What is Yandex Audience?

A service for assembling your own audience segments, which are then used in advertising campaigns. The source can be an uploaded list of contacts, Metrica counter data, geolocation or pixel, and a similar audience is built based on the finished segment.

How many contacts do you need for a segment?

The segment has a filling threshold: a list that is too small will not be processed and will not become available for targeting. It’s worth uploading with a reserve, because not the entire list will match—some of the contacts will not be matched with users.

How does look-alike work in Yandex Audiences?

The system takes the original segment and finds users with similar behavior. The accuracy is adjusted: the higher the similarity, the smaller the coverage and the higher the quality. The quality of the result is entirely determined by the quality of the source.

Why is the segment not gaining coverage?

Three common reasons: too few records matched during loading, the conditions in the segment itself were too narrow, or the maximum similarity accuracy was selected when the source was small. All this is checked by the size of the segment after processing.

Why exclude existing customers from campaigns?

To avoid paying for impressions to those who have already purchased. The customer segment is excluded, and the budget stops going to an audience that will not bring in new sales. This is the easiest savings to set up and the most overlooked.

How often do I update downloaded segments?

Once a quarter at least. Contacts change, interests disappear, some people become clients and must be excluded. A database that has not been updated for a year performs worse than its absence, because it creates the illusion of targeting.

All articles

PerformanceSeptember 5, 20269 min

Yandex Audiences: segments, look-alikes and why they don’t work

I’ll look at what segments can be collected in Yandex Audiences, what is the minimum size needed, how to build similar audiences, and three reasons why a segment refuses to gain coverage or brings in untargeted traffic.

Yandex Audiences is a tool that stands next to the advertising account and is, at best, half used. Usually they download a phone database, build a similar audience once and never come back.

A segment is not a list of people, but a hypothesis about who is worth showing. It needs to be formulated as carefully as an offer.

1. What can a segment be assembled from?

Source	What does it give?	Limitation
Uploaded contact file	Accurate hit to your base	Not the whole list will match
Metrics counter data	Visitors by conditions: pages, goals, time	Only those who have already been on the site
Geolocation	People who regularly visit the point	Works for offline tasks
Pixel	Those who have seen media advertising	Requires pre-posted code
Similar audience	Extension of any of the above	Quality equals source quality

The second line is the most underused. A segment from Metrica can be collected not just by the fact of a visit, but by achieving a goal, by the depth of viewing, by specific pages - that is, by behavior, and not by the fact of arrival.

2. Filling threshold

A segment has a minimum size, below which it is not available for targeting. This is not a limitation of the service for the sake of limitation: with too small a sample it is impossible to either build a similar audience or ensure anonymity.

Practical consequence: need to be loaded with reserve. Not the entire list will match - some contacts will not be matched with users, some are out of date. If you loaded exactly at the lower limit, there is a high probability that the segment will remain inaccessible.

If the segment after processing turns out to be small, the reason is almost always the quality of the data, and not the service.

3. Look-alike: the source is everything

The mechanics are the same as in other systems: the algorithm looks for users similar in behavior to those you provided. And in the same way, he does not know how to distinguish your best client from the person who accidentally placed an order and returned it.

Hence the main rule: the source is taken the narrowest quality segment, not the entire base. The logic is discussed in more detail in the material about audience temperature and look-alike.

Setting the similarity accuracy is the second lever. High accuracy gives small coverage and good quality, wide - vice versa. A reasonable order: start with the exact one, look at the volume and expand only if it is not enough to unscrew.

4. Exceptions that almost no one makes

The existing customer segment is used in two ways. The first obvious one is to build a look-alike based on it. The second one that is skipped is to put it in the advertising campaign exclusions.

The meaning is straightforward: a person who has already purchased should not see the ad “buy for the first time.” Each such display is wasted money and, worse, a spoiled impression.

An exception is issued in a campaign in a minute and usually provides noticeable savings immediately, without any bid optimization.

5. Why do segments go bad?

Loaded database - a snapshot at the time of loading. Then it ages: contacts change, interests fade, some people become clients and must be transferred to another segment.

After a year without updating, such a segment is not only useless - it is harmful because it creates the illusion of precise targeting where it no longer exists. The campaign looks set up, but impressions go by.

Once a quarter: re-upload the current upload, subtract those who bought, check the size after processing.

6. Three reasons why a segment is not working

Few matches when loading. The file is small or the data is old. Checked by segment size after processing.

Too narrow conditions. A segment from Metrica with five simultaneous conditions may not recruit anyone. Conditions are removed one at a time until the segment is full.

Maximum similarity accuracy with a small source. A thousand people and the requirement of high similarity provide an audience that is not enough for promotion. It is treated by expanding the similarity, not by increasing the rate.

7. Summary

Segments are collected not only from the downloaded database: Behavioral Metrics data usually provides more meaningful audiences than a list of phone numbers. Look-alike is built from a narrow quality source. Clients are made exceptions. The databases are updated once a quarter.

Of all the above, the most underrated thing is exceptions. It requires no data or analytics, and saves your budget from day one.