The Algorithm Does Exactly What You Ask, and That Is the Problem

Most advertisers think they are buying attention. Dima Popov would like them to understand that they are entering an auction, and that almost everything that goes wrong downstream starts there.

“Meta and Google, as well as many other modern platforms, are actually auction based,” he says. When someone opens the app, campaigns are told a little about them, and “in microseconds, very quickly, an auction happens that determines which ads will be shown to this user”.

What decides the winner is a bid, and what decides the bid is a prediction: how likely you are to click, how likely you are to buy, and how much that is worth. “All of those three things multiplied together essentially give you the maximum and the most optimal bid.”

The number on the dashboard is not the number that matters

Before anything else, Popov wants advertisers to be honest about what they are measuring.

Return on ad spend, revenue divided by advertising cost, is the figure the platform displays and the figure marketers report. He is blunt that it is not the same as return on investment.

“Even though the return on advertising spend might look attractive, it doesn’t mean that you will get strong ROI if you account for fixed cost. If you account for agency fees. People who manage your ads. If you account for designers who create your ads.”

A campaign can look profitable on the platform and lose money in the business. That gap is where a lot of marketing budget quietly disappears.

It will give you exactly what you asked for

The deepest problem he describes is not that the algorithm is stupid. It is that it is extremely literal.

Targeting used to be something you controlled. Privacy law and platform economics changed that. “They started to hide as much audience segmentation information as they can.” What you can still set by hand is thin: age, gender, geography, and some interests the platform infers and you cannot check.

So the platforms made an offer. Stop trying to describe your customer. Tell us what action you want and pay us for that action. “Let the algorithm figure out the rest.”

The trade is real, and so is the cost. “It’s a black box, and you really need a high quality conversion action available to you.”

Then comes the part that catches almost everyone. Tell the platform you want leads at twenty dollars, and it will bring you leads at twenty dollars, and it will not care what they are.

“The Facebook algorithm will just say, okay, every lead is created equal,” he says. “A lot of advertisers basically have to deal with lead quality deterioration exactly because of this.”

Why would it not? Cheap leads are easier to manufacture than good ones. “It’s a lot easier to create a clickbait creative that would drive a lot of traffic, and then use some baits to trick them into clicking specific buttons.”

At this point he starts speaking as the algorithm, which is the clearest way anyone has put it. It has no human component, he says, so it simply reports for duty: “I’m a mathematical equation. And as long as I’m maximising something they want me to maximise, I’m doing okay.”

The fix is not a better campaign. It is better information going back in. Score your leads, give them different values, and feed that back so the system learns which conversions are worth buying. “If you do not close the feedback loop, if you do not assign conversion value for those leads according to their quality,” you have told the machine that a time-waster and a buyer are the same thing.

Similar is not the same as interested

Of all the targeting tools, the one he thinks is most misunderstood is the lookalike audience, where you ask the platform to find people resembling your existing customers.

“What makes people similar to each other does not necessarily correlate strongly enough with purchase intent.”

His example is himself. Another man of similar age, in his town, also working in a startup, matches him on every visible axis. “But let’s say I recently bought an iPhone, and this person is a super big fan of Huawei or Samsung, and doesn’t want to succumb to the Apple ecosystem. We will be completely different in terms of our purchase intent.”

Creative or targeting depends on how big your market is

There is a fashionable claim that creative is now the only lever that matters. Popov half agrees, then draws a line most people miss.

He accepts that a creative speaking precisely to one customer profile can make up for sloppy targeting. But he warns against judging it by engagement. “The highest click-through rate advertising asset would be something closer to clickbait.” It will bring traffic, and it will bring the wrong people. “They compel people who are easily compelled, rather than people who are the most valuable to your business.”

The line he draws is about market size. If you sell something broadly consumer, creative carries the weight. If you are an enterprise vendor whose entire addressable market is five or ten thousand companies, it does not, because you cannot afford to waste impressions on strangers.

“You can have a pretty lousy creative, but if it says, you have problem X and we can help you solve it, and you show it to the right people enough times, that would move the needle a lot more than if you had a perfect creative wasted on a random audience.”

Small budgets and large ones are different games

His closing advice inverts the intuition that big spenders need more discipline.

“The smaller the budget you have, the more manual and the more strict you need to be.”

A large advertiser can hand the algorithm freedom, because they generate enough conversion data for it to learn from. A small one never will. So a small budget demands one clear hypothesis about the audience and the message, tight manual control, and a set of leading indicators standing in for conversions you do not yet have enough of.

“You will need to invent a proxy metric that basically tells you that you are on the right track.”

Whatever the budget, he thinks the sequence is the same.

“You need to get into your customer’s head first. Even if you have a bigger budget, you need to validate those initial hypotheses about what your ideal customer profile is, and what the exact offer is, before you scale.”

A large budget does not exempt you from that. It only lets you run the test faster.