Behavioral retargeting in Yandex: four traps that break your segments

Behavioral retargeting promises precision. You reach the person who looked at a product, abandoned a cart, read three pages and left without leaving a phone number. The promise can be kept, but only when there is order underneath it.

Messy tagging turns segments into nonsense. An attribution report credits a source that did not really bring the person in. A segment cut too fine becomes statistical noise. And the habit of trimming the top of a distribution makes you undervalue exactly the people who produce the result. These are four conditions without which segmentation leads you in the wrong direction, and each of them looks slightly different in Russia than in the markets where most foreign marketing teams learned their trade.

What changes when you retarget in Russia

A company used to Google and Meta arrives with a familiar toolkit: a GA4 property, remarketing lists in Google Ads, custom audiences built from the Meta pixel. None of it works here. Google Ads stopped serving ads in Russia in 2022, and Meta platforms are blocked. The audiences you spent years building do not move with you.

In Russia behavioral retargeting runs mainly through Yandex. Visitor behavior is recorded by the Yandex Metrica tag, goals and segments are defined in Metrica, and Yandex Direct uses them as retargeting conditions. Uploaded lists, such as customer emails or phone numbers, go through Yandex Audiences. VK Ads has its own pixel and its own audiences, built separately again.

Two practical consequences follow. First, retargeting starts with data you do not have yet. Before the first campaign you need the Metrica tag on every page, goals that match real actions, and enough time for the segments to fill. Second, anything that contains personal data, such as a customer list for upload, falls under Russian personal data law, which requires personal data of Russian citizens to be recorded and stored in databases located in Russia. A CRM export from a server in Frankfurt is a legal question before it is a marketing one. Check with a lawyer before you upload anything.

Retargeting starts with clean tagging

Every segment rests on how traffic is tagged. When tags are typed by hand into every link, drift is only a matter of time: a typo here, a forgotten parameter there, a wrong value somewhere else. A segment built on that tagging groups together people who should never have been grouped.

The fix is a single source of truth for tagging. Yandex Direct now lets you set one tracking parameter at the level of ad groups and individual ads. In Google Ads the same job has long been done by the tracking template, which appends parameters to the final URL automatically. The logic is identical in both systems: fewer manual errors, bulk changes to parameters without rewriting every link, and support for large campaigns where the links run into the hundreds.

As long as tagging is standardised and the analytics actually consume it, the segments are honest. Once tagging becomes chaotic, any neat segment is an illusion of precision. Clean tagging is the foundation retargeting stands on, and treating it as paperwork is how accounts end up retargeting the wrong people with confidence.

For a foreign team there is an extra reason to be strict. Tagging conventions are often inherited from the home market, written in English and designed for GA4 reports. When the Russian account is launched by a local contractor, a second convention appears next to the first. Agree on one scheme before launch and make sure Metrica reads it.

The most valuable segment is often missing from the report

Once the data is tagged, a second trap opens: crediting the result to the channel that sits last in the chain and missing the one that actually started it. An attribution report will honestly show where the last click came from. It will not always show what brought the person to the brand.

The case of one online store makes this concrete. Most of its conversions came from branded traffic, people searching for the store by name. The store ran no ads on branded queries at all. People learned about the brand from sponsored placements with bloggers in video formats and then searched for the store themselves. A conversion from that segment cost less, and a higher share of those visitors went on to pay, than traffic from classic performance channels. In the reports, this cheap and high-converting segment had no link to the budget that created it, because the blogger placements were funded separately.

The lesson for segmentation is plain. When you judge channels, you cannot ignore indirect sources: sponsored placements, influencers, any activity that builds recognition. They often feed the most profitable segment, direct visits and branded search, even when the report draws that segment as if it appeared on its own. Switch off what looks unprofitable on last click, and you risk cutting the source of your cheapest flow.

The branded segment has a flip side. Its numbers can look so good that it is tempting to give it all the credit and wind down everything else. But someone created that branded demand, and if you stop feeding it, the segment will shrink along with recognition. So a segment is judged by two things: how much it brings now, and what feeds it. Retargeting catches people who are already warm, but someone has to warm them first, and in a last-click report that someone nearly always looks smaller than it is.

This matters more for a company entering Russia from abroad. Brand recognition there is usually close to zero at launch, so the branded segment has nothing to feed on until someone builds awareness locally. Metrica offers several attribution models, and comparing them side by side is a cheap way to see which channels disappear under last click.

Do not slice a segment into noise

Segmentation is absorbing, and there is a pull to cut ever finer: by device, region, time of day, specific action. Slicing has a price. The smaller the segment, the fewer events it holds, and the less any percentage calculated on it can be trusted.

A typical scene: a segment had two conversions last week and one this week. In relative terms that is a fall by half, a reason for alarm and urgent changes. In absolute terms it is a difference of one conversion, ordinary random variation on a small sample. Percentages that move “by a multiple” mislead precisely because on a small base a single random visit, or its absence, changes the picture completely while saying nothing about traffic quality or settings.

The practice follows directly. Before you conclude that a segment has got worse and start optimising it, look at the absolute volume of data behind the conclusion. If it is a handful of events a week, the sensible move is to stop tuning settings to noise and either collect more data or use a larger unit of time, a month instead of a week. The threshold where fluctuation stops being noise depends on the base conversion rate, and no universal number exists. There is one guide: the fewer events in absolute terms, the less trust the percentages deserve.

Retargeting segments are small by nature, since they are a subset of people who already visited. A foreign company testing Russia on a modest budget will have smaller segments still. That is a reason to start with a few broad retargeting conditions and split them only once each part has enough events to judge.

A minority of the segment brings most of the result

The last trap is in how the distribution inside a segment is read. Revenue, traffic and sales nearly always follow a power law: a small share of objects concentrates most of the value, and the rest is a long tail of low values. A small share of placements delivers most of the conversions. A small share of customers delivers most of the revenue.

A common substitution is built on this. Remove the top slice from the picture, say the few percent of objects with the highest values, and what remains looks far more even and modest, because the bulk of the spread sat in the slice you removed. Formally the data is not distorted. A convenient cut has simply been chosen, and the conclusion comes out opposite to what the full distribution shows.

For anyone who segments, this is a warning about their own habits. Cutting an “atypical” minority as outliers makes it easy to throw away the very part of the segment that produces most of the result. So whenever you look at a chart with a share of objects trimmed off, ask where the cut was made and how much it weighs in absolute numbers. The minority you want to drop for the sake of a tidy average is often your main source of money.

What to do with your segments

A short route:

1. Put tagging in order. Set one tracking parameter at the level of groups and ads, so segments are built on clean data instead of manual typos. Make sure the Metrica tag and goals are in place before retargeting starts.
2. Look for indirect sources. Before you switch off a channel on last-click numbers, check whether it feeds your branded and best-converting segment. Compare attribution models in Metrica.
3. Do not cut down to single events. Check the absolute volume behind a segment and do not optimise for a swing of one or two conversions.
4. Look at the full distribution. Ask what share of objects was cut from a chart and how much result that share holds.
5. Count money, not percentages. Judge a segment by its contribution to revenue and inquiries, not by how good its relative dynamics look.
6. If you sell into Russia from abroad, rebuild audiences in Yandex from scratch and settle the personal data question before uploading any customer list.

Segmentation and retargeting are as strong as the data beneath them and as honest as the reading of the distributions above them. A precise segment on messy tagging, a verdict based on last click, panic over one conversion and a trimmed top all lead the wrong way with equal confidence. Order in the data comes first, dividing people into groups comes second.

On a free review I can go through which of your segments rest on honest data and which show a mirage. How to build a traffic system where segments and retargeting work for sales is covered on the Yandex Ads page.

Frequently asked questions

Sources

• Yandex Audiences help, building and using audience segments: yandex.com/support2/audience/en
• Yandex Metrica help, attribution models: yandex.com/support/metrica/general/attribution
• Russian version of this article on belokrylovo.ru: Сегментация и ретаргетинг по поведению