Pay per conversion in Yandex Direct: who it actually suits

Paying for results instead of clicks sounds like a fair deal. On search it runs into the same ceiling every time, and the people selling the model rarely mention it.

Here is the mechanism: why cost per lead on search is hard to steer at all, where pay per conversion is the only direct lever, and why the volume for it has to be found somewhere other than search.

Why a bid does not steer cost per lead on search

Take an ad group with similar keywords and one shared bid. It feels like you are controlling price. In practice each phrase inside the group behaves on its own.

Each has its own threshold price at which it enters the auction at all. Its own range of actual click prices. And, more importantly, its own range where the bulk of its conversions sit. Those conversions are by no means always at the bottom of the range; sometimes it is the opposite.

Now you move the shared bid. Every phrase under that bid enters or leaves the auction at once, not the single one you meant to adjust. Standard reports will not even show you how many unique phrases started or stopped receiving impressions: there is no such metric in the interface.

So the change in cost per conversion after a bid edit cannot honestly be attributed to the click price. It is mixed with a change in which phrases are in the auction at all.

The same applies to third-party bidding tools aiming at a target return. If the actual click price at the level of individual queries jumps almost at random, no algorithm guarantees control over the final cost per lead through the bid. Too many factors behind what you are charged sit outside your control.

The one direct lever, and what it costs

There is exactly one model where cost per lead is set directly and predictably: paying on the fact of a conversion.

You pay not for the click but for a completed target action at a price agreed in advance. No attribution confusion: the price of the result is fixed.

The payment for that predictability is volume. On search, the traffic you manage to get under pay per conversion is usually very small. The system releases impressions cautiously, because it is taking the risk, and it opens the tap only where it is confident of a conversion.

So on pure search, pay per result lives in a narrow corner: tasks where volume is not critical, or a combination with other sources that make up the numbers.

Who it suits is already visible. If you need a predictable ceiling on cost per lead and can live with a modest flow, the model is yours. If the task is to build volume, search with pay per conversion will not solve it alone.

Why exact matching is melting away

There is another reason old manual control on search weakens year by year.

The logic of precise structures, a separate ad group per keyword with a headline that repeats the phrase almost word for word, rested on an assumption: maximum relevance means cheaper clicks and higher click-through. The assumption holds while most impressions come from repeating, predictable formulations.

Observation across live search accounts says the opposite. The share of spend and clicks on queries that appeared exactly once and will never repeat grows steadily, even when negative keywords are applied exhaustively at every level. A few years ago such one-off queries gave around a third of traffic. Now it is more than half of spend and clicks.

For that half, a headline cannot mirror the query in principle: the query is unique and unknown in advance.

The meaning for our topic is direct. The higher the share of unique one-off queries in your niche, the less benefit there is in polishing keyword and headline by hand, and the more sense there is in leaning on broad matching, autotargeting and automated bidding with negative filtering afterwards. And since control moves to the algorithm anyway, the payment model question becomes sharp: pay for clicks blind, or pay for the result.

For a foreign advertiser this matters twice over, because the one-off queries are in a language your team does not read. You will not spot the junk by eye in a report. The filter has to be systematic.

Where the volume lives: networks and narrowed audiences

Since pay per result gives little on search, volume is found in the networks. But a network without a bridle means an enormous number of impressions a week and a mountain of irrelevant traffic. A placement whitelist alone often does not cure it.

The working technique is not to target the whole network but to build narrow audiences by intersection.

Custom audiences built from keywords as search queries, keywords as interests, and competitor domains are intersected step by step with an AND operator into combined audiences. You keep intersecting until the forecast weekly reach shrinks to something small: adding another audience beyond that point stops producing meaningful impressions. The threshold differs by niche. The point is to appear not to everyone but to a dense core of people resembling your buyers.

A detail about formats. In some automated campaign types combined audiences are unavailable and only custom ones work. To avoid rebuilding everything in a hurry, prepare a set of several dozen custom audiences by different criteria in advance and start from the narrowest and most precise, widening as needed. And a warning that bears repeating: do not launch a fully automated campaign type on an account without a steady flow of conversions, or the algorithm has nothing to learn from.

Remarketing as fuel for pay per result

Pay per conversion likes a warm audience: the result there is more predictable, so the system releases volume more willingly. Accumulated remarketing is therefore an asset for this model, not a secondary option.

A subtlety appears when campaigns move between accounts. Remarketing can be raised in several ways, and the most economical when migrating is to share the already accumulated audience from the old account into the new one through a manager account. Then you do not collect it again from zero, and it keeps filling up.

The technique has a flip side: if the old account is detached from the manager account, the new one loses access to that audience and all the remarketing built on it stops. Before building a pay-per-result flow on such an audience, make sure the account linkage is under your control. For a foreign company working through a local partner or agency, that is not a technicality. It is the question of who owns the asset.

Who pay per conversion suits

• You need a predictable ceiling on cost per lead and can accept modest volume on search. Suits you.
• The task is to build volume on search right now. Does not suit: the volume will be small.
• You have accumulated conversions and warm audiences the system can draw a predictable result from. Suits you; look for volume in networks and remarketing.
• The account is new, with no conversion history. Too early: neither pay per result nor fully automated campaigns have anything to learn from.
• A niche with narrow, stable demand and a low share of one-off queries. Manual control is still alive; pay per result is optional.

Pay per conversion is not a magic button labelled “I only pay for sales and sleep well”. It is a trade: you get a predictable price for the result and give up volume for it. Whether that trade is good for you is decided by your niche and your goals, not by the marketing of the model.

If it is unclear whether your search will deliver volume at a fixed cost per lead, or whether the money is better spent another way, that is a calculation I run on your data in the free review.

Frequently asked questions

Sources

• Yandex Direct help, pay per conversion and CPA strategies: yandex.com/support/direct/en/strategies
• Yandex Audience, custom and combined segments: yandex.com/support2/audience/en
• Russian version of this article: Оплата за конверсии: когда и кому подходит