer the feasibility question at the start
No assessment can score the feasibility of truly novel work. Rather, a rubric measures familiarity, and familiarity with a capability you have never deployed will always score low. Thus, a skills-gap finding is not a training problem to solve before starting, but the expected state of a team that has not built the thing yet! The assessment reports it back as a finding, which becomes a reason to wait.
To combat such inertia, Li is pragmatic, recommending leaders make a schedule:
"You may not be feasible right now. You may not have the capabilities, the technology may be there, but it should be on your roadmap. Maybe three quarters from now is when you anticipate you'll be ready to do that, but have it on the roadmap and do the steps you need today to get to the point where you can actually implement it."
The commitment goes on the roadmap with a date, and the distance between today and that date becomes named work with owners. This is the difference between planning to be ready and scoring how ready you already are.
You don't know what you need until you build something
Requirements are discovered by building, not by surveying. Anyone who has run a governance program has watched this happen firsthand.
Sangani described what came back when Alation drafted its own AI policy with AI assistance:
"One of the things that came into the meeting when it was first presented was this idea that no data about our customers should ever interact with AI. And we were like, how is this actually going to work?"
On paper, the rule looked responsible. But in practice, it would have blocked most of the work the policy existed to enable. The problem was not careless drafting but sheer novelty, since the policy was written before anyone had built the thing it governed.
Sangani made a similar point about something far more ordinary than an AI system in describing a dashboard:
"Does everybody in the company need to see this information? Is it limited to a few people? But you really don't even know this until you've actually developed the information and the data."
This is the structural flaw in assessing readiness ahead of delivery. The questions a readiness framework asks (Who owns this data? Which fields are sensitive? Which definitions need reconciling?) become answerable only once a real workload pulls on real data.
The organizations that will look competent in 18 months
Those who win with AI will not be those with the highest readiness scores. They will be those that picked strategic outcomes, wrote down what stood in the way of each, and closed those gaps on a published schedule. Gartner found the same pattern on the success side: among leaders whose AI use cases worked, 33% attributed it to embedding AI into the systems and processes people already used⁹ rather than running it alongside them.
Li has a name for what stops such work from ever starting: FOGI, or the fear of getting in,² the counterweight to the fear of missing out. Perfect is the enemy of the good. There is never enough information, so the safer move is to wait for the water to clear, and an assessment is the most respectable available expression of that instinct.
Genuine compliance obligations are a separate matter. Model risk requirements and the EU AI Act answer whether you are permitted to proceed, which is not a question about capability, and the Act's high-risk obligations have since been deferred to December 2027 and August 2028, which buys time to document rather than a reason to stop.
Readiness tends to follow delivery rather than precede it, which is why a durable enterprise AI operating model is built around committed outcomes and governed data products rather than an annual maturity review.
Narrow the scope and the blockers start to rhyme across programs: Nobody can say where a field came from, which of three revenue tables finance actually uses, who owns the definition the model inherited, or whether this dataset is fit for this use case.
Those are metadata questions, not maturity questions. Gartner defines AI-ready data by the ability to prove the fitness of data for a given AI use case, which makes readiness a case-by-case determination that depends on metadata (lineage, ownership, definitions, quality history, usage) to find, govern, evaluate, and justify the data behind each workload.¹⁰ It also explains why readiness follows delivery: the metadata that proves fitness accumulates only as real workloads pull on real data.
That accumulation needs somewhere to live. Alation captures lineage, ownership, definitions, and quality signals as the catalog for the metadata AI builders need, so the gap analysis on each roadmap item has something to resolve against instead of a survey response.
The gaps on your roadmap keep turning out to be metadata gaps, that is a solvable problem. Book a demo with us today.
Sources & notes
Every external claim on this page is independently verifiable. The public sources are listed here.
Gartner AI Maturity Assessment scores maturity across strategy, data, governance, engineering, operating model, culture and AI product/value, and returns a prioritized roadmap, a gap visualization and analyst recommendations per gap. Accessed 17 September 2026. — Gartner ↗ https://www.gartner.com/en/chief-information-officer/research/ai-maturity-model-toolkit
Charlene Li on the feasibility trap, the roadmap alternative, and FOGI; Satyen Sangani on Alation's first AI policy and on dashboard access. — AI Radicals, Season 4 Episode 7, "The 90-Day AI Roadmap" ↗ https://www.alation.com/podcast/episodes/ai-roadmap-charlene-li/
28% of AI use cases in infrastructure and operations fully succeed and meet ROI expectations; 20% fail outright. Survey of 782 I&O leaders, November–December 2025. — Gartner ↗ https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-says-artificial-intelligence-projects-in-infrastructure-and-operations-stall-ahead-of-meaningful-roi-returns
57% of I&O leaders reported at least one AI failure; the 20% failure rate is largely driven by initiatives that were overly ambitious or poorly scoped, with many citing expectations set too high, too fast. — Gartner ↗ [same URL as note 3]
Among I&O leaders who faced setbacks, 38% cited persistent skills gaps and 38% cited poor data quality or limited data availability as direct causes of AI project failure. — Gartner ↗ [same URL as note 3]
ROI from AI is driven not by the sophistication of the model but by how well the technology is integrated, governed, and aligned with real operational needs. — Gartner ↗ [same URL as note 3]
15% of AI decision-makers reported an EBITDA lift for their organization in the past 12 months; fewer than one-third can tie the value of AI to P&L changes. — Forrester, Predictions 2026: AI Moves From Hype To Hard Hat Work, 28 October 2025 ↗ https://www.forrester.com/blogs/predictions-2026-ai-moves-from-hype-to-hard-hat-work/
Forrester predicts enterprises will delay 25% of planned AI spend into 2027 as CEOs pull CFOs further into AI decisions. — Forrester, Predictions 2026, 28 October 2025 ↗ [same URL as note 7]
Among I&O leaders reporting AI success, 33% attributed it to embedding AI into the systems and processes people already use. — Gartner ↗ [same URL as note 3]
Gartner defines AI-ready data as determined by the ability to prove the fitness of data for AI use cases; readiness is therefore assessed case by case. — Gartner, Quick Answer: What Makes Data AI-Ready? (paywalled) ↗ https://www.gartner.com/en/documents/5432763
Analyst attributions
Gartner, "Gartner Says AI Projects in I&O Stall Ahead of Meaningful ROI Returns," Q&A with Melanie Freeze, Director Research, 7 April 2026.
Forrester, Predictions 2026: Artificial Intelligence, Forrester Research, Inc., 28 October 2025.
- AI
- Data Quality
Keep reading
More from the data desk
