The False Dilemma of the Performance Pilot

There’s a dilemma that has become common sense in media buying: either you hand everything over to the algorithm, or you fall behind. It’s a false dilemma. And understanding why it’s false may be the most important investment decision of the next decade in media.

Digital media has evolved enormously over the past 20 years. What used to be signature work — built on sharp knowledge and individual repertoire — is increasingly in the hands of the algorithms of Google, Meta, TikTok and the marketplaces. The work of media is still intense, but today it leans mostly on performance analysis and metric tuning.

It’s not rare to see budgets migrating daily between channels in search of the perfect ROAS. When it’s reached, it creates an unstable equilibrium: any increase in scale or saturation of the base sends it downhill. And so the work of migration and optimization remains — important, but tedious to carry out. A grinding routine of adjustment, optimization and result, an infinite loop for as long as the campaign lasts. Because it’s still invisible work, it depends on the trust placed in whoever holds the active login on the network.

Over the years, platforms made this process more opaque and more automatic. Once the parameters are set, the algorithm pursues continuous improvement to guarantee the largest possible consumption of the budget — but always within the platform’s “walled garden.” At the same time, budget concentrated more and more at the bottom of the funnel, with few test-and-control metrics able to attribute the result to the channel that actually supported the sale. Separating test and control across a population base tagged per campaign was always complex — and became nearly impossible in a world of anonymized data.

If the algorithm’s conveniences brought scale — a single performance pilot now optimizes hundreds of micro-campaigns — they also created a new scenario for anyone who truly wanted to break away from the competition on the same battlefield.

In the usual scenario, where everyone uses the same look-alikes, the immediate effect is more reach and positive results. But with a rising bid right after: the competition plays the same game, and the auction flattens the market’s margin. It’s a checkmate in two moves, straight out of the prisoner’s dilemma. On the first move, the gain seems obvious — I’ll talk to whoever is most likely to buy, using free data. On the second, your competitor did the same, with the same audience, from the same platform, using your conversion data. Your sales drop, unless you accept a higher bid.

When we look at the history of CPC and CPM in digital, few years passed without double-digit growth.

Even with inventory growing, demand grows faster. And it doesn’t grow evenly: it concentrates on the audiences everyone wants to buy, the fruit of look-alike optimization. That clustering of views decouples from the rest precisely because it’s the market’s focus. While 68% of the budget fights over the same inventory, the other side — the media void nobody explores — stays untouched.

Algorithms don't raise CPC through inefficiency. They raise CPC through collective efficiency. The better the machine gets at identifying valuable users, the more advertisers converge on the same audiences.

— Gabriel Villa

The individual gain in performance turns into competitive pressure on the auction. In the short term, the advertiser buys better. In the long term, the whole market pays more for the same user.

There’s another kind of professional who doesn’t surrender to the algorithm. Always after first-party data and proprietary attribution, without delegating strategy to a black box that — despite its countless parameters for experimentation — is opaque in how it executes. This model is more effective in the long run, but it demands articulating the strategy very well: skipping the sales shortcuts to build a base, seeking alternative channels, keeping conversion metrics internal. It also involves cost. Without the media’s free data, you need systems like an internal CDP, or paying a tech fee to import and export data on platforms like DV360.

In the recent past, this was a low-scale model. It’s laborious to personalize campaigns, run test and control, measure uplift across channels, chase attribution. State-of-the-art companies still preferred a few good campaigns, so as not to contaminate the analysis.

If personalizing campaigns was complex, artificial intelligence changed the math. Creative management became automated, with conversion tests requiring no direct involvement. Craft gained far greater scale without losing quality.

On the other hand, there’s the promise of full, end-to-end management by the black boxes — the pilot connects the sales and catalog APIs and hits play for optimization. Having seen this movie before, I’d bet it’s a trend toward higher CPC and lock-in through channel dependency.

But there’s a movement in the opposite direction: the democratization of platforms. AI made it easier, for those who already had repertoire, to accelerate development. And several new avenues for publishing ads, with transparent management, will appear. Precisely because channels have proliferated, and it no longer makes sense to concentrate the whole strategy in a single environment — or to operate without correct attribution.

The challenge becomes a different one: how do you operate more channels without adding complexity? How do you keep attribution, governance and analytical capacity in an ever more fragmented scenario?

That’s the space NexOS occupies. Supporting media through platforms, without replacing the black boxes — but guaranteeing visibility into the data, integrating channels and helping with decision-making. AI enters as support for operation and personalization. Control stays with whoever sets the strategy. It’s a platform built by people who understand flying media with craft: it opens all the data to hit the target audience, with no hidden look-alike, and still uses AI to build the personalization of each place — including the regional channels Google can’t see.

For anyone who makes this decision — investing with proprietary intelligence — comes more control over the future of media costs. Alternative channels sit together in one place, and the competition shifts to how you manage them — not to the black box’s monopoly.

How many times have we watched the simple traded for the simplistic in media buying? The idea that all we needed was to simplify optimization, expecting the algorithm’s intelligence, on its own, to hit the sales and cost goals. Models with little room for inference beyond keyword selection and the input of proprietary data.

Now we have organized access to public, structured data so that media buying can be simple — without giving up the pilot’s repertoire. You are the one who controls and says where the investment goes, in what form, with which creative. And the insight reports are open enough for you to build the next iteration of the campaign yourself, with more focus and profiling.

The future promises a lot. Power to the people. AI is a reality that can support real intelligence — and that’s the trend we have room to occupy. It’s up to us to use what exists to keep ourselves in the loop.

If that materializes, we’ll need to build an MMM — a marketing mix model — able to pull digital data and get attribution right. It’s far more complex to manage channels when the share in the alternatives climbs above the usual 10%. And that trend will arrive when we move beyond the DV360/Meta duet. To this day I haven’t seen anyone bring all the sources together with the fine detail needed to attribute correctly between digital and offline — but AI is here to process all that data and bring some answer.

The algorithm can optimize execution. Strategy remains an exercise of repertoire.

— Gabriel Villa

Meanwhile, I keep betting on proprietary intelligence, channel diversification and the judicious use of AI. Leaning on NexOS to keep my CPC in check, with a search for geographic intelligence and alternative channels. The false dilemma was never between the pilot and the machine. It’s between renting your strategy from a black box or building your own — and staying its owner.

About the author: Gabriel Villa is head of Data Insights, Ads & Martech at Claro, where he builds a data-driven omnichannel marketing stack. As e-commerce head he grew ShopFácil.com’s GMV 41x, and he delivered Brazil’s first transactional chatbot, in 2016.

This essay is part of Tramas — territorial intelligence as method. For the territory reading behind the argument, walk through the NexOS Planner and the map of the Brazilian media blackout.