Record investment, high expectations, invisible ROI. If AI is so transformative, why are so many companies coming home empty-handed?
The number keeping executives up at night
Recent research on enterprise adoption of generative AI keeps converging on an uncomfortable conclusion: the overwhelming majority of projects produce no measurable financial return. The study The GenAI Divide: State of AI in Business 2025, run by MIT's NANDA initiative, found that roughly 95% of enterprise GenAI implementations had not generated meaningful P&L impact at the time of the analysis [¹].
Gartner, for its part, projects that at least 30% of generative AI projects will be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, rising costs and unclear business value [²].
It is not a lack of technology. It is not a lack of talent. It is something more structural, and far more solvable than it looks.
The good news: looking at the initiatives that actually worked, you can identify five dimensions of maturity that separate the winning 5% from the 95% that fall by the wayside. If your company is mature on all of them, the conversation changes completely.
The 5 dimensions of AI maturity
Instead of seeing AI as a technology project, winning companies treat every initiative as a move on five simultaneous fronts. They work as a system: weakness in one brings down the whole.
1. Strategy & Leadership: where you decide to start
The critical question: has your company defined where AI should be applied first?
Most organizations that fail at AI start with the tool, not the problem. "We bought Copilot licenses. Now where do we use them?" is the inverted order that keeps filling the PoC graveyard.
The four stages of maturity:
- Level 1, Improvised: Loose ideas. No criteria for deciding where to start.
- Level 2, Reactive: We react to requests. Choices are one-off and have no model.
- Level 3, Intuitive: We prioritize by perceived impact, but without formal scoring.
- Level 4, Structured: A structured portfolio of priorities, aligned with the company plan.
What changes between levels: at level 4, every AI initiative has a named business owner, a cost and time baseline before the project starts, a stated return target within 90 to 180 days, and a decommissioning plan if it misses the target. At levels 1 and 2, "let's see what happens" is the most dangerous phrase in the committee.
2. Data: the fuel that is (almost always) not ready
The critical question: is the data your AI will need accessible and reliable?
Generative AI is brilliant with language, but its usefulness in the enterprise depends on structured, accessible, governed data. The same MIT study highlights that organizations with mature data architecture capture value much faster, while the rest invest in AI only to discover the real problem was two floors down, in the data lake [¹].
The four stages of maturity:
- Level 1, Fragmented: Scattered across spreadsheets and systems. Every area does its own thing.
- Level 2, Siloed: It exists, but with silos, rework and inconsistent quality.
- Level 3, Partly organized: The main data is organized, but incomplete for AI at scale.
- Level 4, AI-ready: Critical data is integrated and of high enough quality to feed models.
Typical warning signs at levels 1 and 2:
- The same customer concept defined 4 different ways across systems
- Critical documentation in scanned PDFs without OCR
- Access permissions nobody can explain
- "Truth" scattered across private spreadsheets
Until that is solved, any LLM becomes an eloquent generator of wrong answers.
3. People & Leadership: who owns the agenda
The critical question: who owns the AI agenda at your company?
In practice, this is the axis that most separates real maturity from corporate theater. When AI has no executive sponsor with authority, budget and targets, it becomes hallway talk, competing with 20 other IT priorities.
The four stages of maturity:
- Level 1, Orphaned: Nobody. The topic was handed to IT without C-level sponsorship.
- Level 2, Nominal: There is a name on paper, but no real authority or budget.
- Level 3, Partial: Someone is responsible, but there is no formal follow-up cadence.
- Level 4, Structured: An executive sponsor with clear ownership, targets and a defined budget.
What changes between levels: at level 4, there is an executive who answers for the AI agenda in results meetings, with their own targets and a ring-fenced budget. At levels 1 and 2, when a problem appears, everyone points in different directions, and the project dies halfway through.
4. Use Cases & Return: does AI make money here or not?
The critical question: can you show the return on your AI initiatives today?
If MIT finds that 95% of initiatives deliver no measurable P&L impact [¹], this is the axis where that statistic materializes. It is not a lack of value, it is a lack of instrumentation to capture and show value.
The four stages of maturity:
- Level 1, Opaque: No. The impact is felt, but it does not show up in the P&L.
- Level 2, Anecdotal: We have one-off examples, but no formal measurement model.
- Level 3, Partial indicators: The main cases have indicators, but no baseline or target.
- Level 4, Continuous ROI: ROI documented per initiative, with KPIs and continuous tracking.
What changes between levels: among the winning 5%, it is common to see 60 to 70% of project effort going into the data pipeline, versioned prompts and the evaluation system, and only 30 to 40% into "the model itself". Among those that fail, the ratio is usually the exact opposite. Without a baseline and a target, there is no way to defend the next investment in front of the finance committee.
5. Governance & Risk: the axis that stalls mature projects
The critical question: do privacy, LGPD and security factor into your AI decisions?
The same Gartner report identifies inadequate risk control as one of the main factors that kill early-stage AI projects [²]. In Brazil, the LGPD regulatory framework [³] and the international reference of the European AI Act [⁴] make this axis non-negotiable for anyone who wants to scale.
The four stages of maturity:
- Level 1, Neglected: No. The topic only comes up when there is a problem.
- Level 2, Informal concern: There is concern, but no process or formal assessment.
- Level 3, Partial: They factor into use case validation, but in an unstructured way.
- Level 4, Integrated: Risk governance is part of the strategy from conception.
What changes between levels: at level 4, governance is an enabler, not a brake. There are clear criteria (sensitive data? autonomous decisions? customer impact?), a lightweight approval committee, and the team knows before starting which classes of use case are allowed. At levels 1 and 2, legal and compliance discover the project at go-live, and everything stalls.
What the 5% have in common
The companies that escape the MIT statistic are not necessarily bigger, richer or more technical. They are companies with discipline across all five axes at once.
Two behavior patterns are practically universal among the winners:
Focus on "few and deep", not "many and shallow"
Instead of 40 pilots scattered across the organization, 3 to 5 rigorously chosen use cases per cycle, all covered on the five maturity axes before they start.
"Human-in-the-loop" architecture by default
The winners do not try to automate 100% from day one. They start with an "AI suggests, human approves" model, measure real productivity gains, and only then discuss full automation where the risk allows it.
A quick map of where you stand
Before approving your next AI project, it is worth an honest check-up on the 5 dimensions. If any of them sounds too familiar in the right-hand column, pay attention:
| Axis | Critical question | Warning sign |
|---|---|---|
| Strategy & Leadership | Has your company defined where AI should be applied first? | Loose ideas. No criteria for deciding where to start |
| Data | Is the data your AI will need accessible and reliable? | Scattered across spreadsheets and systems. Every area does its own thing |
| People & Leadership | Who owns the AI agenda at your company? | Nobody. The topic was handed to IT without C-level sponsorship |
| Use Cases & Return | Can you show the return on your AI initiatives today? | The impact is felt, but it does not show up in the P&L |
| Governance & Risk | Do privacy, LGPD and security factor into your AI decisions? | The topic only comes up when there is a problem |
A simple rule observed in practice: if three or more axes are on the warning sign, the project is very likely to join the 95%, regardless of the quality of the AI model chosen.
The good news
The same studies that show the 95% failure rate also show the size of the prize for those who get it right: companies at the frontier of adoption are reporting material gains in productivity, customer experience and new revenue models [¹]. The gap between winners and losers is not money, it is method.
The question is no longer "should we invest in AI?". It is "on which of the 5 axes are we behind, and how do we close the gap before the next investment cycle?".
How we can help
We built a free online AI maturity Diagnostic based on exactly these five axes. 5 questions, under 5 minutes, and you get:
- Your maturity level on each of the 5 dimensions: Strategy & Leadership, Data, People & Leadership, Use Cases & Return, Governance & Risk
- Where your initiative stands against market benchmarks
- The 3 priority bottlenecks to solve before scaling
- Actionable recommendations per axis
No commitment, no sales pitch. Just clarity on where you are and what to do next.