The AI Efficiency Illusion: Why Today’s MarTech Workflows Are Just Programmatic History Repeating Itself

Eight years ago, I stood in front of a classroom teaching a specialized curriculum called Programmatic Buying Foundations. The pitch, boldly emblazoned across the syllabus title, was simple yet intoxicating: data and technology could finally deliver highly relevant, effective, and measurable advertising at scale.

If you read Kevin Indig’s recent Growth Memo column on the hidden hours buried inside AI marketing workflows, you will instantly recognize the dark side of my old pitch. The fundamental promise of efficiency—that automation would effortlessly strip away manual labor while multiplying output—was broken then. And, as data continues to mount, it is broken now.

This realization is not merely a cynical nod to the familiar trope that "I’ve seen this movie before." Rather, it is a stark warning about where the current efficiency arguments for generative AI marketing tools are destined to crack. The cracks have already appeared, echoing a familiar fault line in a digital marketing discipline I taught for years.

Main Facts: The Great Accounting Error in Marketing Technology

The core issue plaguing modern marketing teams is a persistent accounting error. When new technologies promise automation, organizations consistently measure the gains on the wrong side of the ledger. They meticulously log the hours saved on visible, manual tasks—such as drafting a quick email or generating dozens of banner ad variations—while completely ignoring the invisible hours spent setting up, prompting, correcting, and babysitting the underlying systems.

The Shift, Not the Elimination, of Labor

When programmatic advertising arrived, it promised to eliminate manual media buying. Instead, it converted manual buying labor into entirely new categories of overhead: ad fraud monitoring labor, brand safety compliance, and privacy regulation navigation.

Today, artificial intelligence is executing the exact same sleight of hand. The labor of writing prompts, building homebrew tools, troubleshooting hallucinations, and editing sub-par output does not vanish. It simply gets reclassified as something that fails to look like "real work" on a project plan—right up until a corporate stakeholder asks why scaled content production has failed to move the revenue needle.

Chronology: A Decade of Broken Efficiency Promises

To understand how the marketing industry arrived at its current obsession with generative AI, it is vital to re-examine the historical trajectory of marketing automation over the past ten years.

2016–2018: The Golden Era of Programmatic Promises

• The Pitch: Programmatic buying was introduced as a seamless five-step workflow designed to optimize media spend instantly. Case studies from global giants like Mondelez, Campbell’s, and Ford India were rolled out to prove the concept.
• The Reality: Within a few short years, the curriculum for these platforms had to expand drastically. Entire units were hastily written to address skyrocketing ad fraud, brand safety crises, and sweeping regulatory frameworks like the European Union’s GDPR. Advertisers were promised sharper targeting, but they instead had to learn how to spot fraudulent inventory and cope with non-transparent dashboards where view-through and cross-device conversions were increasingly difficult to trust.

2023–2024: The Generative AI Boom

• The Pitch: Following the public rollout of foundational large language models, enterprise marketing teams were told that generative AI would write copy, design assets, and code landing pages in seconds, democratizing creativity and slashing headcount costs.
• The Reality: Organizations rushed to build internal AI tools rather than purchasing enterprise-grade solutions. A massive shadow-IT culture emerged within marketing departments, driven by the belief that custom prompts and bespoke wrappers would yield proprietary advantages.

2025–2026: The Reckoning and the "Workslop" Era

• The Present Day: Empirical studies from software development, human resources, and organizational behavior have begun to quantify the true drag of AI integration. Companies are discovering that the time "saved" by AI is frequently clawed back by the painstaking work of review, revision, and internal system maintenance.

Supporting Data: What the Research Actually Says

Skeptics of AI cynicism often demand empirical proof. Fortunately, recent research from top-tier institutions and workplace platforms provides a precise, sobering look at productivity in the age of generative AI.

1. The Developer Speed Paradox (METR Study)

A prominent study by METR put 16 experienced software developers to work on 246 real-world tasks, splitting them between AI-assisted and non-AI-assisted workflows.

• The Expectation: The developers firmly believed that AI assistance would speed up their delivery times by roughly 25%.
• The Result: In reality, the developers finished about 20% slower when using AI. Crucially, even after completing the tasks more slowly, they still walked away convinced that the AI had made them faster. This psychological bias highlights a dangerous disconnect between perceived efficiency and actual output.

2. The Cost of "Workslop" (BetterUp Labs and Stanford Survey)

A comprehensive survey conducted by BetterUp Labs and Stanford University, encompassing over one thousand corporate workers, evaluated the impact of AI-generated content that looks complete at first glance but lacks substance—a phenomenon dubbed "workslop."

• The Time Sink: The data revealed that fixing flawed AI-generated material took an average of nearly two hours every single time an employee received it.
• The Financial Impact: At a large enterprise scale, this hidden correction cycle quietly scales to more than $9 million a year in lost productivity.

3. Workday and Upwork Quantify the Give-Back

• Workday Research: Quantifies the give-back explicitly, noting that for every 10 hours AI theoretically saves a team, roughly four of those hours are immediately consumed redoing weak, generic, or misaligned output.
• Upwork Poll: A survey of 2,500 business leaders and workers broke down how reclaimed time is actually spent. Much of it does not go toward strategic thinking or rest; instead, it is absorbed by verifying AI outputs, learning how to operate shifting toolsets, or absorbing a heavier volume of surface-level work that the technology has enabled.

4. HubSpot Data on In-House AI Building

HubSpot’s state of marketing data indicates that the vast majority of marketing leaders report their teams are already utilizing AI daily. Furthermore, a solid majority note that their companies are actively building internal, custom AI tools rather than deploying standardized third-party software.

As Kevin Indig points out, this in-house building does not magically conclude once a tool is deployed. It morphs into a permanent, largely invisible maintenance job that requires dedicated ownership. When that specific tool owner takes a two-week vacation, the entire workflow grinds to a halt and reverts to manual execution until they return.

Official Responses and Industry Perspectives

As the marketing industry grapples with the fallout of overhyped technological transitions, thought leaders and practitioners are beginning to adjust their messaging.

The Shift in Enterprise Strategy

Chief Marketing Officers (CMOs) who rushed to mandate "AI-first" workflows in 2023 and 2024 are quietly pivoting toward governance frameworks. Rather than celebrating raw output volume—such as the number of blog posts or ad variations generated per hour—forward-thinking enterprises are implementing strict auditing protocols to measure downstream engagement and conversion quality.

The Vendor Counter-Narrative

Software vendors and AI platform creators argue that initial productivity losses are merely part of a standard technological adoption curve. They maintain that as foundational models improve, retrieval-augmented generation (RAG) matures, and prompt engineering becomes more standardized, the friction currently eating into corporate hours will diminish. However, critics counter that even if tools become smarter, the organizational tendency to misallocate saved time into higher-volume busywork remains an enduring human constant.

Implications: Strategic Steps Forward for Marketing Leaders

If you manage a marketing team—or operate inside one that is rapidly building and deploying its own AI-driven workflows—ignoring these lessons will result in burned-out talent, degraded content quality, and unfulfilled business projections.

Drawing directly from both the historical programmatic playbook and contemporary productivity data, marketing leaders must take three decisive actions:

1. Put a Name and a Shutdown Date on Every Internal AI Tool

During the programmatic era, the ad fraud crisis was partially contained through structural standardization like ads.txt, which forced accountability into the supply chain. Your homebrew AI workflows require that exact same level of operational discipline.

• Every custom GPT, internal script, or automated workflow must have a designated owner and a formal review date.
• If a tool cannot prove it is saving more net-positive time than it consumes in maintenance, it must be sunsetted before it quietly hardens into permanent, invisible headcount.

2. Track the Hours Nobody Is Counting

Stop asking your team simplistic, leading questions like, "Did this AI workflow save you time today?" Psychological biases, as demonstrated by the METR study, will lead them to answer affirmatively even when objective metrics say otherwise.

• Instead, pivot your internal retrospectives to ask: "How many hours this month were spent building, troubleshooting, fixing, or maintaining an AI tool instead of executing the core strategy it was meant to accelerate?"
• Measure the shadow labor. Only then will you have an accurate picture of your true return on investment.

3. Purposefully Protect "Slow ROI" Initiatives

Content depth, rigorous digital PR, primary research, and authentic brand mentions—the exact elements that get cited and valued in modern AI-driven search and discovery—share one common trait: they take months to yield results.

• Under intense time pressure driven by efficiency mandates, these slow-burn initiatives are precisely what stressed marketing teams sacrifice first.
• To prevent this, leadership must ring-fence a fixed, non-negotiable percentage of team bandwidth specifically for long-term, human-centric creative and strategic work before AI tooling claims the remaining hours by default.

Conclusion: A New Cover on an Old Syllabus

I taught the programmatic efficiency pitch with absolute conviction a decade ago. But I also taught—in the very same course—a comprehensive unit on why those efficiency promises routinely failed to hold up under real-world pressure.

If your marketing organization is currently chasing AI-driven efficiency without auditing where the uncounted hours are disappearing, you are not wrestling with an unprecedented, futuristic challenge. You are simply reading my old syllabus under a brand-new cover.

It is time to close the ledger, account for the hidden labor, and refocus marketing technology on what it was always supposed to enhance: human ingenuity, not endless busywork.