Yandex Direct account audit, part 2: the technical checklist

The first part of an account audit answers one question: where to look first. The second part begins where the general impression runs out. It lives in the settings the platform switched on for you, in the metrics that jump without a visible reason, and in the data you are about to use for decisions. This is where it is easy to be fooled. An account can look neat and the budget still leaks through a setting nobody enabled on purpose.

For a company managing Russian advertising from abroad, the risk is higher: the account was often set up by a contractor, and the person paying for the clicks rarely opens the settings. Defaults survive for years in accounts like that.

Below are five technical zones where spend hides most often in Yandex Direct, and what to check by hand in each of them.

What the platform switched on without asking

Start with every feature that works on the “auto” principle. Advertising systems present them as a free improvement: auto-applied recommendations, autotargeting, extended geotargeting, automatic connection of new network placements, automatic bid increases. The interface wording makes it sound as if declining them means declining performance. In practice each of these features is a separate tap through which money leaves without your decision.

Go through the list item by item.

• Auto-applied recommendations change settings without a specialist’s confirmation, and the changes do not always work in your favour.
• Automatically created goals distort the attribution picture if nobody checked what exactly is being counted.
• Autotargeting extends reach to queries that are absent from your keyword set, and control falls as reach grows. There is a separate piece on autotargeting and competitor brands.
• Extended geotargeting shows ads outside the region you actually serve.
• Automatic connection of new placements lets network traffic in before anyone has checked its quality. How to keep that under control is covered in cleaning placements in the Yandex ad network.
• Automatic bid increases, built into some strategies, can multiply the cost of a click when limits are loosely controlled.

None of these items means automatic harm. At the testing stage some of them are appropriate. The point is different: each one has to be located in the account and switched on or off as a conscious choice, instead of staying on because the system suggested it.

Google advertisers will recognise the pattern: Google Ads has its own auto-apply recommendations, and Western PPC specialists already treat them with suspicion. The difference in Yandex is practical: the settings sit in different places and carry different names. A specialist who audits Google accounts well can miss half of this list in Yandex simply because they look for the familiar switch and do not find it. A second difference matters for foreign companies in particular: extended geotargeting quietly widens the area you pay for, and if you deliver only to a few Russian cities, clicks from outside them are pure cost.

When a metric jumps: a list of causes instead of panic

A sharp drop in leads or a spike in cost per lead is almost always explained by one cause from a limited set. Keep that set at hand and you locate the source in an hour instead of spending a week testing random hypotheses.

The causes fall into blocks.

Algorithms and learning. A campaign went back into learning after an edit. The strategy does not get enough conversions to learn steadily. Autotargeting mixed junk impressions into the traffic. Switching off one campaign unexpectedly affected the whole account.

Bids and spend. A bid adjustment suddenly poured expensive clicks into a single segment. The algorithm decided that a particular click was “the most likely to convert” and inflated the bid. The strategy spent a week’s budget in a day.

Fraud and quality. Bots inflated a metric. New placements were connected without a quality review. Traffic returned from placements that had already been blocked. The mechanics are described in the article on click fraud in the Yandex advertising network.

External environment. A new competitor entered the auction. Your own campaigns started cannibalising each other. Network campaigns pulled traffic away from search.

Offer and site. The offer is weak, the page loads slowly, or there are not enough landing pages for the keyword set.

Platform limits deserve a line of their own. In the simplified campaign types, some settings, extended geotargeting among them, cannot be switched off by hand at all. This is how the campaign type is built, not an oversight by the specialist. Historical statistics from past months do not guarantee the same cost per lead or the same lead volume going forward either: the auction and the market are volatile by nature. Sometimes a metric jumped because nothing was broken and the market simply moved.

Half of this list consists of real technical causes. The other half is a set of convenient explanations that can cover unfinished work. The task of an audit is to tell one from the other on specific data, instead of accepting the first version offered. If your contractor answers every jump with “the market changed”, ask which of the other blocks they ruled out, and how.

Search retargeting: the number first, the setting second

Search retargeting is tempting: show ads only to people who have already visited the site. Before making it a hard condition for serving, open Yandex Metrica and look at the share of returning visits.

The logic is simple. If returning visits are a few percent, keyword reach under such a condition shrinks by roughly the same factor. You cut off almost all of the traffic for the sake of a thin layer of people who came back. In practice, a hard “show only to retargeting audiences” condition usually restricts results instead of improving them.

The safer route is observation with a bid adjustment. Ads run on every keyword, and for returning visitors you raise or lower the bid without excluding anyone else. Google advertisers know this as the “targeting” versus “observation” audience setting. What differs is where the audience data comes from: in Yandex, the segments are built on Metrica data, so a site without a correctly installed Metrica counter has nothing to retarget to. Check that the counter is on every page, including pages added later. Also check that the campaign type you use allows the adjustment you plan, since the available options are narrower in some newer formats.

Combining autotargeting with search retargeting is worth a try, but the result there depends heavily on who the algorithm happens to pick up.

What data can actually be compared

Comparison is a trap of its own. An owner wants to know which campaign structure works better, a specialist cheerfully compares two setups and draws a conclusion. The problem is that the conclusion is often statistically empty.

For a comparison of structures to mean something beyond random noise, the account needs at least several hundred search conversions. Fewer than that is guesswork. Several kinds of projects fall outside this rule: ecommerce with its own dynamics, accounts where the goal is a micro-conversion instead of a real action, and niches with foreign-language vocabulary where one word is spelled in several variants.

This group matters most for international companies. A foreign brand or product name is searched in Latin letters, in Cyrillic transliteration, and often in two or three Cyrillic spellings at once. Data for one “keyword” is split across variants, each variant looks too small to judge, and a comparison built on them says little.

For an account on pay-per-conversion, comparing by cost per click is also meaningless: the payment model is different. While the data is thin, the honest answer to “which is better” is “too early to say”. That answer is stronger than confident conclusions made up to fill a report.

Competitors in numbers, not by eye

The last technical zone is market analysis. An impression from a couple of ads you happened to see does not replace the picture. When you collect the paid search results across a keyword set, you end up with anywhere from several hundred to fifteen hundred domains, and there is no way to sort them by hand.

A quick first-pass filter: a direct industry competitor shows the main target words of the niche right in the title, description and main tag of its home page. If those words are missing from the home page, the domain belongs to an adjacent topic, works across industries, or landed in the results by accident through broad match.

Then the domains go into a table and are sorted by the total search volume of the phrases each one appears for. Companies investing in advertising systematically rise to the top, and those who appeared at the level of statistical error sink to the bottom.

One detail: collect the data on a working day. A weekend with low advertiser activity will suggest that competition in the niche is weak when it is not. For a team outside Russia there is a second detail. The working day that counts is the Russian one, in Moscow time, and the Russian calendar has its own public holidays. A run scheduled for a working day at home can fall on a Russian holiday and give the same false picture.

What to check in your account this week

1. Find every auto feature in the settings and make a conscious decision on each one, instead of leaving the default.
2. Write a short list of typical causes for metric jumps and walk through it when numbers drop, instead of guessing.
3. Look at the share of returning visits before setting up search retargeting, and choose observation over a hard condition.
4. Check whether the account has enough conversions before comparing campaign structures.
5. Put competitors into a table sorted by search volume, instead of judging by a couple of ads.

A technical audit looks dull: no bright ideas, no attractive slides. Yet it is exactly here, in the checkbox next to automatic placement connection and in the share of returning visits, that it is decided whether the budget works for you or for someone else’s statistics. The first thing to fix is rarely the creative. It is whatever leaks quietly outside everyone’s field of view.

If you want to see which settings in your account work against you, book a free review and we will go through this checklist together. How these checks fit into ongoing campaign management is described on the Yandex Ads page.

Frequently asked questions

Sources

• Yandex Direct help, autotargeting: yandex.com/support/direct/en/keywords/autotargeting
• Yandex Direct help, bid adjustments: yandex.com/support/direct/en/campaigns/bid-adjustment
• Russian version of this article on belokrylovo.ru: Аудит аккаунта, часть 2: технический чеклист