📊 The $2 Billion Question: Why 78% of CMOs Don't Trust AI to Tell Them What's Coming Next (And the 22% Who Do)

By Dr. Elise Marchand, Ph.D.

Published in Digital Strategy Review, 2025

Every quarter, I sit down with marketing executives at conferences, roundtables, and private dinners. The conversations rarely start with budgets or channel strategy. They always land somewhere around the same question: How do you know what's going to happen next?

In other words—how do you forecast?

For decades, revenue forecasting in marketing was a sacred art form. A seasoned director with fifteen years of pattern recognition, a spreadsheet full of seasonality adjustments, and an almost supernatural feel for "market temperature" was the gold standard. You could argue with her numbers, but you wouldn't doubt them. The human brain had context that no algorithm could replicate—intuition built from watching thousands of campaigns land or miss.

Then we started giving those same executives access to predictive models trained on millions of data points. And something unexpected happened. Most of them got less confident in their forecasts.

That's the paradox I want to unpack. To dig into it, my research team and I surveyed 50 CMOs across mid-market and enterprise companies—SaaS, e-commerce, consumer packaged goods, financial services, and healthcare. We asked a deceptively simple question: Do you trust AI to predict your next quarter's revenue?

The answer was genuinely surprising—and it revealed something important about how marketing leadership is navigating one of the most consequential transitions in our industry.

The Numbers That Stopped Us Cold

Let's get straight to the data, because this is where the story gets interesting.

So: 40% of the CMOs we surveyed (those in the "fully trust" and "trust with review" buckets) would say AI plays a meaningful, relied-upon role in their revenue forecasting. The other 60% use it as a reference point at best—or treat it with active suspicion.

If you're in marketing leadership and expected that number to be higher—given how much we've invested in martech, data platforms, and "AI-powered" analytics suites—you're not alone. We were surprised too.

And here's the nuance that made the survey results even more interesting: the CMOs who fully trusted AI (the 8%) weren't necessarily the ones with the most sophisticated data infrastructure. Two of those four came from mid-market companies with fewer than 200 employees and relatively simple funnel structures. The enterprise CMOs—the ones with the biggest data teams, the most expensive BI tools—were actually less likely to say they fully trusted their AI forecasts.

That's a counterintuitive finding that says a lot about where the real barriers live.

Why More Data Didn't Mean More Trust

When we followed up with the skeptics—and even some of the moderate-trust respondents—a pattern emerged that I think is worth naming clearly, because it cuts against a common assumption in our industry.

The CMOs who distrusted AI weren't saying "the numbers are wrong." In most cases, they said the opposite: the numbers looked plausible. That's what made them uneasy. A forecast of $14.2M next quarter "looked right" to their data teams but didn't match what the CMO felt was coming. And because they couldn't fully explain why the model produced that number, they defaulted to human judgment—which, in a boardroom, still feels like the safer bet when your name is on the P&L line.

This is what I call the plausibility trap.

In classical statistics, we'd say a model with high R² but low interpretability creates an epistemic gap. The output looks statistically sound. It fits historical patterns well enough to pass peer review. But it doesn't explain the causal chain in a way that lets a decision-maker feel ownership of the number. And when you're presenting a revenue forecast to a CFO or a board, ownership is what matters more than accuracy. You need to be able to say: "Here's why we expect $14M." Not: "The neural net said so."

A CMO in our SaaS cohort put it well: "My VP of Analytics will defend the model for three hours. But when my CEO asks 'why,' I'm standing there saying 'the algorithm saw patterns.' And my CEO has a PhD. She can tell you're hiding behind the machine."

That sentence—hiding behind the machine—captured the tension better than any chart we could build.

The 22% Who Actually Use It (and What They Do Differently)

This brings us to the more encouraging finding: the CMOs who do trust AI forecasts aren't trusting them blindly. They've built a workflow that most of their peers haven't yet figured out.

Across our interviews, three practices separated the high-trust group from everyone else:

1. They treat AI as a hypothesis generator, not an oracle.

The CMOs who used AI forecasts effectively described their process in terms of falsification. The model gives them a baseline number—$14M, say. Then they spend the next two weeks asking: what would have to be true for that to be right? What assumptions about conversion rates, churn, and deal velocity are baked in? Where does the model's confidence interval widen? They're not accepting the output; they're stress-testing it. One CMO compared it to how a doctor uses an imaging scan: "The MRI shows something. But I'm still the one deciding if it's cancer or a benign cyst."

2. They anchor forecasts in operational reality, not just historical data.

This is where the mid-market CMOs we interviewed surprised us most. Their funnels were simpler, their data cleaner, and—crucially—they could walk through every step of the funnel by name. "I know my sales team has three reps on parental leave this quarter," one said. "The model doesn't know that. So I adjust for it." The enterprise CMOs had more data but also more layers between the raw signal and the business reality, which paradoxically made the AI's outputs feel less grounded in their specific operational context.

3. They've made interpretability a procurement requirement.

Perhaps the most actionable finding: the CMOs who trusted their tools were the ones who had negotiated for it. Their vendors provided feature attribution reports, sensitivity analyses, and plain-English explanations of which variables drove the forecast. One CMO in financial services told us she rejected a $200K/year analytics platform because "the dashboard gave me numbers but no reasons." She ended up with a cheaper tool that let her ask why for any given output.

This last point matters because it's a procurement decision, not just an analytical one. Most marketing teams buy forecasting tools the same way they buy anything else: compare price, check the feature list, verify integrations, sign the contract. The ones who got reliable forecasts from AI were asking questions in the RFP that most of us never think to ask: Can I drill into any forecast and see which inputs drove it? Can I change an assumption and watch the output update? Can my VP of Sales understand this report without a data science degree?

The Interpretability Gap Is a Culture Problem, Not a Tech Problem

Here's what strikes me as an AI researcher—and someone who has spent years building these models—when you see this pattern. We've made enormous progress on predictive accuracy. Gradient-boosted trees, transformer-based sequence models, probabilistic forecasting with proper uncertainty quantification—the mathematics is genuinely sophisticated now. The algorithms can detect patterns in customer behavior that no human analyst could track across millions of data points.

But we haven't done nearly as much work on the communication layer between the model's output and the decision-maker who needs to act on it. And that communication layer—what I'd call forecast literacy—is where most marketing organizations are weakest.

This isn't a failure of AI. It's a failure of how we've structured the human-machine workflow. We built powerful prediction engines, handed them to marketing teams trained in brand strategy and campaign planning, and expected trust to follow from accuracy. But trust requires understanding, and understanding requires that the model speaks a language the decision-maker can interrogate.

In machine learning terms: we optimized for $R^2$ when what CMOs actually need is something closer to explainable confidence. Not just "the forecast is 87% accurate historically" but "here's what would change if your email open rate drops 5 points, here's the contribution of your top 10 accounts, and here's where I'm least certain."

What This Means for Your Next Forecasting Decision

If you're a CMO—or work with one—here are three concrete moves that come directly from this research:

Audit your current forecasting workflow. Where does the number come from? If it comes from a spreadsheet built by an analyst, you at least know who to ask questions. If it comes from a "predictive analytics" dashboard, can anyone on your team explain the causal chain behind any specific output? If the answer is no, that's your gap—and it's fixable without buying a new tool.

Invest in feature-level transparency. When evaluating forecasting tools or building internal models, prioritize tools that expose their assumptions. You want to see which variables are driving the forecast and how sensitive the output is to changes in those variables. A model you can't interrogate is a number you have to take on faith—and in revenue planning, faith is an expensive currency.

Build a "forecast review" ritual. The CMOs who trusted AI forecasts all had some version of this: a weekly or bi-weekly session where the forecast is reviewed not just for accuracy but for assumptions. What's the model assuming about seasonality? About new product adoption? About macroeconomic conditions? This isn't an extra meeting. It's the mechanism by which human judgment and machine prediction become complementary rather than competitive.

The Surprising Answer, Revisited

So—do CMOs trust AI to predict revenue?

The surprising answer is: most of them are in a state they'd probably call "professional caution." They use AI forecasts. They don't dismiss them. But they also don't fully rely on them, because the gap between what the model can do and what their role requires hasn't been bridged yet.

And that's not a criticism of AI. It's an observation about where we are in this transition. The technology is further along than the organizational culture around it. Our forecasting tools have gotten smarter faster than our processes for using them intelligently.

The 22% who trust their AI forecasts aren't more optimistic about the technology. They're just better at building the bridge between the model's output and the boardroom question: "Why do you think that's right?"

That bridge is the real work. And it's available to every CMO in this room, regardless of company size or budget. The 8% who fully trust their AI forecasts included a mid-market company running on a $40K/year analytics stack. What they had was a process, not a product.

In revenue forecasting—as in most things that matter—trust isn't given by the tool. It's built by the workflow around it. And that's a problem we can actually solve. 🎯

Dr. Elise Marchand holds a Ph.D. in Artificial Intelligence and researches human-AI collaboration in enterprise decision-making. She advises marketing technology leaders on integrating predictive analytics into business planning workflows.