Mollick Says HR Should Lead the AI Transformation. He's Half Right — the CFO Signs the 10-K.

By Errin O'Connor — Founder & Chief AI Architect, EPC Group · Microsoft Solutions Partner across all six designations · G2 Leader in BI consulting

At a workforce summit this year, Ethan Mollick made an argument designed to raise eyebrows, and it deserved every eyebrow it raised: HR — not IT — is the function best positioned to lead organizations through the AI transformation. His logic is sound. AI adoption is a people problem before it is a platform problem: incentives, fear, role redesign, hidden power users, change management. His “leadership, lab, and crowd” framework for AI success is fundamentally an organizational-behavior framework, and organizational behavior is HR's home field. Meanwhile the industry data agrees with him from the other direction — LinkedIn's policy leadership has been saying all year that integrating agents into human teams “is not a technology job.”

I agree with all of it. And I have watched enough enterprise transformations across twenty-nine years to add the correction that determines whether Mollick's advice produces a transformation or a workshop series: leading the change and owning the exposure are two different jobs. HR can lead the change. Only one officer owns the exposure — and that officer signs the 10-K.

The three-body problem of AI ownership

Every enterprise AI program lives inside a triangle. The CIO owns the platform — tenants, capacity, identity, integration. HR owns the people — skills, roles, incentives, adoption, the shadow users Mollick talks about. And the CFO owns the boundary — what a probabilistic system is permitted to touch in a financially material workflow, at what threshold, with what evidence. The transformations that fail are almost never missing a leader. They are missing an owner of the boundary. I have sat in the postmortems: the CIO delivered the platform, HR ran a genuinely good enablement program, adoption climbed — and eighteen months later nobody could tell the audit committee which decisions the AI had been making, against which data, under whose authority. Everyone led. Nobody owned.

Here is the asymmetry that settles the question. When an AI-assisted forecast cannot be reproduced in an audit, when an agent acts against the vendor master in violation of a segregation-of-duties control, when a model inserts a number into a financially material workflow — the CHRO is not the officer whose signature and control environment sit at the center of the review. Neither is the CIO. The CFO's signature is on the filings. The CFO's control environment is the one regulators, carriers, and plaintiffs examine. Fiduciary gravity does not follow the org chart of enthusiasm. It follows the signature.

So run it like a three-legged race — with one convener

The operating model I give clients keeps everything right about Mollick's argument and fixes the one thing it leaves loose. HR leads the human transformation — the crowd, the lab culture, the amnesty for shadow users, the role redesign. The CIO leads the platform — identity for agents, governed data, capacity, logging. And the CFO convenes: a two-hour, quarterly, documented, board-visible meeting that decides the six things neither HR nor IT can decide alone — the organization's AI ambition, the data AI may touch, the autonomy it may exercise and at what dollar thresholds, the named executive accountable for outcomes, the insurance posture, and the multi-model position.

Editorial note — methodology: The observation that a substantial share of CFOs at mid-market enterprises have never held an explicit AI-governance meeting is drawn from EPC Group advisory conversations over the trailing twelve months. The original draft cited “nearly half” of CFOs interviewed. Before this page is promoted externally, a formal methodology note will be added here covering: approximate sample size, interview period, organization types and sizes, geography, the nature of the conversations (advisory / sales discovery / formal research), and the precise definition of “held a governance meeting.” Until that block is published, the claim is presented as a directional finding from advisory conversations, not a statistically validated survey result.

And when the named-executive question comes up — item four on that agenda — the honest answer for most organizations between $250 million and $5 billion in revenue is that they need an accountable AI officer and cannot yet justify a full-time one. That is exactly the gap a fractional Virtual Chief AI Officer exists to fill: chairing the quarterly meeting, owning the autonomy matrix, briefing the audit committee, and being the human signature that says this program is defensible. Rent the officer while you decide whether to hire one. What you cannot do is leave the seat empty and call the enablement program a strategy.

What I tell clients to do

One. Give Mollick's argument to your CHRO and ask them to own the crowd — formally, with budget. He is right that the people work is the hard work, and it deserves a real leader.

Two. Give the boundary to the CFO — formally, in the governance charter. The convening authority for the quarterly AI meeting sits with the officer who signs the filings.

Three. Fill the named-executive seat — full-time Chief AI Officer above roughly $5 billion in revenue, fractional vCAIO below it, never “the CIO by default,” because platform accountability is not outcome accountability.

Four. Put all three in the same room quarterly, with the General Counsel, and deliver the minutes to the audit committee. Two hours. That is the entire ask.

Where I land

Mollick is right that this transformation will be led through people, and HR should take that half of the argument and run with it proudly. But enterprises do not get sued, audited, non-renewed, or subpoenaed through their enablement programs. They get sued through their financial controls — and the officer who owns those controls is the only one with the standing to convene the meeting that bounds the exposure. HR leads the change. The CFO calls the meeting. Get both right and the transformation has a spine as well as a heart.

Multiple models. One truth.

The data behind this (sources and verification)

• Mollick's Valence keynote (HR best positioned; leadership/lab/crowd framework) — Argument that HR, not IT, is best positioned to lead AI transformation; the crowd, lab, and leadership framework for organizational AI success.
• LinkedIn policy leadership (VivaTech 2026) — Integrating agents into human teams "is not a technology job"; approximately 70% of job skills changing by 2030.
• EPC Group advisory findings — In our advisory conversations, a substantial share of CFOs at mid-market enterprises have never convened an explicit AI-ambition governance meeting. See methodology note in article body. The live CFO article is the canonical write-up: /insights/cfo-ai-governance-conversation-2026.
• Workday ruling (Reuters, June 22, 2026) — AI hiring tools found reachable under civil-rights law — the concrete case where HR-owned AI tooling produced CFO/GC-grade legal exposure.

Third-party figures above are attributed to their named sources as of the Last verified date. EPC Group audit figures are directional findings from client engagements. Items marked [VERIFY] must be confirmed before external quotation.

Frequently asked questions

Multiple models. One truth.