Key Takeaways
- AI-managed ad spend is on track to exceed a quarter of all U.S. digital spend by the end of the decade, and most brands have no governing framework for it.
- AI optimizes well inside a defined box, but it can’t define the box, and that is exactly where CPG and DTC brands keep losing money.
- The brands winning with automation right now treat AI as a bid and allocation engine.
- Agency models that sell AI autonomy as a feature are often selling less accountability, not more performance.
- Human judgment on audience architecture, creative strategy, and channel sequencing is a competitive advantage.
The uncomfortable truth about AI-managed ad spend isn’t that it works or that it does not; it’s that most brands have already automated significant portions of their media without a clear decision about what should be off-limits. As forecasts from Digiday and others show AI campaigns capturing more than a quarter of U.S. ad dollars by decade’s end, the question is not whether to use these tools. The question is which decisions you’re willing to let a machine make on your brand’s behalf, and which ones you absolutely cannot afford to.
The Machine Is Good at One Thing (and Brands Keep Asking It to Do Everything)
AI bidding and allocation tools are genuinely excellent at pattern recognition inside constrained environments. Feed a well-structured campaign with clear signals, defined conversion events, and consistent creative, and a machine learning system will find efficiencies a human optimizer can’t match at scale.
Where CPG and DTC brands consistently bleed money is when they let AI systems make decisions that require contextual brand judgment. Audience sequencing across a funnel, for example. Or the choice to pull back spend on a channel that is hitting efficiency metrics but eroding brand perception. Or how to weight upper-funnel investment when a new competitor enters the category.
These aren’t optimization problems. They’re strategy problems. No amount of campaign data fixes a strategy that was wrong at the start.
What AI-Managed Ad Spend Actually Looks Like
It helps to get concrete about what “AI-managed” means, because the term covers a wide range of actual control.
At one end, you have AI-assisted buying: a human media strategist sets audience parameters, budget allocations, channel mix, and creative logic, and machine learning handles bid adjustments and pacing in real time. That is a reasonable division of labor.
At the other end, you have fully automated campaigns where the platform, Google’s Performance Max, Meta’s Advantage Plus, or an AI-driven DSP, controls nearly every variable including who sees the ad, when, at what frequency, and on what placements. The human sets a budget and a goal. The machine decides everything else.
The second model isn’t inherently wrong, but it requires something most brands skip: a pre-commitment to what success looks like beyond ROAS, a clear audience exclusion strategy, and someone accountable for reviewing what the machine actually did.
Without that, you aren’t running AI-managed campaigns. You’re running unmanaged ones.
How to Decide What to Automate
Here is how we think about it at Junction 37. Before any campaign element gets handed to an AI system, it has to pass three questions.
- Is the optimization objective measurable and attributable? If yes, AI can help. If the outcome you care about is brand equity, category perception, or consideration lift, the machine will optimize for a proxy metric that may not reflect reality.
- Is the creative strategy already locked? AI will distribute and test creative variations efficiently, but the message, the audience, and the brand position is a decision that comes from humans.
- Do you have enough first-party signal to trust the model? AI systems trained on thin data make confident-sounding mistakes. CPG brands with limited direct consumer relationships need to be especially careful here.
If all three answers are yes, automation is a genuine asset. If any answer is no, putting AI in the driver’s seat is a risk management problem, not a media efficiency gain.
The Agency Model Problem No One Is Talking About
There is a structural incentive issue worth naming. For large holding company agencies, AI automation is a margin play. Fewer junior staff hours, same management fee, better-looking efficiency numbers on a dashboard.
The result is campaigns that look clean in reporting and underperform in the market. We have inherited enough of these accounts to say this plainly.
Automation should reduce cost and complexity for the client. When it primarily reduces cost for the agency, something is broken. If you want to see what a different model looks like, take a look at how we work with CPG and DTC brands and why accountability is built into every engagement.
Human Expertise Is Not the Opposite of AI Efficiency
We aren’t making an argument against these tools. We use them. We built our own thinking around them, and if you want to go deeper on where the line sits, our post on the right media buying split for CPG brands lays it out specifically.
The argument is simpler than it might sound. A quarter of U.S. ad spend flowing through AI systems is a milestone that should prompt brands to get explicit about their own framework, not just follow the forecast.
The brands that will win this decade won’t be the ones who automate the most. They’ll be the ones who are clearest about what they are automating and why.
That clarity is a human job. Protect it.
—
Frequently Asked Questions
What does AI-managed ad spend actually mean for CPG brands?
It means a growing share of your media budget is being allocated, bid on, and optimized by machine learning systems with limited human input at the execution layer. For CPG brands, this creates efficiency gains in constrained environments but real risk when those systems are making decisions that require brand or category context.
Which campaign decisions should CPG and DTC brands never fully automate?
Audience architecture, channel sequencing, creative strategy, and any decision that connects media to brand positioning should remain human-led. AI can execute against a defined strategy with precision, but it cannot construct the strategy in the first place.
How do I know if my agency is using AI automation in my interest or theirs?
Ask for transparency on who is actively reviewing campaign decisions versus what the system is making automatically. If the answer is vague or if you aren’t getting regular human-authored strategic rationale alongside your performance reports, that is a signal worth taking seriously.
What first-party data do I need before trusting AI to manage my campaigns?
At minimum, you need clean conversion data tied to real purchase or intent signals, audience suppression lists to avoid wasted spend, and enough volume for the model to learn from. Thin data environments produce confident-sounding AI decisions that are often wrong. Build the data foundation before expanding automation.
—
Ready to build a media strategy where humans and AI both do what they are actually good at? Talk to the Junction 37 team.
—
Chris Pyne, Founder and CEO of Junction 37 and its sister venture Series A. He built Cortex, J37’s proprietary AI-driven planning ecosystem, and pioneered the integration of predictive marketing science into client strategy. Previously, Chris held C-suite roles at OMD USA and MediaCom, where he led planning for $7B in billings and 700+ employees.