Only 16% of CMOs feel able to defend their AI spend, and the price of it is still subsidized

Boards have started asking marketing teams to prove what their AI spending buys, and most marketing leaders say they cannot. A column published this week on MarTech by Scott Gillum, founder of the agency Carbon Design, gathers three surveys describing the same hole. Comviva's 2026 Global CMO Survey put the share of marketing leaders who have faced a board-level challenge on AI spending at 86%. The share who said they could meet it with clear business evidence was 16%. Duke University's 2026 CMO Survey rated marketing technology activities on a seven-point scale and found nothing above 5, including generating a return. MIT NANDA's report on enterprise generative AI put the share of organizations seeing no measurable return at 95%, against $30 billion to $40 billion of enterprise investment. MarTech is owned by Semrush and Gillum runs an agency, so read the in-housing argument in that light. The survey figures are not his.

The proof gap has a particular shape in software

Gillum's reading of the MIT number is the part that should land hardest on a software marketing team. Of that enterprise budget, he writes, the biggest slice went to sales and marketing, for the plain reason that it was the easiest place to win internal approval. Operations and finance pilots got less money and less attention, and returned more. Marketing bought the most and can show the least.

That is not a tooling failure. It is what happens when a function adopts a technology on the strength of how obviously it saves time. Time saved is easy to feel and easy to show in a meeting. It is not what anyone promised the board. A content team shipping four times the drafts has not, on that evidence alone, produced four times the pipeline, and nobody at the board table is confused about the difference.

The cost side is not settled either

Every argument about proving return treats the spending figure as a known quantity. It is not. In a post published the same week, the B2B SaaS agency Kalungi argued that today's AI pricing is a land grab rather than a cost structure: infrastructure providers are eating the difference to buy adoption and lock in switching costs, and expect to correct the price once the market has settled. Kalungi sells growth services to software companies and has an obvious interest in clients preferring durable systems to cheap workflows. The mechanism it describes is still the ordinary one for platform subsidies.

Its test question is sharper than most of what is being written about AI budgets: "which parts of your current plan would still make sense if your AI costs tripled?" For a software marketer that question has a second edge. You are not only buying tokens through your tools. Your own company may be selling AI features on a flat-rate plan priced against the same subsidized inference, which means a correction would hit acquisition cost and gross margin in the same quarter.

Agencies are already rewriting the paperwork

The clearest sign that nobody has settled the cost question is in the contracts. Digiday reported this week that agencies are drifting toward something resembling a software business model, and are handling it through addendums and added clauses rather than reopening master service agreements. Four agency executives told Digiday they are not ready for a full overhaul. Changes get triggered narrowly: a client asks for a seat in a proprietary AI tool, wants specific AI language written in, or a new tool enters one project.

What clients want written down is instructive. Per Digiday's reporting, the recurring items are data storage and use, human oversight, indemnity, disclosure of which tools were used, and who owns the intellectual property in AI-generated output. Mile Marker is building metadata and IP terms into existing agreements. Go Fish has added brand safety and data protection clauses over the past six months. Keri Bruce, who heads the advertising group at law firm Reed Smith, told Digiday that "Ultimately, people want disclosure. They want to know what tools are being used."

The market is changing so quickly that at least my advice is to make sure you're building some flexibility into that — Brian Yamada, global chief innovation officer at VML, to Digiday

Nobody quoted there claims to know what AI-assisted work should cost. An industry that has billed for marketing labor for a century has not agreed a price for it.

What a software marketer should change

• Split the efficiency claim from the results claim in board reporting. File hours and output volume under operations, and defend the spend on pipeline, win rate or retention.
• Reprice every AI line item at three times today's rate and see which programs survive. The ones that do not were never programs, they were arbitrage on someone else's subsidy.
• If your product sells AI features on flat-rate plans, get finance to model the same shock. Marketing owns the pricing page even when it does not own the margin.
• Put tool disclosure, IP ownership and human review into agency and freelancer agreements now, as clauses rather than a rewrite. Your buyers' procurement teams will ask before your own legal team does.