Ask yourself a simple question about any AI feature running inside your GTM stack right now: What number would prove it’s working, and who’s checking it?
If you can’t answer immediately, you’re not alone. I couldn’t either, on a project I sponsored myself.
At one company, I pushed for an initiative to pull customer call transcripts out of Zoom, Clari, Mindtickle, and Teams into Snowflake, then use AI to surface patterns and route relevant insights to CS, Product, and Marketing. The idea made intuitive sense to everyone who heard it. Renewal risk sitting in a call nobody replayed. Product feedback buried in a transcript nobody read twice. Patterns across customers no person could find on their own. I worked with the data team to scope the MVP, and I had executive support to build it.
What I never built was a business case naming the number the project was supposed to move, or the person accountable for reading it once the project launched. I left the company before it reached production. I don’t know what happened to it, but I’d guess it’s dead, somewhere in a backlog, not because the idea was bad but because nobody, including me, had written down what success meant.
That’s the more common failure mode in GTM AI right now. Not a bad result. The absence of any result anyone can name.
Intuition Isn’t a Business Case
“Feels important” has been doing an enormous amount of work across GTM AI over the past two years, standing in for the harder step of naming a metric and a date to check it.
MIT’s Project NANDA studied 300 enterprise AI deployments in 2025 and found that 95 percent produced no measurable return, despite $30 to $40 billion in enterprise GenAI spending across the initiatives it covered. The five percent that worked weren’t running better models. They were measuring against an agreed number from the start.
A year later, the picture hasn’t improved. McKinsey’s 2026 State of AI survey, published this August, found that only 37 percent of the 1,719 executives it surveyed attributed any earnings impact to AI, essentially unchanged from 2025. Just 6 percent qualified as what McKinsey calls AI high performers, organizations attributing at least 5 percent of EBIT to AI and calling the impact significant, also flat year over year. Eighty percent of individual employees said AI made them more productive. Almost none of that productivity had reached the income statement.
Two companies made this expensive and public this year. Klarna spent 2024 telling the market its AI assistant handled the work of 700 support agents, then spent 2025 quietly rebuilding its human support team once complaints accumulated. Ford leaned on automated quality-control systems for years, then rehired hundreds of the engineers those systems had replaced once defects and warranty costs made the gap too large to explain away. Both got their measurement eventually. It took a public, expensive wrong answer to force what should have existed on day one.
This is exactly the kind of gap our AI Revenue Risk Assessment is built to surface, before an AI initiative disappears without ever proving its value: see how it works →
What Happens When the Sponsor Leaves
My transcript project didn’t fail because the plan was flawed. It was vulnerable because its entire case for existing lived in my head, not in a document anyone else could reference once I was gone.
A project with a sponsor and no metric survives exactly as long as the sponsor does.
Most GTM organizations are carrying several versions of this today: An AI feature a VP championed at a QBR. A pilot a founder greenlit after a conference demo. A tool procurement approved because a rep swore by it. None of these are failures yet. Each one is a single departure away from becoming one, quietly, because the metric that would have caught the problem was never recorded.
The Same Mistake, Different Side of the Table
This same failure shows up on the vendor side of the table, described in different language. Jamin Ball, a general partner at Altimeter Capital, coined the term ERR in 2024, experimental recurring revenue: Dollars booked from customers who are still trying a product, not yet committed to keeping it. Cassie Young, a general partner at Primary Venture Partners, picked up his argument this fall and pushed it further, predicting what she calls a “gross retention apocalypse,” a wave of AI vendors about to lose the customers who signed on to experiment and never got proof the experiment was worth keeping.
Young has her own example of what happens when a company measures the wrong number instead of the right one. Early in her career, she watched a company post net revenue retention above 120 percent while gross retention sat at 60 percent. The topline figure looked outstanding. The number underneath it said the company was losing well over a third of its customers every year, and covering the gap with expansion revenue from whoever stayed. Measuring the wrong number turns out to be its own version of not measuring at all.
ERR and an unmeasured internal AI initiative are the same mistake, seen from two different seats. A vendor selling AI without proof it holds up over time. A buyer running AI without proof it’s earning a return. Neither side has done the work, and both are operating on the same unexamined assumption: AI is obviously worth it, so proving it can wait.
Where This Leaves the ROI Question
Every post in this series has pointed at some version of the same blind spot. This one is the most avoidable version of it: You can’t prove ROI on an initiative to which nobody assigned a metric in the first place.
List every AI feature or initiative currently running inside your GTM stack, purchased or built. Next to each one, write the metric that would prove it’s working and the name of the person accountable for checking it. If either column comes up blank, you have more work to do.
This is the fifth post in an eight-post series on why most B2B SaaS companies can’t answer the AI ROI question, and what closes the gap. If you want a scored, five-minute read on where your own organization stands, the AI Revenue Risk Assessment gives you the diagnostic: take the assessment →
Andrea Mulligan is a B2B SaaS executive and advisor with 30 years of experience building Customer Success, Professional Services, and GTM organizations. She works with PE-backed and growth-stage companies on CS transformation, revenue retention strategy, and post-sale model design. Start a conversation →