The problem
Mid-sized companies are the backbone of the Mexican economy, yet their AI adoption is structurally stalled: only 38% of Mexican businesses have adopted AI in any meaningful form, and just 3% have reached advanced levels (AWS, 2025). The prize for closing the gap is quantified — 83% of Mexican adopters report revenue growth averaging 16%, and McKinsey (2025) documents a 37% average productivity gain in administrative functions. Applied across Mexico's ~228,000 mid-sized enterprises, systematic acceleration of AI adoption would carry national-scale economic impact.
The gap is not a technology gap. Existing frameworks were built for large enterprises in developed economies; mid-sized Mexican companies operate under a fundamentally different constraint profile, and generic playbooks routinely fail them.
The research
COMPASS was developed through a qualitative multiple-case study of five organizations across Mexico City and Guadalajara, spanning three AI maturity levels and four sectors — from early-stage adopters to an AI-native consultancy — with 65 semi-structured interviews across five stakeholder roles per organization (CEO, IT manager, functional manager, end-user, client/partner), plus document analysis and structured longitudinal observation. It synthesizes three established theories: the Theory of Constraints (where are you blocked?), the Resource-Based View (what do you have?), and Dynamic Capabilities (how do you move?).
The central finding
Constraint-aware, maturity-calibrated strategy — not technology selection — is the primary determinant of AI transformation success.
Four constraints dominate, and one binds all the others:
- Leadership commitment & AI literacy — the binding constraint across every case and maturity level. No technical investment compensates for its absence. Commitment (willingness to sponsor) and literacy (understanding what AI can and cannot do) are distinct, and success requires both.
- Digital talent scarcity — universal, but stage-specific: from zero internal competence at early-stage firms to specialized role shortages at advanced organizations.
- Data governance deficits — the most consistent technical barrier. COMPASS reframes it as a parallel workstream, not a prerequisite that delays the first pilot.
- Cultural resistance — rarely open opposition; it manifests as cautious detachment, passive non-engagement, and evaluation paralysis, each demanding a different countermeasure.
The framework
COMPASS locates an organization at one of three maturity stages and prescribes stage-proven responses:
| Stage | Dominant constraints | Priority responses | Success indicators |
|---|---|---|---|
| Explore (early) |
Leadership AI illiteracy; no internal competence; data not ready; consulting dependency; no champion | Discovery & diagnostic phase; narrow pilot on one high-value use case; designate internal champion; structure consulting for knowledge transfer | Pilot initiated; champion designated; literacy program launched; partnership structured for capability transfer |
| Implement (intermediate) |
Data fragmentation; OT–IT gaps; passive non-engagement; talent scaling; ROI measurement gaps | Minimum viable data standard; phased modular deployment; behavioral adoption co-design; internal AI learning program; ROI tracking | Governance workstream active; adoption rate tracked; second-tier champion; ROI evidence documented |
| Scale (advanced) |
Bureaucracy slowing execution; specialized talent scarcity; AI trust & explainability; sustaining learning culture | Capability platformization (reusable frameworks/SDKs); governance pre-engagement; iterative cycles; ethics & explainability frameworks; innovation metrics | Reusable assets deployed; governance redesigned; learning infrastructure institutionalized; ethical standards embedded |
What consistently works
- Build a learning & experimentation culture — encode each project's lessons into shared organizational assets.
- Sequence investments — one high-value, low-complexity use case; prove ROI; let evidence fund the next move.
- Empower an internal AI champion — real authority, leadership endorsement, co-leadership of delivery.
- Design partnerships for capability transfer — knowledge-transfer obligations and internalization milestones, not delivery-only dependency.
About the researcher
Daniel Risso Elliot is a doctoral candidate at Signum Magnum College (Austria) and a technology executive at a global enterprise software company, and advises Mexican mid-market firms on AI adoption. More at rissell.me.
Sources: AWS / Strand Partners, "Unlocking Mexico's AI Potential" (2025); McKinsey & Company, "Superagency in the Workplace" (2025); dissertation cross-case findings (Chs. 4–5).