The 40% Problem: Carl Eschenbach on Why Most AI Bets Fail — and What Smart Revenue Leaders Do Differently

Carl Eschenbach has watched more platform shifts than most people have had careers. Thirty-eight years in enterprise technology — from PBX phone systems to VMware's rise from two hundred people to twenty thousand, from Sequoia Capital to the CEO chair at Workday — have given him a particular kind of clarity about what separates companies that transform from ones that merely dabble. We sat down with him at Captivate '26 in Austin to talk about AI, the pressure cooker every revenue leader is sitting in right now, and the single question he says most teams forget to ask before they spend a dollar.

<hr>

There's a Gartner projection making the rounds that should give any revenue leader pause: by 2027, forty percent of agentic AI projects will be canceled due to a combination of poor governance, unclear ROI, and runaway costs. When Mark Schopmeyer, Co-CEO of CaptivateIQ, put that number to Carl Eschenbach on stage in Austin, Carl didn't flinch.

"It doesn't surprise me," he said. "There was an MIT report last year that found ninety-five percent of corporate AI pilots were failing. That was when everyone was in proof-of-concept mode, just experimenting."

And the problem, he argues, hasn't changed much. Now, it just wears a more expensive suit.

"One of the issues with experimentation is people don't have an outcome in mind. They start because there's a lot of top-down pressure to just go do it. But you also have to be careful — if you over-experiment and it doesn't drive business value, what did you really accomplish?"

It's a pointed observation, and one that lands differently coming from someone who spent three and a half years as CEO of Workday, one of the most consequential enterprise software companies in the world, before returning to venture at Sequoia. Carl has been in the rooms where these decisions get made — and more importantly, where they go wrong.

Start at the Finish Line

Most teams start with the experiment. Carl opts for the finish line.

"I start with the end goal in mind and work backwards from there," he says. "What are we trying to achieve during this experimentation? And at the end of that evaluation, if we meet the deliverable and it's driving business value, then we move forward."

That means, before a single agent gets deployed or a proof of concept gets scoped, there should be a clear answer to why.

"Any company buying a technology, there's one of three reasons it'll make it through the sales cycle and into production," he says. "Show me how it saves money with a measurable ROI. Show me how it accelerates growth — no CEO or CFO in the world will say no to that. Or make the case for risk avoidance. Think about security: people aren't always buying to save money. It's an insurance policy."

For revenue and comp leaders, this framework isn't just useful for evaluating vendors. It's a forcing function for internal AI projects too. What bucket does your automation initiative fall into? If you can't answer that clearly, Carl suggests you don't have a project — you have a hobby.

The Build vs. Buy Trap

One of the thorniest decisions revenue ops teams face right now is whether to build internal AI tooling or buy it from a vendor. Carl has watched this play out at every company he's touched, and his answer may frustrate the engineering optimists in the room.

"Most of the time when you bring a technology decision to a technical team, they say, 'We can do it ourselves,'" he says, with the weariness of someone who has heard this approximately nine hundred times. "And my next question is always: How long will it take, and how much will it cost?"

The follow-up is where the conversation gets interesting. What's the net present value of moving in six months instead of eighteen? What happens to your competitive position while your team is building something a vendor has already hardened across hundreds of enterprise deployments?

"It's not an ‘or’ answer, it's an ‘and’," he allows. "There are times when you have domain-specific knowledge where you want to build your own agents. But I would always challenge people: if you're gonna do it yourself, how long will it take and how much will it cost? And then ask, is this part of your core competency? What is your business? You should be using your resources to grow your business. Not rebuilding what's already out there."

Find Your 20%: The Change Agent Theory

Even leaders who get the strategy right often underestimate the human side of AI adoption. Eschenbach reaches for a change management framework he's used for years, and it's blunter than most.

“When you're driving change, your people will respond in one of three ways. Sixty percent will be paralyzed — they don't know what this means for them. Twenty percent will fight it and resist. And the last twenty percent are what I call change agents. They'll drive things forward.” 

His advice: stop trying to convert the sixty percent first. Find the last twenty.

"Identify those change agents within your organization. Highlight them. Put them front and center. Build a center of excellence around them to drive the transformation."

For comp and revenue ops leaders, this is practical. The change agents on your team are probably already tinkering — building small automations, prompting their way through tedious reconciliations, asking questions about what AI could do to the quota-setting process. They don't need permission to be curious. They need visibility and resources.

Carl is also direct about the cost of misreading this dynamic.

[BLOCKQUOTE
| Quote: The most dangerous thing you can do in regard to your future is to rely on your past success. There's a reason the windshield is bigger than the rearview mirror.
| Author: Carl Eschenbach
| Title: Partner, Sequoia Capital
]

On AI Sprawl and the Governance Problem

Speed is Carl’s north star — he calls it, alongside simplicity, one of the two greatest business strategies of all time — but he flags a risk that enterprise leaders are only beginning to reckon with.

"What's happening in the enterprise today is that it’s very easy for any of your employees to get or build an agent, start using it on your computer, and give it access to privileged data." He calls this AI sprawl, and he's seen it accelerate fast. "Don't let that happen. Because if it's not under the purview of a center of excellence, someone who has governance over AI, or your security teams, you can get in trouble."

For comp and revenue ops teams specifically, this matters a lot. Compensation data, including  individual payouts, plan mechanics, quota attainment, and pipeline assumptions, is among the most sensitive in any organization. The convenience of a rogue AI tool is not worth the exposure.

"I'm talking out of both sides of my mouth," he admits in a moment of characteristic candor. "Speed, speed, speed. Leverage AI. But have the right security framework around it to protect the enterprise and your data.” 

What Stays Human

This kind of conversation often ends with a prediction about how many sales roles AI will eliminate. Carl isn’t interested in that version.

"I think AI is going to amplify human potential, not replace it," he says. "If you go back and look at what drives corporate growth, the single biggest driver over the last thirty to forty years is productivity gains. And the biggest driver of productivity gains is technology. But we've only averaged one to two percent gains a year. I think AI changes that — maybe three or four percent. That sounds small. It's not small."

"People still want to do business with people,” Carl continues. “You want to know who's behind the technology. You want to trust them. Are they running their business with integrity? Do they have security and compliance controls? That human factor — that's not going away."

What AI does, he argues, is clear the decks. The mundane work, the reconciliation, the reporting, the administrative weight that buries sales and comp teams — that's what gets automated. What's left is the work that compounds: relationships, judgment, strategy.

"AI is the great equalizer," he says. "It takes all those trivial tasks away from what we're doing every day and allows us to focus on higher-level things."

For revenue leaders sitting on the fence about how aggressively to move, Carl’s message is simple, and he's been delivering versions of it since 1987: the future doesn't wait.

"If the world outside your company is moving faster than you are, the end is near. Change is inevitable. Whether you grow from it — that part's up to you."

Want more from Carl? Watch Captivate 2026 on-demand to view his entire keynote, as well as sessions from other industry leaders. 

<hr>

Carl Eschenbach is a partner at Sequoia Capital and a board member at Palo Alto Networks, Grafana and formerly on the board of Snowflake, UiPath, Workday and Zoom. He served as CEO of Workday from 2022 to 2026 and spent fourteen years as President and COO of VMware.

‍