Ad account metrics that lie: what the business pays for

The marketing team comes to the owner with good news. Click-through rate is up, there are more clicks, the ads are “working better”. A month later that team is let go, because sales are exactly where they were. There is nothing paradoxical here. The numbers inside an advertising account and the money in the company’s till are two different realities, and the first one lies about the second on a regular basis.

A business does not hire a specialist so that CTR grows. It hires one to earn more. Between pressing buttons in an account and money arriving in the till lies a whole layer of other people’s responsibilities: the site, the offer, the sales team, the prices. So an isolated channel metric says very little about the result, however pretty it looks. And for a company that sells into Russia from abroad, the layer is thicker than usual, because the account is in Yandex Direct, the analytics are in Yandex Metrica, the reports arrive through a contractor, and each of those adds its own way of making a number look better than it is.

Here are the figures that mislead most often, why the Russian setup makes them slipperier than in Google Ads, and what to look at instead.

CTR, CPC and “cheap leads” protect nobody

Intermediate account metrics describe a process, not an outcome. Click-through rate, cost per click, the number of inquiries, the cost of an inquiry are gauges on the dashboard, not the destination. The specialist who is pleased with a low cost per lead often genuinely does not understand why the client stopped working with them: “there were leads, the price was fine, the queries were relevant”. The client stopped because the leads never turned into sales, and the business saw that before the specialist did.

The trap is not that account metrics are useless. They are needed, as gauges. The trap is mistaking them for the goal. As soon as the focus shifts from “my CTR is good” to “the client is earning more”, everything changes: the decisions inside the campaigns, the conversation with the customer, and how long that customer stays.

In Russia this trap has a specific mechanical form. Yandex Direct pulls its conversion column from Yandex Metrica goals linked to the campaign, and Metrica since 2021 creates automatic goals on its own: a click on a phone number, a click on an email address, a form submission, a click on a messenger link. A contractor who ties every one of those goals to the campaign gets a conversion count that looks generous and a cost per conversion that looks modest. It is simply that “conversion” in that account means “someone touched something”, and the automated bidding strategy, if it is set to optimise toward those goals, will happily chase touches instead of buyers. A foreign head office reading the translated report sees conversions and a price and has no way of knowing which events stand behind the word.

The practical shift is simple to state. Stop reporting account numbers and start talking about the business. Ask regularly about sales, about margin, about what happens after the inquiry. Even where end-to-end analytics is not in place and there is no exact revenue data, it is more honest to say up front that account metrics are a proxy rather than a result, than to defend a rising CTR later.

Cost per lead without a definition of a lead means nothing

“What does a lead cost you?” is a question people ask all the time, and it is not a valid question without a second one: what do you call a lead. Under that one word hide events of entirely different weight.

A file download is a lead. A request for a quote is a lead. A site visit by a company representative is also a lead. These events are not comparable: in the effort the person made, in the quality of the contact, in the probability of reaching a deal. Their cost can differ by orders of magnitude, and that is normal. A cheap light lead and an expensive heavy lead are not “bad” and “good”. They are different things.

From this follows something that breaks the habit of comparison. Cheapness on its own says nothing. A large flow of cheap inquiries can bring in less money than a handful of expensive ones, each of which closes into a large contract with add-on sales. The right criterion is not the price of a lead but its contribution to revenue, through conversion into a deal and average order value.

The main practical consequence: cost per lead benchmarks from other people’s chats and case studies do not transfer. Behind the same word, different projects have different events. Comparing your cost per lead with someone else’s is comparing the incomparable. Only one comparison makes sense: your own cost per lead over time, with the definition of a lead held constant. Change the target action, add a qualification step, and the series is no longer comparable even with itself.

For a company entering Russia the definition problem arrives twice. Once in the usual way, and once across the border. A lead on the English site is a form or an email. A lead on the Russian site is very often a message in WhatsApp or Telegram, or a phone call, because that is how Russian buyers prefer to make first contact. Write the Russian definition down in the contract with the contractor, tie it to the price a deal can carry, and treat any number reported without that definition as decoration.

The average hides exactly what you came to see

A separate category of lie is the averaged metric. An average is convenient to drop into a report, but it conceals the distribution, and the truth lives in the distribution.

Take average visit duration. Logic suggests: whoever spent longer on the site is more interested, so let us build a segment of “visits of several minutes or more” and push it with retargeting and higher bids. In practice that segment cuts off almost the entire target audience, including the people who actually buy. The overwhelming majority of visits, converting ones included, fit into far shorter intervals than several minutes. A sensible threshold for segmenting by time is on the order of ten seconds, not minutes. But this is visible only if you build a report that groups visits by duration, rather than looking at a single average figure.

Yandex Metrica gives you that grouping out of the box, in the “time on site” report with visits bucketed into intervals. It also gives you a related trap. Metrica’s bounce rate counts a visit as a bounce if it lasted under fifteen seconds and had a single pageview, which is a different definition from GA4’s engaged sessions with their ten-second threshold. Read each figure inside its own tool, and for decisions use the distribution, which does not depend on anyone’s threshold.

The same story with scroll depth. “Percentage scrolled” is a relative value, and comparing it between pages is meaningless: half of a short page is a couple of blocks, half of a long page is a dozen. The same number means different things. And even on one page the scroll percentage does not answer the main question: on which screen does the person leave, or, on the contrary, submit the inquiry. To see that, you need to track the visibility of specific meaningful blocks, not a general percentage. Metrica’s scroll map gives the same relative picture as any other heat map, so the block-level tracking has to be set up as goals on purpose. Then it becomes clear where the real hole is, and hypotheses for improvement stop being guesswork.

The conclusion from both examples is one and the same. Before drawing a conclusion from an average figure, look at how the distribution underneath it is arranged. Almost always the average hides exactly what you went into that metric to find.

What to look at instead

A short set of bearings that puts the numbers back in their place:

• Keep point B in mind. Not “a good CTR” but “the client earns more”. Every account metric is a gauge on the way there, never the finish line.

• Agree on the definition of a lead. Fix in writing what counts as an inquiry, and tie it to the pass-through price of a deal. For a Russian project, decide whether a messenger conversation and a phone call count, and make sure they are tracked. Without this, cost per lead means nothing.

• Count contribution, not cheapness. Look at what an inquiry brings through conversion into a deal and average order value, not at what it costs on the way in.

• Do not trust the average. For the key metrics build the distribution: how visits are spread by duration, on which sections people leave. The average hides, the distribution shows.

• Own the measurement. If you sell into Russia through a contractor, the Yandex Metrica counter should belong to your company, with the contractor as a guest. Then the list of goals, and therefore the meaning of the word conversion in every report, is something you can open and read rather than something you are told.

• Set up a conversation with the business. Regular feedback about sales is worth more than any dashboard. If there is no revenue data, say out loud in advance that account metrics are limited.

Metrics are not the enemy. The enemy is the substitution, when a gauge on the dashboard is taken for the destination. A furious client who was shown a rising CTR instead of rising revenue is angry with good reason: they were promised money and shown a chart.

The distance between the two is longer for a foreign company than for a local one, because the chart arrives in translation, from a tool the head office does not use, built on goals someone else defined. That is not a reason to distrust the Russian channel. It is a reason to insist on the definitions before the numbers.

If you look at your reports and cannot tell whether the business is better off or not, you are measuring the wrong thing. Which figures in your case are connected to money and which merely decorate the report is something we can sort out in a free review. How to build advertising where the path from click to sale is visible, I set out on the Yandex Ads page.

Frequently asked questions

Sources

• Russian version of this article on belokrylovo.ru: Метрики кабинета, которые врут: за что платит бизнес