AI Media Buying Is Changing Who Makes the Decisions

AI media buying refers to the use of machine learning systems to automate and optimise the planning, purchasing, and placement of advertising across digital channels. These systems ingest performance data, adjust bids in real time, and allocate budget across placements without waiting for a human to pull a lever. The question worth asking is not whether AI can do this. It clearly can. The question is what that means for the people who used to do it manually, and whether the outcomes are actually better.

The short answer: AI media buying delivers genuine efficiency gains at scale, but it optimises for what it can measure, which is not always what matters most to a business. Understanding that distinction separates marketers who use these tools well from those who hand over the wheel and hope for the best.

What AI Media Buying Actually Does

Strip away the vendor language and AI media buying is doing a few specific things. It is adjusting bids in real time based on signals like device, time of day, audience segment, and historical conversion rates. It is allocating budget across placements and ad sets based on which combinations are performing. It is testing creative variations and shifting spend toward the ones generating better results. And it is doing all of this faster and at greater scale than any human team could manage manually.

Google’s Performance Max, Meta’s Advantage+ campaigns, and programmatic platforms like The Trade Desk all operate on versions of this logic. You define an objective, set a budget, feed in creative assets and audience signals, and the system takes it from there. The pitch is compelling: less manual work, more efficient spend, better results.

In some cases, that pitch holds up. When I was at iProspect managing large-scale paid search accounts, the shift from manual bidding to automated bidding strategies was genuinely meaningful. Accounts we had spent months manually optimising were being matched or exceeded by smart bidding within weeks, particularly on high-volume campaigns where the machine had enough conversion data to learn from. The efficiency was real. But so were the failure modes, and those were less visible.

If you want a broader grounding in how paid media channels work before going deep on automation, the paid advertising hub covers the fundamentals across search, social, and programmatic.

The Efficiency Gain Is Real. The Risk Is Misunderstood.

The efficiency argument for AI media buying is straightforward. A human optimising a paid search account is checking in periodically, making decisions based on a snapshot of data, and limited by cognitive bandwidth. An automated system is making thousands of micro-decisions per day, across every auction, informed by a much larger data set. At scale, that is a structural advantage.

But efficiency and effectiveness are different things. A system can be very efficient at doing the wrong thing. I have seen this play out more than once. A Performance Max campaign running with a poorly defined conversion event, something like a page view or a time-on-site threshold, can spend aggressively and show a strong in-platform ROAS while delivering almost no actual revenue. The machine is not lying. It is optimising for exactly what you told it to optimise for. The problem was upstream, in the objective-setting, not in the algorithm.

This is the part of the AI media buying conversation that vendors tend to gloss over. The system is only as good as the signal you give it. Attribution and campaign analytics are already a fraught area in paid media, and when you introduce an automated system that learns from your attribution data, any flaws in that data get amplified rather than corrected.

What Happens to the Human Skill Set

When I grew the team at iProspect from around 20 people to over 100, a significant part of what we were hiring for was tactical execution capability: people who could build and manage paid search accounts, set bids, structure campaigns, and optimise based on data. That skill set took time to develop and was genuinely valuable.

Automation has compressed that. A junior practitioner today can manage a Performance Max campaign with a fraction of the technical knowledge that would have been required five years ago. That is not a bad thing for clients. But it does change what agencies and in-house teams need to be good at.

The skills that matter now are upstream of the platform. Knowing how to define the right objective. Understanding what a conversion event actually represents in business terms. Being able to assess whether the machine’s reported performance reflects real commercial outcomes. Knowing when to override the algorithm rather than defer to it. These are strategic and analytical skills, not platform skills, and they are harder to develop than knowing where to find the bid adjustment settings.

There is also a real risk that as automation handles more of the execution, organisations lose the institutional knowledge that comes from doing things manually. When something breaks or behaves unexpectedly, you want people who understand the mechanics well enough to diagnose it. If your entire team has only ever worked with black-box automated systems, that diagnostic capability is thin.

Where AI Media Buying Works Well

There are conditions under which AI media buying genuinely outperforms manual approaches, and it is worth being specific about what those conditions are.

High volume with clean conversion data is where automated systems have the clearest advantage. If you are running an e-commerce account with thousands of transactions per month and a well-defined purchase event as your conversion goal, smart bidding strategies have enough signal to learn from and enough room to optimise. The machine can identify patterns across dimensions that no human analyst would have the bandwidth to track manually.

Early in my career, running a paid search campaign for a music festival at lastminute.com, the speed at which a well-structured campaign could generate revenue was striking. Six figures in revenue within roughly 24 hours from a campaign that was, by today’s standards, quite simple. The principle has not changed: when the targeting is right and the conversion path is clean, paid search moves fast. Automated bidding accelerates that further when the data supports it.

Retargeting is another area where automation adds genuine value. The logic of retargeting campaigns is well-suited to machine optimisation: you have a defined audience of people who have already shown intent, and the system can identify which segments within that audience are most likely to convert and at what bid level. The decisions are data-rich and repetitive, which is exactly where automation earns its keep.

Broad audience prospecting with a large creative library also benefits from AI optimisation. Advantage+ campaigns on Meta, for instance, can test combinations of creative and audience at a scale that manual A/B testing cannot match. If you have invested in creative production, automation can help you find out what works faster than a structured test would allow.

Where AI Media Buying Creates Problems

The failure modes are less frequently discussed, partly because they are less visible in platform reporting and partly because vendors have no incentive to highlight them.

Low-volume accounts are a persistent problem. Automated bidding systems need conversion data to learn from. If you are running a B2B campaign generating 10 to 15 conversions per month, the algorithm does not have enough signal to optimise meaningfully. It will still spend your budget, it will still report on performance, but the optimisation is largely illusory. You are better off with a simpler manual or enhanced CPC approach and focusing your energy on the offer and the targeting rather than the bidding strategy.

Brand safety and context are also areas where automation falls short. Programmatic buying, in particular, has a well-documented history of placing ads in environments that are actively harmful to brand reputation. The efficiency of buying at scale across thousands of placements comes with the cost of reduced control over where your ads appear. For some brands that trade is acceptable. For others it is not, and the in-platform brand safety controls are blunt instruments.

Over-reliance on platform-reported metrics is perhaps the most widespread problem. Platform-level data is not neutral. Google and Meta are both reporting on performance using their own attribution models, which have a structural tendency to credit their own platforms generously. When an AI system is optimising based on those signals, it is optimising toward a version of reality that may look quite different from what your own analytics show. The gap between platform-reported ROAS and actual business outcomes is one of the most consistent problems I saw across client accounts throughout my agency career.

Clients who handed the keys to automated systems without understanding this distinction often came back to us six months later confused about why revenue had not grown in proportion to their ad spend. The machine had been doing its job. The job it had been given was just not quite the right one.

The Innovation Problem in AI Media Buying

There is a version of the AI media buying conversation that is mostly theatre. Vendors presenting at conferences, agencies positioning themselves as AI-first, platforms rolling out features with names that suggest a level of intelligence that the underlying mechanics do not quite support. I have judged the Effie Awards and seen behind the curtain of what actually drives marketing effectiveness. It is rarely the technology. It is the quality of the strategic thinking that precedes the technology.

The question I always come back to is: what problem is this solving? Not in the abstract, but for a specific business with a specific challenge. Automated bidding solves the problem of making real-time auction decisions faster than a human can. That is a genuine problem worth solving. But “we are using AI for our media buying” as a positioning statement, without being able to articulate what it is actually doing differently or better, is not a strategy. It is a talking point.

Clients sometimes come in asking for AI-driven media buying because they have read about it or heard a competitor is doing it. That is a reasonable starting point for a conversation, but it is not a brief. The useful conversation starts with what they are trying to achieve commercially, what their current performance looks like, and where the specific gaps are. Sometimes AI-driven automation is the right answer to that. Sometimes it is not. The technology should follow the diagnosis, not precede it.

How to Work With AI Media Buying Systems Effectively

If you are running paid media at any meaningful scale, you are probably already using some form of automated bidding whether you think of it that way or not. The question is how to use it well.

Start with the conversion event. This sounds obvious, but it is where most problems originate. Your conversion event should represent something that is genuinely valuable to the business, not a proxy that is easy to measure. A lead form submission is a proxy. A qualified sales conversation is closer to the real thing. A closed deal is the real thing. The further your conversion event is from actual business value, the more the machine will optimise toward something that looks good in the platform but does not move the needle commercially. Tools like the Google Ads traffic estimator can help you sense-check volume expectations before you commit budget to a campaign structure.

Give the system enough room to learn. Automated bidding strategies need a learning period, and interfering with them too early undermines the process. Set a budget you are comfortable running for at least two to four weeks without making significant changes, and resist the urge to adjust targets or creative before the system has had time to stabilise. The instinct to intervene when early numbers look uncertain is understandable, but it is usually counterproductive.

Maintain creative input as a strategic lever. The one area where human judgment remains clearly superior is in understanding what will resonate with an audience at a brand and message level. AI can tell you which creative variation performed better within a campaign. It cannot tell you whether the creative is saying something worth saying. Invest in creative quality and variety, and treat the machine as a distribution mechanism for your best ideas rather than a replacement for having them.

Triangulate your measurement. Do not rely solely on platform-reported data. Cross-reference with your own analytics, with incrementality testing where possible, and with actual business outcomes like revenue and margin. The machine is telling you what it sees from its vantage point. Your job is to understand whether that matches reality.

If you want to go deeper on how paid media fits into a broader acquisition strategy, the paid advertising section covers channel strategy, budget allocation, and how to think about performance measurement across different campaign types.

The Structural Question for Agencies and In-House Teams

AI media buying is not just a tactical question. It is a structural one for anyone running or working within a marketing team.

For agencies, the compression of execution work means that the value proposition has to shift. If the machine is doing the bid management, the agency’s value is in strategy, creative, measurement design, and the kind of commercial judgment that comes from working across multiple clients and industries. Agencies that have not made that shift and are still positioning primarily on execution capability are in a difficult position, because the execution is increasingly commoditised.

For in-house teams, the risk is different. Automation can create the illusion of control. The dashboards look good, the platform is reporting strong ROAS, and no one is looking too hard at whether those numbers connect to actual business performance. The discipline of interrogating your own results, of asking whether the machine is optimising toward the right thing, requires a kind of intellectual rigour that can atrophy when the day-to-day execution is handled automatically.

The marketers who will get the most from AI media buying are the ones who understand it well enough to know when to trust it and when to question it. That requires knowing how the systems work at a mechanical level, having a clear view of what good commercial outcomes look like, and being willing to override the algorithm when the evidence suggests it is headed in the wrong direction. That is not a technology skill. It is a judgment skill, and it is the one that matters most.

About the Author

Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.

Frequently Asked Questions