If Google and Meta Are Doing the Media Buying, What Exactly Are You Paying Your Agency For?

In the early days of paid search, the work was extraordinarily manual. Advertisers moved data around on spreadsheets. Bids were adjusted by hand. Product feeds were primitive. Promotions required constant intervention. And if something quietly went wrong in a large account, you might not know until a human being found it.

At Voltage, we thought that was ridiculous.

So we built technology to do it better.

Our third iteration was called Voltage s3. The name was admittedly a little corny, but the technology was anything but.

s3 sat on top of the advertising platforms and automated much of what sophisticated advertisers were still doing manually. It managed bids. It built and optimized product feeds. It understood promotional calendars. It watched for sudden changes in spend or ROAS. It surfaced problems before they became expensive problems.

It was customizable. It consolidated enormous amounts of information. And at the time, it was genuinely ahead of the automation curve.

There were clients who would hire us for the technology alone.

For years, that was an enormous competitive advantage.

Then the platforms started catching up.

When your competitive advantage gets automated

As Google became more sophisticated, I started getting pressure to abandon s3 and move completely onto Google’s increasingly automated bidding systems.

Not just from sales reps. From senior people at Google.

Their argument was compelling.

Google had access to signals we didn’t. Their machine learning was improving rapidly. Their bidding technology could process an extraordinary amount of information in real time.

But there was something about the proposition that made me deeply uncomfortable.

The company selling my clients the media was also asking me to let its black box decide how much of that media they should buy, and then use its reporting to tell us how well it worked.

I argued about this with Google for years.

We held onto s3 as long as we could.

Eventually, though, the technology landscape changed underneath us.

It was a little like owning an older iPhone that can no longer run the newest version of iOS. The phone didn’t suddenly become bad. But eventually it stopped receiving the signals and capabilities necessary to compete with what came next.

That’s essentially what happened to s3.

Google’s ecosystem became increasingly closed and increasingly automated. The advantages we’d built began moving inside the platform itself.

Eventually, we made the switch.

And I lost some sleep over it.

If Google could increasingly do the things clients had historically paid us to do, what exactly were they paying us for?

That question ultimately changed our company.

Our job description changed

One of the first clients to really force the issue was Dennis Kirk, a large powersports retailer we’d already worked with successfully for seven or eight years.

For years, we’d driven significant growth for them using the tools and methodology we’d developed around paid search and Shopping.

Then CEO Bob Behan and VP of Marketing Mac McKenzie sat across from me in our Santa Monica office and essentially said:

We need more.

They had made significant investments in the business. They needed meaningful growth to support those investments, and they were willing to pay to acquire it.

There was an irony to the conversation.

Collectively, agencies like ours and Google had spent years teaching advertisers to obsess over ROAS.

Now Dennis Kirk wanted us to move beyond the very framework we’d helped reinforce.

Fortunately, we knew something important about their business.

We had years of data, a massive data warehouse and a client willing to rethink what acceptable customer-acquisition economics looked like.

So instead of simply trying to squeeze another few percentage points out of the existing campaigns, we started asking a different question:

How do we create more customers?

That took us beyond traditional paid search.

We moved higher in the funnel. We experimented with video and in-market audiences. We expanded channels. We learned how paid social and paid search could support one another. We used years of customer data to determine how aggressively we could acquire new customers while remaining inside an economically responsible range.

And we blew through previous revenue high-water marks while maintaining profitability.

I remember thinking:

Our job description just changed.

There is a significant difference between the two.

The machines got really, really good

That distinction matters even more today.

Google and Meta’s technology is extraordinary.

Their algorithms can process signals at a scale no human being could remotely replicate. They can identify audiences, predict behavior, adjust bids, allocate budgets, choose placements and optimize creative combinations millions of times faster than an account manager working through a spreadsheet ever could.

That’s progress.

I don’t want my people doing work a machine can do better.

But there’s another side to that sophistication.

Almost every week, I hear some version of the same sentence from an advertiser:

“What worked six months ago isn’t working anymore, and I can’t figure out why.”

A founder we spoke with recently put it almost exactly that way. His company had historically produced strong returns on Meta. Now ROAS was deteriorating and customer acquisition was getting more expensive.

Meta hadn’t forgotten how to buy advertising.

Something had changed.

The difficult question is what?

And the answer may not be inside Meta at all.

More often, it’s some combination of several of them.

That’s where the value has migrated.

The media buyer is also the media seller

If I were sitting across from a CEO today and he asked me:

“Mark, if Google’s AI is doing the bidding, targeting and optimization, why the hell do I need you?”

I’d give him a pretty simple answer.

Google and Meta may increasingly function like extraordinarily sophisticated media buyers.

But they are still, first and foremost, media sellers.

That’s not an accusation. It’s simply an important distinction.

Neither company is responsible for understanding every economic and operational constraint inside your business.

And most importantly, whether spending the next dollar the platform is capable of spending is actually the right decision for your company.

If simply turning Google’s and Meta’s algorithms loose were enough, most advertisers would already have the accelerated profitable growth they’re looking for.

They don’t.

That’s why experienced operators still matter.

Sometimes the right answer is not to do what Google tells you

We’ve learned this lesson more than once.

One of the clearest examples came during Google’s major push toward Performance Max.

To be clear: I thought PMax was compelling.

Smart Bidding was becoming extremely powerful. Google was bringing together signals, automation and inventory in ways that offered enormous potential.

What made me uncomfortable wasn’t the technology.

It was the urgency with which advertisers were being told to surrender what already worked.

For many of our ecommerce clients, Google Shopping had been the single biggest driver of customer acquisition.

And now we were effectively being told:

That meant giving up significant amounts of transparency, history, segmentation and customization around something we knew was extraordinarily valuable to our clients.

We moved portions of client feeds into PMax while maintaining traditional Shopping campaigns where we believed they still made sense. We compared performance. We learned where the new automation was stronger and where the existing architecture continued to produce better outcomes.

We took a fair amount of pressure to move faster and go all-in.

We held the line.

Eventually, Google’s own guidance evolved and validated that traditional Shopping could continue to make sense for certain portions of an advertiser’s product catalog.

The hybrid approach we’d been defending became accepted practice.

I don’t tell that story because Voltage was “right” and Google was “wrong.”

If PMax had crushed Shopping in our controlled tests, our responsibility would have been to switch.

Our responsibility wasn’t to be right about PMax. It was to make Google prove it was right with our clients’ money.

There is no substitute for experience

Years ago, my boss and mentor Dave Moore at 24/7 Real Media, himself a veteran of the birth of the cable industry, told me something remarkably simple:

“There is no substitute for experience.”

I’ve thought about that sentence a lot lately.

Because you could argue that AI should make experience less valuable.

I think the opposite is happening.

A machine can inspect an enormous advertising account at a scale no human ever could.

Our own systems can now find things like a huge group of ad groups quietly spending money with no approved ads running, something that could remain practically invisible to a human being trying to manually inspect an account of that size.

Every meeting, decision, test and change can be logged and timestamped. The expected outcome can be recorded. The actual performance can be monitored in the background. If reality diverges from the expectation months later, the system can surface it again.

That’s incredible.

And I want as much of it as we can get.

The expensive humans shouldn’t be data janitors.

Then put that information in front of someone who has spent years watching businesses succeed and fail.

Because the machine still has to know what matters.

A seasoned operator can recognize that the apparent advertising problem is actually an inventory problem. Or a merchandising problem. Or a margin problem. Or that an apparently healthy ROAS is concealing deteriorating customer acquisition.

The machines are getting extraordinarily good at answers.

Human value is increasingly in knowing which questions to ask.

Three questions I’d ask before hiring an agency today

If a CEO friend told me he was interviewing three performance agencies next week and asked how I’d separate the media buyers from the growth operators, I’d tell him to pay attention to three things.

First:

Watch their diligence process carefully.

That’s the classic make-sick sales process.

I’m much more interested in the make-well part.

And here’s my favorite:

What would cause you to tell me not to hire you?

You’ll learn a lot from the answer.

Second:

If the conversation never gets meaningfully beyond Google and Meta, you have your answer.

They should want to understand the economics of the business.

Margins. LTV. Inventory. Merchandising. Creative. Conversion. Customer mix. Operational constraints.

Nobody has a crystal ball.

You’re testing whether they’re thinking beyond next month’s campaign report.

Third:

I don’t think enough companies ask this question.

Agencies love talking about their dedicated account teams.

But people leave.

If John and Andrea have managed your account for two years and both disappear tomorrow, do two years of decisions, conversations, experiments, failures, wins and accumulated understanding disappear with them?

Or has the agency actually built institutional knowledge around your business?

This is one area where I think AI has the potential to fundamentally improve the agency model.

We’ve built our own system so that every conversation, meaningful decision, test, result, winning strategy and failed experiment becomes part of the institutional memory around that client.

And we’re now having experienced strategists periodically evaluate accounts managed by other strategists, not because something is wrong, but because fresh eyes see different things.

That’s a powerful combination.

So what exactly are you paying your agency for?

And increasingly, not access to some magical piece of technology that the platforms themselves will eventually replicate.

You’re paying for judgment.

You’re paying for someone capable of understanding your business well enough to determine what is actually preventing the next stage of profitable customer acquisition.

You’re paying for a system that gets smarter as it learns.

You’re paying for experience across hundreds of companies, business models, economic cycles, platform changes, successes and expensive mistakes.

You’re paying for someone willing to embrace technology when it produces a better answer, and willing to challenge it when the evidence says otherwise.

Twenty years ago, we built s3 because machines could do certain things better than people.

Today, we’re building systems for exactly the same reason.

The best operators aren’t the ones who resist automation. They’re the ones who know what not to automate.