Almost every marketing team we talk to is faster than it was two years ago. Campaigns ship in days instead of weeks. Copy, creative, and audiences get generated, tested, and swapped without anyone touching a spreadsheet. On paper, this is the most productive marketing has ever been.
So it is worth asking why so many of the leaders running those teams quietly admit they feel less sure of what's actually working, not more.
That gap between more execution and less understanding is the automation paradox. It's the part of the AI story most of the industry is choosing not to talk about.
Marketing automation has made execution easier and interpretation harder at the same time. AI can now run the tests, write the variants, and allocate the budget, but it has also moved the decisions that used to be human inside platform algorithms that even experienced practitioners can't fully see. The result is a strange kind of progress: performance metrics improve while confidence in what's driving them drops. In an environment where everyone has the same tools and the same "good enough" output, the scarce resources become the two things automation can't manufacture: human judgment and distinctive craft.
For most of marketing's history, the chain from decision to outcome was legible. You chose a channel, set a budget, ran a campaign, and could reason about cause and effect. Automation didn't just speed that chain up. It absorbed the middle of it.
.jpg?width=1020&height=572&name=Blog%20internal%20images%20(2).jpg)
The scale of the shift is real. BCG's 2025 CMO survey found that 71% of CMOs plan to invest more than $10 million a year in generative AI over the next three years. McKinsey reports companies using AI-driven automation seeing 15 to 20% higher campaign ROI, with systems autonomously running tests and deploying winners with less manual oversight. Deloitte found 64% of brands planning to use AI to automate content and improve efficiency.
Every one of those numbers is a good thing on its own. Together they describe a system where more of the work, and more of the judgment, sits behind an interface. A 2026 analysis from the Poznań University of Economics and Business put the human consequence bluntly: in many organizations, marketers "shifted from being decision-makers to validators — or, in less mature contexts, executors — of algorithmic recommendations."
That is the paradox in one sentence. The tools got more capable, and the people got further from the decision.
Here's the part that doesn't show up on a dashboard. The same automation that lifts your numbers can lower your confidence in them, because it widens the distance between a result and an explanation.
The BCG survey captured this precisely. The same CMOs pouring millions into GenAI ranked one use of it dead last: measuring whether their marketing is actually working. Read that twice. The industry is scaling the execution and underfunding the understanding, knowingly.
.jpg?width=1020&height=572&name=Blog%20internal%20images%20(1).jpg)
Aaron Ward, Major Tom's media director, sees the downstream version of this constantly. As he puts it:
"Performance looking okay is just a poor excuse of not knowing how performance is truly impacting the business at the end of the day."
A green dashboard is not the same as a clear answer, and automation is very good at producing the first without the second.
This is the same pattern we've written about as measurement debt: the quiet accumulation of reporting you can't quite stand behind. AI doesn't retire that debt. It raises the interest rate, because there's now a model between you and the number. Closing that gap is doable — it's the kind of attribution rebuild behind our data-led work with Rieker — but the automation won't do it on its own.
The second half of the paradox shows up in the work itself. When everyone can generate three times the content in the same window, the marginal value of another "good enough" asset falls toward zero. Volume stops being a differentiator the moment it's free.
Caleb Maurice, Major Tom's data and analytics lead, names the failure mode without flinching: "the AI tools are really good at creating word salad." His more useful observation is what happens next. "There's now a premium on high quality communication in our industry because of the inundation of AI tools creating AI slop. So really good communicators are now a premium."
That is the counterintuitive move most AI commentary misses. In a market flooded with automated output, the human capabilities that don't scale, like a real point of view, a distinctive brand voice, and creative that's actually brave, get more valuable, not less. The evidence from outside our walls agrees. Analysis of the 2024 IPA Effectiveness Awards found that creative distinctiveness and long-term brand building continue to drive business outcomes even as media gets more data-rich and fragmented. Deloitte's read is the same, pointing marketers toward "distinctive brand experiences" as the thing to protect.
| What AI automation accelerates | What AI automation obscures |
|---|---|
| Producing variants, tests, and content at volume | Explaining what actually drove the result |
| Real-time optimization and budget allocation | Keeping the work distinctive from competitors using the same tools |
| Speed from idea to live | Human ownership of the decision |
| Hitting surface metrics | Trusting that the surface metric means what you think |
There's a subtler risk than slow reporting or generic creative, and it's the one that catches experienced teams. Automated insight can feel like clarity while quietly doing the opposite.
An AI system will happily generate more outputs, more segments, and more "insights" than any human can act on. Caleb's warning is that this abundance can amplify tunnel vision rather than correct it. The volume of analysis creates a feeling of rigor that the thinking underneath hasn't earned. The Poznań researchers reach the same place from theory: without reflective judgment, AI "does not optimise marketing — it automates flawed assumptions and decisions." A wrong prior, executed faster and at scale, is not progress.
Our Innovation Architect / UX Strategist, Olu Osunrinde, describes the on-the-ground version: teams reading confident, automated recommendations about a user journey that is itself broken, then optimizing harder in the wrong direction. The tool sounds sure. That's the trap. Confidence is the one output you should never accept from a model without checking the room it was generated in.
Kantar frames the boundary well: AI is "probabilistic, not deterministic," can sit on "biased, outdated, or structurally flawed" data, and misses "context, incentives, consequences, or emerging cultural meaning." Their conclusion is a line worth keeping on the wall: "human judgment must override the AI model" when optimization conflicts with brand strategy or long-term growth.
None of this is an argument against automation. It's an argument for keeping a human in the seat where judgment actually matters, and being deliberate about which seat that is. A few practical moves separate the teams that stay clear from the ones that drift.
Fund the interpretation, not just the execution. If you're investing in AI to run marketing, invest at least as seriously in your ability to explain what it's doing. The teams under strain are almost always the ones that automated the doing and left the understanding to chance.
Decide, in advance, where the human overrides the model. Brand meaning, cultural timing, and the strategic risk a "bad" score can't see. Then protect those decisions from being quietly handed to the algorithm.
Treat distinctiveness as infrastructure. As Aaron Ward puts it, "a lot of companies really forget that they are advertising to a human at the end of the day." When output is infinite, the point of view is the moat.
Keep asking the question the dashboard won't. Not "are the numbers up?" but "can we explain why, and would we stake the next budget on that explanation?" That's the same question underneath the Clarity Gap, and AI has made it more urgent, not less. As Aaron says, "a tool isn't just going to magically make everything better."
The paradox doesn't resolve. Marketing will keep getting easier to run and harder to read, and both of those trends will accelerate. The teams that win the next few years won't be the ones that automate the most. They'll be the ones that stay clear about what the automation is actually doing, and keep a human judgment premium exactly where the machine runs out of road.
If you can run your marketing faster than ever but can't confidently explain what's driving the results, that's not a tooling problem to solve with another platform. It's a clarity problem. If you're at the point of wanting to fix it, that's where our data and analytics work begins. If you're still weighing it up, the thinking here is yours to take into your next planning meeting.
Often, yes — not because the tools are inaccurate, but because they move decisions inside algorithms that sit between you and the outcome. Execution speeds up while the chain from action to result gets harder to trace. Performance can improve at the same time your confidence in explaining it drops. The fix isn't less automation; it's investing in the measurement and judgment needed to interpret what the automation is doing.
Because a better number and a clear explanation are two different things. BCG's 2025 survey found CMOs scaling GenAI aggressively while ranking GenAI-for-ROI-measurement as their lowest priority. The industry is funding execution over understanding. When you can't say why performance moved, every budget conversation gets harder, regardless of what the dashboard shows.
Both, which is the problem. AI attribution models are more sophisticated than last-click, but they're also more opaque and more dependent on the quality of the data underneath them. Kantar notes AI is "probabilistic, not deterministic" and can sit on flawed data. More precise-looking numbers can be harder for a non-specialist to interrogate, so precision and clarity don't always move together.
AI slop is high-volume, low-distinctiveness content generated because it's cheap to produce, not because it's worth saying. It matters because when everyone can produce three times the output with the same tools, undifferentiated content stops working. The scarce, valuable thing becomes a genuine point of view and distinctive craft, the parts of marketing that don't scale automatically.
Not necessarily. The teams getting this right aren't slowing adoption; they're being deliberate about it — automating execution while deliberately funding the interpretation, measurement, and judgment around it. The risk isn't moving fast. It's moving fast with no one able to explain where you're going.
On brand meaning, cultural timing, long-term strategy, and any moment where the optimal short-term metric conflicts with long-term growth. Kantar's guidance is direct: human judgment must override the model when it misses context a probability score can't hold. Decide those boundaries before you need them, not in the middle of a campaign.
By treating distinctiveness as strategy, not decoration. Evidence from the IPA Effectiveness Awards continues to show creative distinctiveness and long-term brand building driving outcomes. Use AI for volume and iteration, but protect the point of view, the brand voice, and the creative decisions that make the work recognizably yours — those are what a competitor with the same tools can't copy.
Think big, act bold, and let the world catch up.
.jpg)