Digital advertising was supposed to kill the black box. Instead it has built a bigger one and convinced marketers that anything dressed up as artificial intelligence deserves to be trusted.
Digital promised transparency and control. We were told that we would finally know who saw the ad, where they saw it, what they did next, what it cost and whether any of it actually mattered. And to be honest, for a while digital did feel like progress. Search gave us keyword control. Social gave us audience targeting. Programmatic gave us bidding logic, frequency caps, site lists and the intoxicating feeling that marketers had finally been allowed to touch the machinery. We could see the levers. We could pull them. We could even pretend we fully understood them.
Somehow, the next stage of progress has become fewer levers, fewer explanations and one large button that says: “Trust the algorithm.” This is the great irony of modern digital media. The technology that was supposed to democratise advertising is becoming more restrictive, more automated and more opaque. The platforms are not asking marketers to become better pilots. They are asking them to become passengers. And most marketers have quietly taken the window seat.
Google App Campaigns were the warning shot
Google App Campaigns were the warning shot. Launched a decade ago as Universal App Campaigns, they took app install advertising and quietly removed most of the controls buyers had grown used to. No manual keyword targeting. No real audience building in the old sense. No meaningful choice over whether your spend goes into Search, YouTube, Display, Discover or Google Play. You give Google your app, your assets, your budget and your conversion goal. The system decides the rest.
In theory, elegant. In practice, it means the platform controls targeting, bidding, inventory selection, creative assembly and attribution while the advertiser gets a dashboard and the comforting illusion of command. Since then, the direction of travel has been obvious. Performance Max extended the same logic across Google’s wider ecosystem, bundling Search, YouTube, Display, Discover, Gmail and Maps into one automated campaign type. AI Max for Search pushes even traditional search further towards automated query matching, creative adaptation and bidding. Meta’s Advantage+ and TikTok’s Smart+ follow the same playbook: fewer manual levers, broader algorithmic control and more faith placed in platform-reported outcomes. The names differ, the pitch decks differ, but the promise is always the same. Give the machine your goal and budget, then please stop asking exactly where the money went.
The algorithm is not your auditor
Here is what makes this particularly worth scrutinising. These algorithms are not designed by neutral monks living in a monastery of media efficiency. They are built by publicly listed companies with revenue targets and shareholders who are not famous for their patience. Of course they are designed to deliver results for advertisers, but only within a system where the platform also needs to grow revenue, protect margins and keep spend flowing through its own pipes.
That does not make the platforms evil. It just makes them businesses. The algorithm is not sitting there thinking, “How do I get this brand the absolute maximum value for every dollar?” It is operating inside a commercial environment designed, measured and monetised by the same company that owns the inventory. That is a significant amount of responsibility for a single black box and a significant conflict of interest that the industry largely chooses to discuss in the abstract rather than confront in practice.
Attributed results are not the same as growth
The problem is not that algorithms are bad. Often, they are very good. The problem is that the algorithm is judge, jury, trader, auction participant, measurement provider and occasionally creative director. When platforms optimise toward conversions, they naturally seek the easiest conversions. That can mean remarketing pools, branded search, existing customers, high-intent users or people who were likely to convert anyway. Without incrementality testing, clean holdouts, path-level transparency and meaningful placement reporting, advertisers can mistake capture for creation. They think they are growing demand when they may simply be buying credit for demand that already existed.
Better technology should not mean less control
Of course, the future cannot be a return to manual everything. Nobody sane wants to go back to adjusting thousands of bids by hand like medieval monks illuminating manuscripts. Automation is necessary. The auction environment is too complex, too fast and too signal-rich for humans to manage manually.
But automation without transparency is not progress. Fundamentally, better technology should give advertisers more transparency, not less. It should make platforms work harder for every dollar, not ask brands to surrender control in exchange for a prettier dashboard. If the machine is making thousands of decisions in real time, that should be a reason to report back more clearly, not an excuse to report back less.
Marketers need to stop being polite
The platforms will continue to push automation because it works beautifully for them. It reduces friction, increases spend velocity and keeps optimisation inside their walls. The role of agencies and marketers is to put constraints around it, not abandon it.
That starts with demanding more transparency from the platforms themselves. Advertisers should not have to beg to understand where their money went. Placement reporting, query visibility, audience quality, creative combinations and clear separation between prospecting and remarketing should not be treated as luxury features. They are basic requirements if brands are expected to hand over larger parts of their media decision-making to an algorithm.
It also means asking harder questions internally. First, ask for placement-level reporting, not just channel-level aggregates. If you cannot see where the money went, you cannot assess whether it was well spent. Second, run holdout tests before accepting attributed results as proof of growth. A clean geographic or audience holdout, held for long enough to be meaningful, is the minimum bar for claiming that automation is actually working. Third, ask the one question that the current measurement paradigm is almost entirely designed to avoid: what share of my conversions came from users with no prior brand interaction or purchase history? If the answer is low, the algorithm is harvesting existing demand, not building new.
The future of media buying will not be won by marketers who reject automation. It will be won by marketers who know how to interrogate automation, challenge its outputs and refuse to confuse attributed results with real business growth.
So yes, use the machines. Use the algorithms. Use the automation. But don’t worship them.
When a platform says, “Trust us, the algorithm has worked it out,” the right answer is not blind faith.
It is: “Show me the receipts.”
Kabeer Chaudhary is the global CEO of M&C Saatchi Performance
