Eight years ago, I taught a course called Programmatic Buying Foundations. The pitch, right there in the course title, was that data and technology could deliver highly relevant, effective, and measurable ads at scale. Read Kevin Indig’s new Growth Memo column on the hidden hours buried inside AI marketing workflows, and you’ll recognize the dark side of my old pitch. The promise of efficiency was broken then, and broken now, too.
That recognition isn’t a claim that “I’ve seen this movie before.” It’s a warning about where the efficiency argument for AI marketing tools is likely to crack, because it already cracked once, in a discipline I taught for years.
The Promise I Used To Teach
Programmatic Buying Foundations walked marketers through a five-step workflow:
- Organize your audience insights.
- Design compelling creative.
- Execute with integrated technology.
- Reach audiences across screens.
- Measure the impact.
Each step came wrapped in case studies from brands like Mondelez, Campbell’s, and Ford India showing the workflow paying off. The framing was always efficiency first, with the assumption that automating the buying would drag effectiveness and measurability along as a byproduct.
Here’s the part that gets left out when people wax nostalgic about programmatic’s golden era. The same course had a full section on ad fraud and brand safety, and another on GDPR, because both had become live problems for the exact audiences the efficiency pitch was made to. Advertisers were told automation would make their targeting sharper. They also had to learn, in the same curriculum, how to spot fraudulent inventory, how “lack of transparency” ranked as one of the field’s top concerns, and how new privacy rules made view-through and cross-device conversions harder to trust than the dashboards implied. The tool that was supposed to make measurement more reliable came with an entire unit on why measurement was getting less reliable.
See also: Reporting Uncertainty Without Losing Credibility
The Same Crack, Showing Up In AI Marketing
Indig’s argument is that AI hasn’t removed marketing work. It moved it. His numbers back that up with more precision than programmatic ever got. METR put 16 experienced developers to work on 246 real-world tasks, a portion with AI assistance and a portion without. The developers expected AI would speed them up by roughly a quarter. Instead, they finished about a fifth slower, and they still believed afterward that it had sped them up.
Marketing has its own version of that gap. A BetterUp Labs and Stanford survey of over a thousand workers found that AI-generated “workslop,” content that looks finished but isn’t, took an average of nearly 2 hours to fix each time someone received it, a cost that scales past $9 million a year at a large company. Workday’s research quantifies the give-back too, since for every 10 hours AI saves, roughly four end up spent redoing weak output. Upwork, which polled 2,500 leaders and workers, breaks that reclaimed time down further, with much of it going to checking and fixing the AI output, learning the tools themselves, or simply taking on more work than before.
HubSpot’s data shows why this matters at scale. The vast majority of marketing leaders report that their teams already use AI, and a solid majority say their company is building internal AI tools rather than buying them. Indig’s point is that this in-house building doesn’t disappear when the tool ships. It becomes a permanent, mostly invisible maintenance job, and somebody has to own it. When that owner takes a vacation, the workflow reverts to manual until they’re back.
I should have spotted this a decade ago. Programmatic promised that automation would replace manual media buying labor. What it actually did was convert manual buying labor into fraud monitoring labor, brand safety labor, and compliance labor, none of which showed up in the original pitch deck. AI marketing tools are giving the identical spiel. The labor of prompting, building, and maintaining doesn’t vanish. It gets reclassified as something that doesn’t look like “real work” on a project plan, right up until a stakeholder asks why scaled content production hasn’t paid off yet.
My unvarnished take is that this isn’t a story about AI being oversold, exactly. It’s a story about efficiency claims in marketing technology consistently getting measured on the wrong side of the ledger, the hours saved on the visible task, never the hours spent setting up, and babysitting the system that saves them. That was true of programmatic in 2018. It’s true of homebrew AI workflows in 2026. The technology changed. The accounting error didn’t.
See also: Why Digital PR Fundamentals Beat AI Tactics: What Marketing Leaders Should Focus On
What To Do About It
If you manage or work inside a marketing team building its own AI tools, three moves come directly out of both the programmatic playbook and Indig’s data.
1. Put a name and a shutdown date on every internal AI tool. Programmatic’s ad fraud problem got solved partly through ads.txt, an industry standard that made accountability structural rather than optional. Your homebrew AI workflows need the same discipline. Every tool gets an owner and a review date, or it becomes permanent invisible headcount.
2. Track the hours nobody’s counting, not just the hours saved. Ask your team not “did this AI workflow save you time” but “how many hours this month went to building, fixing, or maintaining an AI tool instead of doing the work it was meant to speed up.” The METR study shows people will answer the first question wrong even when they’re being honest.
3. Protect the slow return on investment work on purpose. Content depth, digital PR, and the mentions that actually get cited in AI answers are exactly the work programmatic-era marketers under similar time pressure let slip first, because it doesn’t show results for months. Ring-fence a fixed share of your team’s time for it before AI tooling gets to claim the rest by default.
I taught the efficiency pitch once. I also taught, in the same course, a whole unit on why it didn’t hold up on its own. If your marketing team is chasing AI efficiency without asking where the hours it isn’t counting went, then you’re not looking at a new problem. You’re looking at my old syllabus with a new cover.
More Resources:
Featured Image: HobbitArt/Shutterstock
