Paid results are wobblier than they used to be. Conversion rates feel flatter. ROAS is harder to predict. It’s easy to say “Google’s changed again” or “Meta doesn’t work like it used to” or blame AI.

But most of the time, that’s a convenient excuse.

Platforms are smarter than ever. The likes of Performance Max, Advantage+, Spark Ads are built to do more of the heavy lifting than traditional campaigns ever could. The problem is what they’re being asked to work with.

The mistake we see most often

We audit a lot of paid media accounts at Tank. The most common issue isn’t budget, bidding, or even creative. It’s missing or meaningless audience signals.

The platform is told to “find converters” without ever being shown who a converter actually is. Campaigns go live with no CRM data, no high-value customer lists, no behavioural audiences beyond “all site visitors”, and no distinction between browsers and buyers.

That isn’t AI failing, it’s just bad briefing. You wouldn’t hire a strategist, give them no context, and expect miracles. The algorithms work the same way.

AI needs direction, not hope

Modern media platforms aren’t psychic. They optimise towards what you show them matters.

That means:

• Uploading real customer data, not vague demographics
• Building remarketing audiences based on meaningful engagement
• Using video views, time on site, and past purchases as signals
• Creating lookalikes from people who actually convert, not everyone who clicked

Strong inputs create strong outputs. Weak inputs create expensive guesswork. There’s no shortcut around that.

Automation doesn’t mean “hands off”

Automation has convinced a lot of marketers that paid media is now set-and-forget. It isn’t.

You still need to do the groundwork, the “real thinking” of it all. Among other things, you have to decide what a good user looks like, design creative that attracts the right attention, and give the platform enough signals to learn quickly.

If those pieces are missing, AI campaigns won’t just underperform, they’ll burn your budget on low-intent users and tell you very little about why.

So when someone says “Performance Max doesn’t work for us”, the follow-up question I often ask is: how did you brief it?

Like prompting AI: you just need to speak the language

Think of the platforms like junior strategists. Fast, capable, and adaptable – but totally reliant on the quality of the brief.

That means giving them the right signals to learn from. Not just broad goals, but real examples of what good looks like. It’s about feeding it the right data, the kind that sharpens targeting and improves decision-making over time.

A good way to start:

• Brand campaigns warm up your audience and build recognition, so the platform can distinguish between cold users and those already likely to convert. That improves how it prioritises who sees what – especially when budgets are tight.
• Conversion-led activity gives the system strong success signals – not just clicks, but actual outcomes. That helps AI spot patterns in behaviour and target more people like the ones who actually bought.
• Structured testing lets you control the signals you feed back in. Instead of throwing everything at it and hoping the system catches on, you’re deliberately shaping how it learns – and what it learns from.

Underpinned by a few core principles:

• Feed in real customer data – not just demographics, but lists of actual purchasers, loyal customers, and high-value segments. That tells the platform who to look for.
• Define meaningful success actions – optimise for outcomes that matter, like purchases or qualified leads, not soft metrics like clicks or time on site. That teaches the system what to aim for.
• Use structured exclusions – remove users who’ve already converted, or who bounced fast. This sharpens intent signals and stops the system learning from the wrong behaviours.

Each of these elements strengthens your targeting signals. And, better signals mean less guesswork, lower cost per acquisition, and a platform that actually knows what to optimise for.

Take the reins

Automation is the future of paid media. That part isn’t up for debate.

But the winners won’t be the marketers who trust it blindly. They’ll be the ones who understand how to speak to it properly and how to guide it better.

So next time performance dips, don’t start by blaming the algorithm. Start by asking whether you gave it anything worth optimising against.