Three Leaders, Not One: MIT CISR's Repsol-and-BBVA Benchmark for Scaling AI Beyond the Single Champion

Executive Summary

• MIT CISR’s February 2026 research briefing names a structural pattern that separates organizations that scale digital and AI innovation from organizations that consume budget without impact: three coordinated leadership roles, not one executive owner. (Fonstad, Mocker, Salonen — n=277 organizations surveyed with EY, 50+ executive interviews, 2016–2025 case studies.)
• Initiative leaders generate evidence that an offering is desirable, feasible, and viable. Shared resource leaders build reusable platforms (data, AI, UX, cybersecurity, legal, risk) that initiatives draw on. Portfolio leaders realign and rank initiatives quarterly against strategic priorities and reallocate scarce talent and funding.
• The Repsol benchmark is the hard mid-market target: 505 digital initiatives launched, 76% scaled successfully over five years, €800 million realized, 20% profit increase, 2x return on investment. The data and AI hub helped over 60% of successful initiatives use analytics for cash-flow contributions.
• The BBVA benchmark anchors portfolio discipline: more than 2,000 initiatives realigned to strategic priorities each quarter, about 10% discontinued every quarter when no longer aligned. 40 staff globally manage the Single Development Agenda; 250+ staff sit in business units.
• The core recommendation challenges the most common American mid-market deployment pattern: “Assigning responsibility to a single executive or unit risks underinvesting in a broader leadership capacity required to deliver impact systematically.” The single AI champion model is the failure mode this framework names.

The Single-Champion Failure Mode

Most American mid-market AI programs in the 200–2,000 employee band run on one executive. A “Head of AI,” a CIO who took on AI as an additional brief, or a CTO who hired an AI lead. This person drives the early wins, the board updates, and the vendor selection. The pattern delivers two or three pilots with measurable improvements, then stalls. McKinsey’s State of AI (November 2025, n=1,993) finds only 6% of organizations are high performers at >5% EBIT impact from AI, and McKinsey’s “Seizing the Agentic AI Advantage” reports ~80% see no P&L impact despite 78% adoption.

MIT CISR names what fails when a single executive owns innovation: “Most organizations struggle with digital innovation, as their approach consumes resources out of proportion to the impact it delivers.” The single-leader pattern compresses three different decisions — what to build, what to share, what to fund — into one person’s calendar. The decisions are different, the trade-offs are different, and the time horizons are different. Forcing them into one role produces under-investment in shared platforms, weak portfolio discipline, and initiative leaders who lack the autonomy to test and learn.

The MIT CISR finding pairs with Pass 451 MIT CISR Enterprise IT Operating Models (Thorogood) and Pass 449 Forrester AI CIO (Moccia), which both name the CIO operating-model shift toward outcome governance. Leading Digital Innovation supplies the structural answer to “what does outcome governance look like as an org chart”: three leadership roles, coordinated against strategic priorities.

The Three Leadership Roles

Initiative Leaders

Initiative leaders own offerings end to end. The Repsol pattern is two co-leaders per initiative: a product owner from the business who is accountable for desirability and viability, and a technical lead from the Digital Program who is accountable for feasibility. The product owner answers “do customers, employees, or partners actually want this and does it contribute to strategic ambition.” The technical lead answers “will it work reliably at scale.”

Initiative leaders progress through phases by generating evidence — not by writing more comprehensive plans. Repsol uses a five-stage funding process with a venture-style investment committee that reviews evidence between stages. The trade-off this role manages: increasing reliability at scale versus first-to-market speed.

Shared Resource Leaders

Shared resource leaders run the platforms initiatives draw on. At Repsol, “hubs” concentrate technology and talent in data and AI, robotics, and user experience design. Each initiative’s technical lead is staffed from the relevant hub. The data and AI hub helped over 60% of successful initiatives use analytics for cash flow contributions.

The functional scope is broader than IT: human resources, risk and compliance, and legal sit in the shared resource layer alongside cloud, data, AI, UX, and cybersecurity. The trade-off this role manages: speed to market versus building shared, reusable resources. A shared resource leader who is too accommodating produces bespoke work that does not productize; one who is too rigid blocks initiative velocity.

Portfolio Leaders

Portfolio leaders are top-level executives who realign initiatives against strategic priorities at a regular cadence. BBVA does this quarterly across 2,000+ initiatives. About 10% of initiatives are discontinued each quarter when no longer aligned with strategic priorities. The portfolio team is small relative to the operation: 40 staff globally for the Single Development Agenda process, with 250+ staff sitting in business units.

The trade-off portfolio leaders manage: attracting customers quickly versus reducing cost to serve. The discipline is not picking winners; it is using transparent criteria to reallocate scarce talent and funding away from initiatives evidence has weakened, and toward initiatives evidence has strengthened.

What the Benchmarks Mean for a 500-Person Company

The Repsol and BBVA case data is large-enterprise scale. The structural pattern is mid-market transferable, but the resourcing is not. Three translation rules:

Initiative leaders are part-time roles, not new hires. A 500-person company with 8–12 active AI initiatives has business unit leaders who already own the customer or operational outcome. The shift is naming them as the desirable-and-viable owner with explicit authority to kill an initiative if evidence weakens.

Shared resource leadership starts with one hub, not three. Repsol concentrated talent in data and AI, robotics, and UX. Mid-market starts with one — data and AI — staffed by 2–4 people, accountable for the platform every initiative draws on. The cybersecurity, legal, and risk shared services already exist; the work is connecting them to AI initiative reviews, not standing up new functions.

Portfolio cadence matters more than portfolio scale. BBVA’s 10% quarterly discontinuation rate is the executable signal. A 500-person company running 10 AI initiatives should expect to kill one each quarter when evidence does not support continuation. If zero initiatives are killed in a quarter, the portfolio is not being managed — it is being preserved.

How This Connects to the Workforce Cluster

Pass 462 Deloitte Global Human Capital Trends (n=9,000+, Mar 2026) finds organizations using human-centric AI strategies are 1.6x more likely to exceed investment-return expectations. Pass 457 Deloitte team-dynamics (n=1,394, Feb 2026) finds team-structure variables — size, cognitive diversity, cross-functional composition — predict AI-derived gains. MIT CISR’s Leading Digital Innovation completes the cluster: leadership-role structure is the third dimension. Workforce orchestration, team structure, and leadership coordination are the operational answer to McKinsey’s 6% high-performer gap.

Key Data Points

What This Means for Your Organization

If your AI program runs through one executive — a Head of AI, a CIO with the AI brief, or a CTO who hired an AI lead — MIT CISR’s data names the structural ceiling. The single-leader model produces early wins and stalls at scaling. The three-role pattern is what the high performers do differently. The translation work for a 200–2,000 person company is naming who plays each role with the authority to act, not adding three new hires.

Three diagnostic questions executable Monday morning. First: for each active AI initiative, who has explicit authority to kill it if evidence weakens? If no one does, you have research projects, not initiatives. Second: which one shared resource — data and AI is the most common starting point — has a named leader accountable for the platform every initiative draws on? If no one is accountable for the shared layer, every initiative builds its own and the productization advantage Repsol captured is unavailable. Third: at what cadence are AI initiatives realigned against strategic priorities, and what was the discontinuation rate last quarter? If the rate is zero, the portfolio is not being managed.

If these three questions raised structural decisions specific to your organization — the politics of naming kill-rights, the build-out of the first shared hub, the establishment of a quarterly portfolio review with real teeth — I’d welcome the conversation. brandon@brandonsneider.com.

Sources

Primary source:

• Fonstad, N. O., Mocker, M., & Salonen, J. (2026, February 19). Leading Digital Innovation. MIT CISR Research Briefing No. XXVI-2. https://cisr.mit.edu/publication/2026_0201_LeadingDigitalInnovation_FonstadMockerSalonen
• Methodology: 277 organizations surveyed in partnership with EY; 50+ executive interviews; case studies developed 2016–2025.
• Credibility: HIGH. MIT CISR is academic source with primary survey, longitudinal case studies, and named case organizations (Repsol, BBVA, Posten Bring). Disclose: EY partnership on the survey component. The case data is large-enterprise; mid-market translation requires explicit re-scoping rather than direct application.

Triangulating sources from corpus:

• McKinsey (2025, November). The State of AI: How organizations are rewiring to capture value. n=1,993. 6% high-performer rate (>5% EBIT from AI).
• Deloitte (2026, March 4). 2026 Global Human Capital Trends. n=9,000+. Human-centric AI strategies 1.6x more likely to exceed investment-return expectations.
• Deloitte (2026, February 27). Bridging the AI value gap: Are team dynamics the missing link? n=1,394 U.S. workers. Team-structure predictors of AI-derived gains.
• MIT CISR (2026). Enterprise IT Operating Models in the AI Era (Thorogood). CIO operating-model shift to outcome governance.
• Forrester (2026, April 9). The AI CIO Will Govern Outcomes At Scale (Moccia). CIO as outcome governor, not tooling buyer.

Brandon Sneider | brandon@brandonsneider.com April 2026