This article features a Q&A with Jay Combs, VP of Marketing at ModelOps
Ask a marketing leader how many AI initiatives their organization is running and you’ll usually get a number. Four, maybe nine. The copywriting assistant, the personalization model somebody built on top of the CRM, whatever the agency is piloting this quarter. The number feels solid because it corresponds to decisions people remember making.
Jay Combs thinks that number is off by an order of magnitude, and that the gap isn’t made of decisions at all. Combs is VP of Marketing at ModelOp, which sells enterprise AI delivery and governance software — worth stating up front, because nearly everything he argues here describes a problem his company charges to solve. His evidence is ModelOp’s 2026 AI Governance Benchmark Report, a survey of 100 senior AI leaders that gave the condition a name: an “AI value illusion,” where portfolios expand and delivery timelines compress while the ability to say what any of it is worth falls quietly behind.
Marketing Already Lost This Argument Once
We know the shape of this. Marketing spent decades reporting whatever was easiest to produce, and it took years of CFO pressure to drag the conversation toward revenue and retention. Combs sees the same sequence running again, faster.
“For years, marketing measured impressions, clicks, and MQLs because they were easy to count. Eventually the focus moved to what actually mattered: revenue, retention, growth. AI is going through the same transition, but with far more at stake.”
What’s different is the cost of being wrong. Over-counting MQLs wasted attention, mostly. An AI initiative with no outcome attached burns engineering hours, vendor fees, and — increasingly — a metered bill that arrives monthly whether anyone reads it. Combs’s prescription is unglamorous. Tie every initiative to a specific business use case and the outcome it’s meant to improve, and do it “up front, not after the fact.” He’d go further: any initiative that can’t name its use case should raise questions with executives, because those initiatives “can quickly become very expensive.” For a CMO, that’s less a compliance step than a budgeting habit, and it produces a defensible answer when finance asks what the AI line bought.
The AI You Didn’t Buy
Combs splits marketing AI into two categories. Only one of them went through procurement.
“There’s the AI your team reaches for directly including frontier models like Claude or ChatGPT for design, copywriting, competitive research, and building decks. And there’s the AI already embedded in the platforms you run every day: your CRM, content, advertising, and analytics tools, which can get enabled through a routine update, and often at a premium or metered token cost.”
The second category is the one that should keep marketing leaders up at night, because nobody evaluated it. It showed up in a release note. Multiply that across the ten or twelve platforms in an ordinary enterprise stack, then across business units and geographies, and Combs’s arithmetic starts to look reasonable: “a marketing org can have dozens, even hundreds, of AI initiatives running with no single way to see all of them.” His first move isn’t a policy document. It’s a question a CMO ought to be able to answer without commissioning a two-week audit — what AI is influencing our content, personalization, segmentation, and campaigns today, and what business use case is each one tied to? Answer that and you can be proportionate. Most marketing AI is low risk and should move fast. The few use cases touching pricing, transactions, or support are the ones that earn real review.
Governance Is a Sequencing Problem
Every marketing leader has the objection ready: governance means legal review, and legal review means the campaign ships in November instead of September. Combs doesn’t argue with the caricature. He says most programs earned it.
“When governance is bolted on at the end and fragmented across teams, it slows everything down. The work finishes, and only then do missing documents, new approvals, and open compliance questions surface. The problem isn’t governance, it’s the process.”
His fix is to run policies, testing, and documentation continuously across the AI development lifecycle instead of staging them at the finish line, with the depth of review matched to what’s actually at risk. Drafting campaign copy takes the fast path. AI that touches a price takes the slow one. It’s fair to name the self-interest here, because a continuous enterprise-wide governance standard is exactly what ModelOp’s platform exists to operate, and “embed it into the full AI lifecycle” reads as much like a product description as a recommendation. The underlying claim is still testable inside your own organization, though, and checking it costs nothing. Look at where AI review time actually goes this quarter. If most of it is spent reconstructing decisions after the work is finished, the delay isn’t coming from governance. It’s coming from late governance. Combs thinks the real shift is cultural — that CIOs, CMOs, and other business leaders have to stop treating oversight as a gate and start treating it as part of how AI gets delivered.
Ask Where the Tokens Went
Marketing tends to hear “AI Factory” and assume it’s built for credit models and clinical trials. Combs argues the opposite. Marketing adopts AI faster than almost any other function, new capabilities land every week, and once you scale that across regions the case for a consistent operating model gets hard to dismiss. Then he adds the part that’s genuinely new this year.
“As GenAI platforms meter by the token, every campaign, draft, and experiment carries a cost. Leaders need to see where that budget goes and ask a blunt question: are we funding real priorities, or just generating expensive AI slop?”
Token-based pricing rewires marketing’s cost structure in a way that hasn’t fully landed yet. Content production used to be a fixed cost — salaries, retainers, a per-project fee you negotiated once. Metered generative AI makes a chunk of it variable, consumption-driven, and nearly invisible, distributed across dozens of tools that each bill separately. “AI slop” is a pointed phrase for a vendor CEO to use about his own category, and it lands because most marketing organizations can’t currently distinguish between a thousand dollars of useful generation and a thousand dollars of filler. Combs’ three priorities for the year follow from that. Build an AI system of record covering internally developed models, embedded vendor AI, generative applications, and autonomous agents. Define the business outcome before deployment rather than after. Then measure outcomes instead of activity — “don’t celebrate prompts written or content generated,” he says, when what you can measure is revenue, conversion, retention, or operating efficiency.
The system-of-record item is, again, his product category. But he attaches a benefit to it that has nothing to do with oversight: a visible inventory lets marketers find what already exists “before reinventing the wheel, duplicating costs and efforts.” Anyone who has watched two teams in the same department separately commission the same subject-line model will recognize the value of that.
Combs’ closing point is the one most likely to get skipped, and it’s the one with the most organizational teeth. He doesn’t think AI value belongs to any single executive — not the CIO, not the CMO, not the CFO. The organizations he sees doing this well treat AI as a shared capability, with technology, marketing, finance, legal, and risk all working from one operating model. Which is convenient for a company selling an enterprise platform, and also the only arrangement that survives contact with a stack where marketing’s tools and IT’s tools are increasingly the same tools.
So start with the count. Not the aspirational count from the AI strategy deck, but the real one: pull the release notes from every platform your team logged into last month, find the features that arrived switched on, and find out what each one is billing you. The number that comes back is your actual starting position. It’s probably the most useful thing you’ll learn about your AI program this quarter, and nobody else in the building is going to go get it for you.
