AI Will Run a Third of Your Marketing Work by 2028 — Most Teams Are Planning for the Wrong Third
Marketing leaders expect AI to automate 36% of marketing work by 2028, up from 16% in 2026. The number isn't the interesting part. Which 36% gets automated — and which 64% becomes the whole job — is the part teams aren't planning for.
A demand gen director showed me her 2026 headcount plan in March. It had a line for a new content automation platform and, two rows down, a backfill she'd decided not to fill — a mid-level campaign manager who'd left in January. Her logic: the platform automates the campaign manager's work, so the role doesn't come back. Clean math.
Then I asked her what the campaign manager actually did all day. About a third of it was the production work the platform would genuinely absorb — building emails, scheduling, variant setup. The rest was deciding which campaigns were worth running, reading the qualitative signal in why one segment converted and another didn't, and the unglamorous coordination that kept six teams pointed the same direction. The platform automated the third. She'd eliminated the role that did the other two-thirds.
This is the shape of the mistake the whole function is about to make. Marketing leaders expect AI-driven automation of marketing work to more than double — from 16% of the work in 2026 to 36% by 2028. That projection is probably right. The problem is what teams are inferring from it: that 36% of the people, or 36% of the roles, become unnecessary. The automation lands on tasks, not jobs, and it lands very unevenly. The teams that win the next two years are the ones who get specific about which tasks.
The Automatable Third and the Stubborn Two-Thirds
Not all marketing work is equally exposed to automation. The 36% is not a random slice — it has a clear shape, and so does the part that resists.
The automatable work is bounded, repetitive, and judgment-light. Producing campaign variants, drafting first-pass copy, building landing pages from a template, scheduling, list segmentation by explicit rules, performance reporting, A/B test setup. These tasks have clear inputs, clear outputs, and a definition of "done" that a machine can check. This is where the 16%-to-36% growth is coming from. It is real, and it is happening fast.
The stubborn work is unbounded, contextual, and judgment-heavy. Deciding what is worth producing at all. Reading why a campaign worked when the numbers alone don't say. Knowing the difference between a segment that didn't convert and a segment that converted later through a channel you didn't instrument. Positioning. Narrative. The taste that separates on-brand from generic. None of this has a checkable "done."
The automatable work is the visible work. This is the trap. The bounded, repetitive tasks are the ones you can see someone doing — they look like the job. The judgment work is quiet and intermittent. So when leaders look at a role and ask "what could a machine do," they see the visible third, conclude the role is mostly automatable, and miss that the value was always in the parts they couldn't watch.
Why the Two-Thirds Gets More Valuable, Not Less
The instinct is that automating a third of the work shrinks the function. The opposite is the more likely outcome, and the reasoning matters.
Automation removes the constraint on volume — which moves the constraint to judgment. When producing a campaign took two weeks, the production capacity was the bottleneck. When it takes two hours, the bottleneck moves upstream: which campaigns, for whom, saying what. A team that can produce ten times more work needs ten times better answers to "what should we produce." The judgment work doesn't shrink. It becomes the entire game.
Cheap production raises the cost of bad strategy. When everything is slow and expensive, a mediocre campaign idea dies in the backlog before it costs much. When production is instant, the mediocre idea ships, runs, and consumes budget and attention. Automation doesn't make bad strategy cheaper — it makes it faster and more expensive. The strategic check on what gets made matters more, not less.
The output of an automated task still needs an owner. An AI drafts the email; someone decides it's right. It builds the landing page; someone owns whether it converts. Automation changes the work from doing to directing and judging — but the headcount that does the directing and judging does not approach zero. It approaches the number of distinct things worth directing, which goes up as production gets cheaper.
Where This Shows Up in Practice
Content teams. The writer who spent four days producing one piece now produces eight in the same time. The skill that matters shifts from "can write" to "knows what's worth writing and can tell good from generic at a glance." Teams that hire and promote on production speed are optimizing the part that just got automated. Teams that hire on editorial judgment are staffing the part that didn't.
Campaign and lifecycle operations. Build, schedule, and QA collapse toward minutes. What remains is orchestration — sequencing, segmentation logic, knowing which lifecycle moment deserves a human-crafted touch and which is fine on autopilot. The ops role doesn't disappear. It moves up a layer, from executing the campaign to designing the system that executes campaigns.
Performance and analytics. AI generates the reports and surfaces the obvious correlations. The remaining work — the hard part — is interpretation: which signal is causal, which is noise, what the qualitative story is behind a number. The analyst who only ran reports is exposed. The analyst who explained what the reports meant is now the most leveraged person in the function.
Brand and product marketing. Almost entirely in the stubborn two-thirds. Positioning, narrative, message architecture, competitive framing — judgment work with no checkable "done." This is where automation arrives slowest and where the relative value of headcount rises fastest as the rest of the function gets cheaper to run.
What to Actually Do About It
Audit your roles by task, not by title. For each role, split the work into the bounded-and-automatable bucket and the judgment-heavy bucket. Put a rough percentage on each. You will find the split is wildly uneven across roles — and that some of your most "junior" roles are quietly two-thirds judgment.
Reinvest the freed capacity instead of banking it. When automation frees a third of a person's time, the default move is to cut. The better move, in most cases, is to redeploy that time into the judgment work the team never had enough hours for — the strategy, the analysis, the experimentation. Cutting captures a cost. Reinvesting compounds an advantage.
Hire and promote for the stubborn two-thirds. Stop screening for production speed and tool fluency — those are the commoditizing skills. Screen for editorial judgment, strategic reasoning, and the ability to tell good from generic. Those are the skills whose relative value rises every quarter as automation spreads.
Build the "what's worth doing" function deliberately. As production gets cheap, the prioritization layer gets overwhelmed — more ideas are now feasible than are wise. Someone needs to own the filter. If no one does, the team will simply produce more of everything, and the volume will bury the signal.
Re-baseline your success metrics. Output volume was a proxy for effort when output was hard. It is now nearly free and nearly meaningless as a measure. Shift the team's scoreboard to outcomes — pipeline influenced, conversion lift, narrative that moved a deal — the things the stubborn two-thirds actually produces.
The Stakes
The teams that read the 36% number as a headcount target will spend 2026 and 2027 cutting roles by their visible third and discovering, late, that they amputated the judgment. Their output volume will rise and their results won't, and they will struggle to explain the gap because the thing they lost was never on a dashboard.
The teams that read the same number as a task-mix shift will use automation to delete the bounded work and pour the recovered hours into the judgment work that was always understaffed. Same automation, same 36%. One team gets a cheaper function that does more undifferentiated work. The other gets a function that finally has time to be good.
The projection — 16% to 36% — is not the decision. The decision is which 36% you let the machines have, and whether you treat the remaining 64% as overhead to trim or as the actual job, finally given room to breathe. Most teams are planning for the wrong third. Plan for the right one.