RoBoLeave a Comment on Nobody Was Lying

Nobody Was Lying

What the research actually says about the changing CMO and CRO role, revenue operations, attribution, and AI

I have sat in the QBR and watched four leaders present four different numbers for the same quarter. Marketing had one. Sales had another. Finance had a third that was technically correct and completely useless. Customer success had a fourth that nobody bothered to ask about.

Heck, I might have even been one of those leaders.

It is easy to tell this story as though I were the one in the room noticing the problem. Some days I was. Other days I was the person defending a number I had not personally traced back to its source, because the dashboard said what I needed it to say and I had nine other things to get through before lunch.

We have all been there. Making sense of fragmented data in our own way. A spreadsheet held together by three people who know which tab is current. A CRM dashboard that does not track the number quite the way you need it, so you export it and fix it by hand. Finance is handing over the figures behind CAC and CPL with their own definitions attached. Customer success defending CLV and churn. Sales showing closed-won that never makes it through implementation.

Then the fingers start pointing. The board is unhappy. And one person is working very hard to control the conversation and the story.

Nobody in that room is lying. Everybody is pulling from a different system.

That is the reason I care about this more than almost anything else in commercial leadership right now. Because now, those same four teams are running AI on top of those same four numbers.

What is actually happening to the CMO, CRO, and chief commercial officer roles?

The short version is that the job got bigger and the ground under it got softer.

We hold commercial leaders accountable for revenue. Then we hand them a CRM that sales owns, billing that finance owns, product data that engineering owns, and a marketing platform with its own private definition of a lead. Four owners, four definitions, one person on the hook.

The pressure shows up in the tenure data. Spencer Stuart’s 2026 CMO tenure analysis looked at 346 named CMOs across the S&P 500 and found average tenure at 4.1 years against 5.0 years for the C-suite overall. In healthcare, my home turf, it drops to 3.9. About a third of those companies have no enterprise marketing leader at all.

I have watched people read that stat as a death notice for the CMO. It is not. Look at where those leaders go.

Are the CMO and CRO roles really converging?

Yes. This is the one claim in this piece that the research backs without much argument.

Spencer Stuart’s Fortune 500 study found that 11% of top marketing leaders have no version of the word marketing in their title. The words showing up instead are commercial, growth, customer, brand, and strategy.

Their 2026 analysis puts industry names to it. Hospitality companies are naming chief commercial officers who own sales and marketing. Software companies are designating chief revenue officers. Retail is naming chief customer officers who take on omnichannel activation, retail design, and the in-store experience, in addition to the traditional brand and insights remit.

And the short tenure is not what it looks like. In that same 2026 data, of 218 CMO exits between 2021 and 2025, 62% were either promoted internally or moved to a similar or bigger role elsewhere. Nine percent became CEOs. Another 13% became divisional CEOs, presidents, or COOs.

The job is not disappearing. It is absorbing sales. Which means if you cannot read a pipeline the way a CRO reads a pipeline, you are not being considered for the version of your job that exists in 2026.

Do leaders who understand attribution actually outperform?

Partly. And here is where I have to say something that works against my own resume.

Deloitte Digital surveyed 650 B2B sales executives across 13 industries and found that organizations with a firmly established RevOps model were 1.4 times as likely to exceed revenue goals by 10% or more. Real research, real sample, fielded by Lawless Research at US companies with at least 500 employees and $500 million in revenue.

But sit with what that finding can and cannot tell you. It is correlational and self-reported. Companies that already run well are the companies with the discipline and the budget to build RevOps in the first place. So the result is equally consistent with RevOps producing performance and with performance producing RevOps. Probably both are true and they feed each other. What the study cannot establish is that installing a RevOps function will move your number, which is exactly the claim it gets used to support in board decks.

Now the uncomfortable part.

Brett Gordon, Robert Moakler, and Florian Zettelmeyer analyzed 663 large-scale randomized experiments at Facebook and asked a clean question. If you already know the true causal effect of an ad because a proper controlled experiment measured it, can sophisticated modeling of observational data recover that same answer? They had access to more than 5,000 user-level features, which the authors note is richer than what most advertisers or their measurement partners can access.

The models did not get close. Against true median lifts of 29%, 18%, and 5% for upper-, middle-, and lower-funnel outcomes, the double machine learning estimated 83%, 58%, and 24%, respectively. Stratified propensity score matching did worse still, at 173%, 176%, and 64%. The authors conclude that despite large-scale experiments and rich user-level data, they were unable to reliably estimate a campaign’s causal effect. The study is Close Enough? A Large-Scale Exploration of Non-Experimental Approaches to Advertising Measurement, published in Marketing Science in 2023.

Read those numbers again. At the bottom of the funnel, the modeled estimate was nearly five times the real effect.

The reason matters more than the number because the reason tells you whether your own situation is any different.

The authors are explicit about the mechanism. Ad platforms use complex, evolving processes to decide which users see which ads, so any non-experimental method must undo that selection before it can isolate the campaign effect. The platform is not showing your ad to a random slice of the market. It is showing your ad to the people its own system has judged most likely to respond, which means a share of them would likely have converted without you. So the data you look at afterward has two things tangled together, the effect of your campaign and the effect of the pre-selection. The paper’s conclusion is that the untangling does not work with the data platforms currently log, and that it may not work until platforms record the auction-specific features behind their targeting decisions.

The research studied ad platforms. It did not study B2B funnels, and I am not going to pretend otherwise. But the mechanism it describes is not unique to Meta, and that is the part worth thinking hard about.

Your SDRs prioritize the accounts that look most likely to close. Your ABM program targets companies already showing intent. Your nurture logic routes the most engaged contacts to sales fastest. Every one of those is a selection mechanism you built on purpose, and every one of them means the touchpoints receiving credit are systematically attached to the deals most likely to have closed anyway. You built a system to find people who are ready to buy, then measured the system by whether the people it found bought.

I have stood in front of a board and said marketing sourced 83% of closed revenue. That number was accurate as a description of what my tracking system recorded. It was not proof that marketing accounted for 83% of that revenue, and I was not careful enough about the distinction for a long time. The budget moved on that number. Headcount moved on that number.

So what does attribution fluency actually look like if it is not the ability to run the model?

It is knowing which of your numbers are descriptive and which are causal, and never letting the two get spoken about in the same tone of voice. Sourced pipeline is descriptive. It tells you what your system recorded and where. Incremental revenue is causal, and the Gordon study is a useful reminder of how you get at it, because the researchers treated randomized experiments as the ground truth against which everything else was judged. In practice, that means holding something back and comparing. Geographic holdouts, staggered launches, a segment you deliberately do not touch for a quarter. Those tests are uncomfortable to propose because they look like leaving money on the table, and they are the closest thing you have to finding out whether the money was ever there.

The leaders I would bet on are not the ones with the most sophisticated model. They are the ones who can say to a CFO, this number is what we observed, this other number is what we tested, here is the gap between them, and here is what I would need to close it.

I cannot prove that sentence earns more trust than a dashboard. I can tell you that every time I have said some version of it, the conversation got easier, and the follow-up questions got better.

Does AI make data problems worse?

Yes, and not for the reason most vendors tell you.

Salesforce found that 84% of data and analytics leaders say their data strategy needs a complete overhaul before AI can succeed, and 49% say their companies sometimes or frequently draw wrong conclusions from data missing business context. Fair finding. Also, a company is selling the remedy, so hold it loosely.

MIT complicates it. Their Project NANDA team reviewed 300 deployments, conducted dozens of executive interviews, and found that roughly 95% of GenAI pilots delivered no measurable P&L impact. But they did not blame data quality first. They blamed a learning gap, tools that never adapt to the actual workflow, and integration that stops at the demo. The report is The GenAI Divide: State of AI in Business 2025.

So, clean data is necessary but not sufficient. You can have one source of truth and still burn a year on a pilot that never touched a real process.

Buying tools without correct data will just make you wrong faster.

Does it matter where RevOps reports?

I have been in this argument more times than I can count. Put RevOps under the CRO or it will never have teeth. Put it under finance or the numbers will always flatter sales. Everyone in the room is certain, and the certainty is the tell.

I went looking for what sits underneath it and came up empty. Every source I could find arguing for a particular reporting line was a vendor or a consultancy stating a preference. I did not find research measuring outcomes by reporting line. If it exists, I want it.

Why is the hole there? I have a guess, not an answer. Comparing companies by reporting line means comparing companies that already differ in size, leadership bench, and whichever executive had capital the quarter it got decided, and those same differences plausibly drive performance. Separating the structure from everything bundled with it is the same causal problem the advertising research ran into. I do not know whether it is solvable. I know I could not find anyone who had solved it.

The vacuum is also how a single consulting article became the evidence base for a function.

BCG’s 2020 article on go-to-market operations is where several of the most-quoted RevOps figures come from. The 100% to 200% increase in digital marketing ROI. The 10% to 20% lift in sales productivity. The 30% reduction in go-to-market expense. Read how BCG frames them. They introduce the list by saying top B2B technology companies are reporting these benefits. That is a consultant describing what clients told them, offered plainly as that, with no control group and no claim of causal proof.

Then watch a number like that travel. I found those same figures recirculating across vendor blogs and agency posts, and by the time they land the hedges are gone, “clients are reporting” has become “research shows,” and a field observation is doing duty as a benchmark. I have put that number in a deck. If you work in this space, you probably have too.

That is the same error the attribution section is about, running on a different track. Something descriptive gets promoted to something causal because the promotion is useful and nobody has an incentive to check.

The part almost nobody quotes is that BCG argued against the org chart obsession in the same article. Centralization, they said, suits many companies but is not a universal answer, and reorganizing by itself accomplishes little because results come from processes and tooling rather than boxes. The source people reach for to justify moving boxes said the boxes are a small part.

Which brings me to what I think the reporting line is really standing in for. This next part is my position, not a finding.

When people argue about where RevOps sits, I think they are arguing about who gets overruled. If RevOps reports to the CRO but sales can still override the definition of a qualified opportunity, nothing changed. If RevOps reports to finance but owns the field definitions, the routing logic, and the forecast methodology across all three functions, it will work from what everyone calls the wrong place. Authority is the variable. The org chart is one way to grant it, but not a dependable one.

So the question I would ask instead. When marketing and sales disagree about what counts as a qualified lead, who decides, and does that decision survive the next QBR?

Here is what that failure looks like in the room. A CRO googling the textbook definition of an MQL, or an SAL, or an SQL, then walking into a meeting to use it against their own CMO and their own RevOps team.

Think about what has to be true for that to happen. The definitions living in their own systems are contested enough that nobody trusts them. There is no internal answer to point to, so the search goes outside, because a definition off the internet feels like neutral ground. And they are now arguing against the two functions that should have built that definition with them in the first place.

No org chart fixes that. A lead stage definition is not a fact you can look up. It is an agreement between the people who generate demand and the people who close it, and it only means anything if both sides wrote it and both sides are held to it.

If you cannot name the person who decides, your reporting line is decoration. If that person has to google the answer, you never had an agreement. You had two teams working from two dictionaries, and a QBR where they find out.

What about teams that split into AI adopters and AI holdouts?

Every leader I talk to has some version of this. One group runs at it, another group will not touch it, and the second group gets called resistant.

I think that label is usually wrong, and the Gallup data points at why.

Gallup’s Q4 2025 workplace survey, nationally representative, found that 69% of leaders use AI at work at least a few times a year, against 55% of managers and 40% of individual contributors. The people setting AI strategy are its heaviest users. The people who have to run it inside a workflow are its lightest.

That is not a motivation gap. It is a gap in what the tool is being asked to do.

An executive uses AI to summarize, to draft, to reframe a deck, to think through a memo before writing it. That work has no system of record, no audit trail, and no consequence when the output is wrong, because the executive reads it before it goes anywhere. In that context the tool is genuinely excellent. It feels like the future.

Now take a rep updating an opportunity, a CSM logging renewal risk, an ops analyst reconciling two systems that disagree. That work has a right answer, a downstream consumer, and a real cost when it is wrong. Same tool, entirely different job, and a much harder one.

So the rollout stalls and leadership reads resistance. My read is that the two groups are not evaluating the same thing, and the group closer to the work is closer to being right about it.

MIT points the same direction. Their GenAI Divide report found that over 40% of knowledge workers use AI tools personally, and that those same users describe the tools as unreliable once they meet them inside enterprise systems. That is not a workforce refusing technology. That is a workforce that has used the good version at home and can tell the difference.

There is a governance problem underneath all of it. In that same Gallup data, 38% of employees said their organization had integrated AI, 41% said it had not, and 21% did not know. Your people cannot agree on whether the company has an AI strategy at all. That is the four numbers in the QBR again, wearing different clothes.

The honest limit. I could not find credible independent research connecting the adoption split to revenue outcomes. The studies claiming AI power users are several times more productive are published by companies selling AI platforms, and I am not putting a number in front of you that I would not defend under questioning. Treat the mechanism above as my reasoning, not a finding. Same answer on data gatekeeping between marketing, sales, and customer success. Widely asserted, and I could not find it properly studied.

What I would do anyway. Before you label anyone a holdout, sit with one of them while they try to use the tool on their real work, in their real system, with their real data. If it fails there, you do not have an adoption problem. You have the problem this whole article is about, wearing a different name.

What should a CMO or CRO do about fragmented revenue data?

Four moves, in order. None of them need budget.

Start by tracing one number. Pick the one your board sees every month and follow it back to the system it came from, not through the dashboard and not through the analyst who built the deck. You are looking for two answers. Where it originates, and who can change its definition without telling you. If you cannot name that second person, you have found your first problem. If you want the fastest version, trace closed won that never makes it through implementation. Sales, finance, and customer success can each report that number honestly and still land in three different places, because nobody owns the space between the signature and the go live.

Next, get the definitions written down by the people who have to live with them. A lead stage definition is not a fact you look up. It is a treaty between the team that generates demand and the team that closes it. Marketing and sales draft it together, in one document, with one named person who breaks ties. Then it gets reviewed on a schedule, because definitions drift the moment someone adds a field.

Third, separate what you recorded from what you proved. Sourced pipeline is descriptive. It tells you what your system captured. Incremental revenue is causal, and the way to get near it is to hold something back and compare. One geographic holdout. One staggered launch. One segment you leave alone for a quarter. Run a single test rather than rebuilding your whole measurement stack, and report the result next to the attributed number instead of in place of it. Saying which is which also protects you. When a number turns out to be soft and you never said which ones might be, the room stops trusting all of them.

Last, before you call anyone an AI holdout, sit beside them while they use the tool on their real work, in their real system, with their real data. If it fails there, you do not have an adoption problem. You have this problem wearing a different name.

Now go back to that QBR. Four leaders, four numbers, nobody lying, and one person working very hard to control the story.

I have been that person. It works right up until someone traces the number, and eventually someone does.

If you have sat in that meeting, I want to hear about it. Which number was it, and who owned the definition?

Find me on LinkedIn or at robo@robinbowling.com.

Sources

Leave a comment

search previous next tag category expand menu location phone mail time cart zoom edit close