Proprietary asset of the Mendoza FuturIA observatory. What is published here is the model, the scoring and the methodology —design artifacts, 100% mine and defensible. No field organization data appears: the example profile is illustrative and synthetic, and is labeled as such.
Approach
A case of the first and fourth links of the method —the sociology of organizations and AI architecture and governance— but before writing a single line of AI: the instrument that decides whether an organization is ready for AI to add to it rather than subtract from it. It is born from a question that recurs across the Mendoza and Latin American productive fabric: “I don’t know where to start with AI”. The honest answer is almost never “start with the model”; it is “first measure where you stand”.
IMIA is that measurement instrument. And its central design decision is a single one: governance is not just another grade in the average; it is a gate. Without it, an organization does not advance, however good the other dimensions may be.
The problem
Most “AI maturity” diagnostics are marketing checklists: they return a single, optimistic, non-actionable number. Worse: they treat governance as an optional appendix, when in practice it is the variable that decides whether an organization moves from isolated pilots to value in production. An instrument was needed that (a) measured the real organizational configuration, not the declared intention, and (b) made governance a structural requirement, not an ornament.
The context makes it urgent: in Latin America, AI penetration is below 4% (against more than 20% in Europe), and the gap within the region is enormous —in Brazil, 41% of large firms use AI against 11% of SMEs (Jung and Katz, ECLAC, 2024/25). And where adoption exists, it tends to be superficial: in Argentina, a national survey of SMEs (nadIA — UTDT and Fundar, 2025) found that 41.6% use some AI, but concentrated in basic tools and with very low governance and internal-capability indicators. The problem, then, is not only that AI is missing: it is that the bottleneck is organizational, not technological. What is missing is the criterion to know who is ready, for what, and what to unblock first. Measuring maturity is the tool to close that gap, not to certify the few who have already arrived.
What I designed
A maturity model —IMIA (AI Maturity Index)— made up of 7 dimensions assessed on a 0–100 scale, from which a global level 0–5 is derived (0 Denial → 5 Augmented organization), with a scoring that produces a radial profile, the bottleneck and a leap potential indicator.
The 7 dimensions: Leadership · Data · Culture · Capabilities · Processes · Governance · Automation potential. The first six measure configuration (what the organization is); the seventh measures opportunity (what it could gain) and informs without penalizing —an immature organization may have very high potential, and that is a signal of priority, not a defect—. Potential distinguishes deterministic automation from the agentic layer: when software decides and acts on its own, the governance bar rises, because human judgment moves to the design and audit of the agent’s rules.
The six levels are the adoption ladder turned into measurement: levels 0–2 describe the silo (denial, shadow AI, pilots without value), level 3 the first real but isolated value, and 4–5 the governed architecture. The model says not only which rung an organization is on, but which is next and what to unblock to climb it.
That there are seven dimensions and not a single score is not arbitrary: the evidence on organizational readiness for AI is explicitly multidimensional —strategy, culture, talent, data, governance, infrastructure (Jöhnk, Weißert & Wyrtki, 2021, identify 18 factors across 5 categories). Having technology is not enough; you need the conditions to absorb it.
The key decision: governance as a gate, not as an average
The global level is not a simple average of the dimensions. The scoring implements capping rules (gates) that prevent inflating it with good grades on the easy parts:
| Gate | Rule | Why |
|---|---|---|
| Production | No Level ≥ 3 is assigned without real cases in production | Without real value there is no “functional adoption”: it cuts through the smoke of eternal pilots. It is Moore’s chasm turned into a rule —the pilot that thrills the early adopters dies crossing into the majority, and the jump from level 2 to 3 is that crossing. |
| Governance | No Level ≥ 4 is assigned if Governance < 40/100 | Without verifiable rules, there is no sustainable integration. It is the moat. |
| Data | Level ≥ 4 requires Data ≥ 50/100 | Without usable raw material, integration does not scale. |
The governance gate is the central piece: an organization with excellent leadership, data and culture but without AI usage policies, without risk management (privacy, security, bias, compliance) or clear rules, is capped at Level 3. Governance stops being discourse and becomes a hard constraint of the model. That is what separates an AI Governance Engineer from a policy consultant: governance is operationalized.
Its elegance lies in a dual mechanism: governance weighs in the index and governs as a cap. Weighing alone would make it negotiable (it can be offset by another dimension); governing alone would make it binary. The two together say: governance adds when it is present, and blocks when it is absent.
How it is calculated (scoring v1.0 summary)
- Each dimension: average of its items (0–4 scale) rescaled to 0–100.
- Global index: weighted sum of D1–D6 (Data 22% and Leadership 20% weigh more, as the strongest theoretical predictors; Governance weighs high because it is the moat). The Potential dimension informs but does not penalize.
- Conversion to level 0–5 applying the gates above.
- Leap potential = Data · 0.5 + Leadership · 0.3 + Capabilities · 0.2 → detects immature organizations with a usable base (“high potential return”): the case where the bridge pays off fastest.
- The v1.0 weights and cutoffs are well-founded but provisional by design: they are recalibrated empirically (reliability per dimension, factor analysis, regression on “cases in production”) and scoring v2.0 is versioned. Discipline of versioning and validation, not fixed opinion.
Output per organization (example profile — ILLUSTRATIVE, synthetic data)
Fictional data to show the output format. It does not correspond to any real organization.
Maturity profile — Example org. (synthetic)
Leadership ███████░░░ 72
Data █████░░░░░ 48
Culture ██████░░░░ 60
Capabilities ████░░░░░░ 40
Processes █████░░░░░ 52
Governance ███░░░░░░░ 32 ← bottleneck
Automation pot. ████████░░ 80 (high latent opportunity)
IMIA global: 51/100 → Level 3 (Functional adoption)
Cap applied: governance gate prevents Level 4 (Governance 32 < 40)
Diagnosis: high leap potential, held back by governance and data.
Recommendation: before scaling AI, establish usage and risk-management
policies (D6) and improve data accessibility/quality (D2).
The actionable reading is not “they are at Level 3”, but “their bottleneck is Governance and Data, not Leadership; their value ceiling is high” —exactly what a governed adoption roadmap needs in order to prioritize. Not a number: what to unblock first.
Why this case proves the role
- Governance: I designed a model where governance is a measurable gate, with auditable gates —operational governance, not theory. Direct study material for IAPP AIGP.
- Adoption: the output is an adoption roadmap prioritized by bottleneck and potential —the central conversation of an AI Adoption Architect.
- Rigor: versioned scoring, justified weighting, empirical recalibration plan; the weights that are a design decision are declared as such.
- Honesty: the method is published, not client data; the example is synthetic and labeled.
What I would do differently at scale
- Recalibrate weights and gates with real observatory data (scoring v2.0) and report reliability per dimension.
- Map each governance item to concrete controls of the EU AI Act and the NIST AI RMF, turning the diagnosis into a regulatory readiness assessment.
- Automate the radial profile and the individual report as a reproducible artifact.