AI Adoption Metrics Are Broken. Watch Your CMO.

TL;DR

• In the best companies Keith Rabois sits on, the #1 consumer of AI tokens is the CMO. Not engineering.
• This is the right signal: AI has dissolved the deputy chains that forced executives to outsource the execution of their own judgment.
• Measuring adoption by seats and training completions misses the actual shift. The real metric is whether senior people have started producing work product directly.

---

The number one consumer of AI tokens in high-performing companies isn't engineering. It isn't data science. It's the CMO — and that's the right signal for measuring whether AI adoption is real.

Keith Rabois sits on the boards of Stripe, Ramp, Airbnb, and DoorDash. His observation from a recent appearance on Lenny's Podcast: the best organisations he's seen share this pattern. Their CMOs top every other function on token consumption, because they've stopped routing work product through deputies and started producing it directly.

"They don't need to rely upon deputies and deputies and deputies to get actual work product."

He named two companies where this is happening, both large and performing well. The pattern explains something that most AI adoption playbooks miss.

What the deputy chain actually costs

Every senior leader routes work through deputies. The maths never worked otherwise. A CMO holds strong judgment about what a campaign should say, where the positioning should land, what the market data means. Acting on that judgment takes copywriters, analysts, designers. People who have to be briefed, given feedback, revised, and managed.

The deputy chain exists to bridge the gap between a leader's judgment and the output that expresses it. At every handoff, some of the original intent gets interpreted, filtered, softened. The CMO approves a brief that's 80% of what they had in mind because 100% would require two more rounds of feedback nobody has time for.

The effective output of a CMO's judgment has always been capped by their deputy chain's throughput.

How AI dissolves the deputy chain

The CMO who knows what they want now has a mechanism to produce it directly. AI isn't smarter than a deputy. The difference is that the CMO's business context never has to be explained to an intermediary. Judgment and execution happen in the same brain, in the same sitting.

When the bottleneck was always the execution gap, and AI closes the execution gap, the person with the sharpest judgment gets the most upside. The CMO has broad scope (campaigns, positioning, briefs, analysis, executive communication), the clearest sense of what good output looks like, and was previously the most constrained by the chain between intent and result.

Under this model, the CMO produces the draft and refines it themselves in a single session. Output is better for the same reason good editors make better first drafts than junior writers: the person with taste and the person producing are the same person.

Why this is the signal that matters

Most organisations measure AI adoption through tools deployed, seats licensed, and training completions. These measure whether people have access to AI. They don't measure whether AI has changed how work actually happens.

The real signal is whether AI has reached the decision layer. Whether the people with the broadest business context and the clearest sense of what good looks like are now producing output directly, rather than reviewing and approving work that filtered up through deputies.

The CMO-as-top-token-consumer is that signal. It means AI has dissolved a deputy chain, not just accelerated the people already in it.

I see a version of this from a different angle. Building two production SaaS platforms solo over the past year, there was no marketing function, no design team, no analyst pool. Everything that normally flows through a deputy chain was mine to produce or skip. What I found was that the constraint was never execution. It was clarity of judgment. When I knew precisely what I wanted, AI materialised it quickly. When I didn't, no volume of token spend helped.

The builder-leader identity is partly about this: closing the gap between strategic judgment and its expression in output. The CMO who has bypassed three deputies is living that identity at scale.

What AI adoption actually looks like when it's working

If you want to know whether AI is working in your organisation, stop looking at platform dashboards. Three questions give you the real picture.

Who is producing work product they previously delegated? A CMO writing their own campaign brief. A CFO synthesising the board narrative directly. A CPO drafting the product brief before briefing a designer. These are the signals that matter. They show AI has reached the decision layer, not just the execution layer.

Where has the approval cycle shortened because the approver is also the producer? The normal handoff creates delay and interpretation loss. When the same person produces and approves, both disappear. If a function's review cycles have shortened without obvious quality loss, someone closed their deputy chain.

Which functions are generating more output from the same headcount? Not faster junior work, which AI also enables. More output from senior people. Marketing shipping more campaigns without adding writers. Strategy producing more analysis without adding analysts.

Those signals matter more than monthly active users on your AI platform. Seats tell you people have access. Token consumption at the executive layer tells you AI has dissolved a deputy chain. The constraint-driven adoption strategy works partly because it forces people to close the execution gap themselves rather than waiting for a deputy to close it. The CMO who does this voluntarily is ahead of the curve. The organisation that measures whether it's happening is ahead of most.

For leaders who haven't yet built that hands-on intuition, AI strategy without personal exposure is the most common reason adoption stalls at the leadership layer.

---

Frequently Asked Questions

Doesn't this mean executives are just doing work that should be delegated?

The deputy chain exists to scale execution, not primarily to develop junior talent. When a CMO writes a campaign brief directly, they're removing an interpretation layer between their judgment and the output, not blocking a junior writer's development. Production work (design, detailed formatting, sourcing, channel execution) still lives in the team. The chain that translated executive thinking into a first draft is what AI has replaced.

What happens to mid-level roles if executives start going direct?

Roles that primarily served as conduits for executive intent are under pressure. Roles that provide genuinely distinct value (specialist production capability, institutional memory, coordination across complex systems) are not. The people in the middle who were mostly managing the translation from judgment to output face the sharpest pressure.

Is this pattern specific to CMOs?

Rabois described it in the context of CMOs, but the underlying mechanism applies wherever the bottleneck between judgment and output is an execution chain rather than a genuine complexity ceiling. CFOs synthesising financial narratives, CPOs drafting product briefs, strategy teams building research summaries. Any role where senior judgment has historically been filtered through deputies before becoming work product is a candidate for the same shift.