Last-click attribution and the touches that come before a lead

Someone clicks an ad, fills in the form a minute later, and the report hands all the credit to that click. The picture looks honest: here is the source, here is the inquiry, here is its price. The trouble is that the inquiry did not ripen at the moment of the click. Before it came a search, somebody else’s post in a feed, a banner, a conversation with a colleague, a billboard on the way to work. The last touch took credit for work that several earlier ones had prepared.

Last-click attribution is convenient for reporting, and it almost always misleads you about what actually brought the person in. Below is where exactly it goes wrong, what never reaches your analytics at all, and how to judge the contribution of each touch by the data rather than by who happened to be last in line.

For a company selling into Russia from abroad, the problem has a local shape. The paid search journey runs through Yandex, not Google. Measurement usually lives in Yandex Metrica. And a large part of the conversation with a customer happens in channels that a Western analytics stack was never set up to record.

Why the last click misleads

Start with the fact that “source of the inquiry” is already a simplification inside a single channel. Take search. One and the same query can trigger impressions across a dozen different keywords, including keywords from different ad groups. The priority that the interface promises to exact matches works in practice for only a small share of impressions, so you cannot treat it as a guarantee. When a report assigns an inquiry to one particular keyword, that row hides a whole bundle of keywords that competed for the same impression.

You can see the scale of this overlap if you build a pivot table on a data model rather than on a flat export. Put the search query in the rows. In the values, count the distinct keyword IDs and distinct ad group IDs per query. Not the plain row count: the number of different elements. For every query you then see how many keywords actually fought for the impression and how each of them performed on its own.

The conclusion is simple. If attribution drifts inside a single search channel, gluing the whole customer journey to one source is even less defensible. The last click is not the cause of an inquiry. It is the door the person walked through at the end of a long road.

What changes when the channel is Yandex Direct

Teams coming from Google Ads tend to assume that keyword-level reporting in Yandex Direct is as tidy as what they are used to. It is safer to assume the opposite. Autotargeting and broad matching mean that the query a person typed and the keyword that got the credit can be further apart than the report suggests. Before you shift budget between keywords on the strength of last-click numbers, check the search query report and the overlap described above. The keyword you are about to cut may be the one doing the work that another row is credited with.

The blind spots of analytics

Tags, pixels, cookies and link parameters are good at catching channels where a visit is recorded technically. There is a whole class of touches they cannot see in principle: television, radio, outdoor and digital outdoor advertising, word of mouth, offline events. For these you might notice a spike in inquiries while a campaign is running, but you have nothing to tie a specific customer to a specific channel.

In Russia this blind zone is wider than a Western team expects. A great deal of recommendation happens in messengers and Telegram channels, where a link is forwarded, retyped or described in words, and whatever parameters it carried are gone by the time the person reaches your site. The visit arrives as direct traffic or as a branded search, and the report credits the brand query for demand that a channel post created.

What works here is a method many people consider too simple to take seriously: ask the customer directly at the first contact. “Where did you hear about us?” is asked in the conversation with the sales team, in the inquiry form, in the chat. The answer goes into a separate CRM field and is collected alongside the digital sources. This is not precise data. It is an indicative estimate: memory fails, and people name the most memorable channel rather than the decisive one. But where the alternative is a complete analytical blind spot, even an approximate signal is worth more than nothing.

One condition makes the method work or makes it useless: discipline. Everyone who takes inquiries has to ask the question the same way. If half the managers ask and half forget, the sample skews, and you get a picture of which employees fill in the record more carefully rather than a picture of your channels.

For a foreign company there is a practical detail. The question has to be asked in Russian, by whoever actually answers the phone or the messenger in Russia, including a partner or distributor if they handle first contact. And the field has to reach your CRM rather than stay in someone’s notebook. Where that CRM stores data about Russian customers is a question for your lawyers, since Russian law has requirements on where personal data of Russian citizens is kept.

When touches erase themselves

There is also a reverse process. Some touches disappear from analytics because the tracking technology itself keeps changing, not because you count badly. Browsers limit cross-site tracking based on cookies and move towards aggregated interest categories: a site gets access to one generalised topic rather than to the detailed history of specific pages a person visited.

For paid traffic this hits two particular tools. Search retargeting, which followed people based on their earlier queries, loses the material it builds audiences from. Dynamic retargeting, which showed a person the exact products they had viewed, loses the precision of the data it depends on. Both tools as a class risk weakening noticeably, and for an online store they are the cheapest and best-converting touches.

The practical meaning is that the customer journey becomes less observable over time, not more. The more you depend on a single tool that relies on cookies, the more exposed your whole attribution model is to the next browser update. Safari and Firefox already restrict this kind of tracking by default, and the browser mix of your Russian audience is not the one you see at home. Build the model on the assumption that part of the path will stay dark.

The sensible insurance is to collect your own customer data and to rely on context and direct contact rather than on pixel-based pursuit whose lifespan does not belong to you. In Russia this also means treating Metrica’s own reports as one view among several, and reconciling them against the CRM rather than trusting either alone.

How to read contribution from the data

When there are many touches and some are visible while others are not, the temptation is strong to collapse everything into one average and stop worrying. An account-wide average misleads as badly as the last click. It spreads the result in an even layer where the real distribution is sharply uneven.

Here is how it looks on a performance cut. Build a pivot table whose rows are consecutive ranges of a metric, from low values to high. In the values, show the share of impressions, spend and inquiries in each range. The picture is almost always asymmetric: a small share of impressions and spend brings a disproportionately large share of inquiries. Roughly speaking, a little over half of the inquiries come from where less than half the budget was spent and less than a fifth of the volume was shown. That segment is the de facto norm for the niche, and the averaged figure is not.

The same principle carries over to touches. Stop asking “which source produced the inquiry”. Ask “how is contribution distributed across the whole chain, and where is the result concentrated”. A cut instead of an average, a distribution instead of a single point. Then you can see which touches actually move a person towards an inquiry and which were simply last in the queue and took credit for someone else’s work.

Metrica helps here in one specific way: it lets you switch the attribution model in its reports and compare the same period under different models. The gap between a first-touch view and a last-click view is the most informative number you will get from it. A channel that looks strong early and weak late is preparing demand. Cutting it because the last-click report says so is how companies switch off the part of the funnel that feeds everything else.

What to do about attribution

A short route to walk before you trust a report:

1. Do not treat the last click as the cause. Use a pivot on a data model to check how many keywords and ad groups one query triggers, and you will see how conditional the link between an inquiry and a single row is.
2. Close the blind spots with a question. Add “where did you hear about us” to the form and to the sales script, create a CRM field for it, and make sure everyone asks it every time, in Russian, including partners who take first contact.
3. Reduce your dependence on cookies. Collect your own customer data and do not build the whole model on retargeting whose accuracy is decided by browser vendors.
4. Look at distributions, not averages. Build cuts by ranges, find where the result is concentrated, and do not average something uneven into one misleading figure.
5. Compare attribution models in Metrica. Read the difference between them as information about which channels work early and which work late.

Behind the words “this channel doesn’t work” there is often short-sighted attribution rather than a failing channel: the inquiry was credited to whoever opened the door instead of whoever prepared it. As long as you pay by the last click, you are switching off exactly the touches that do the main work earlier and more quietly.

You can work out which touches actually move inquiries in your case, and which only claim someone else’s result, in a free review. How to set up advertising so that contribution is read from the data rather than from the last click is covered on the Yandex Ads page.

Frequently asked questions

Sources

• Yandex Metrica help, reports and attribution settings: yandex.com/support/metrica
• WebKit, tracking prevention in Safari: webkit.org/tracking-prevention
• Russian version of this article on belokrylovo.ru: Атрибуция и путь клиента: касания до заявки