Yandex Direct has spent five years moving toward the same place Google Ads landed with Performance Max — an automation-first world where the network picks placements, formats, and audiences from a single campaign. The autostrategies (Оплата за конверсии, Максимум конверсий, Целевая доля рекламных расходов) do work. But they only work when the campaign they run on is a coherent unit: one intent, one placement type, one goal, one audience shape.
The default UX pushes the opposite. New campaigns turn on Search and RSYA together, add auto-targeting, and let you attach three or four goals of wildly different values. Autostrategy then averages across all of it and spends the budget on whichever combination hits its target most cheaply, usually RSYA display clicks on the loosest goal. Reported CPA looks fine. Real qualified leads collapse.
The Three Splits That Matter
Every account we've rebuilt from a “one bucket” structure needed the same three splits before autostrategy started producing real performance. In order of impact:
| Split | What it does | Why autostrategy needs it |
|---|---|---|
| Search vs RSYA | Separate campaigns per placement network | Different CTRs, CVRs, and CPCs by 5–10×; averaging breaks bidding |
| Brand vs generic | Split branded queries into their own campaign | Brand CVR is 5–15×; leaving it in poisons the CPA target |
| Goal per campaign | One primary conversion goal per bidding unit | Autostrategy needs one signal, not five weighted averages |
Two of these are old rules Google Ads managers already know. The third — “one goal per campaign” — is more specific to Yandex, because the platform allows a much richer goal setup in Metrica and encourages throwing all of them at every campaign. That's the default trap.
Split 1 — Search vs RSYA Are Not the Same Thing
In 2020 you could still run a decent Search + РСЯ combo campaign because autostrategy was primitive and manual bids dominated. In 2026 the platform runs almost everyone on autostrategies, and autostrategy uses the campaign as its learning unit. If your one campaign contains 40% Search clicks at 120 ₽ CVR 6% and 60% RSYA clicks at 8 ₽ CVR 0.3%, the strategy will chase whichever cost-per-conversion looks cheaper on the day and starve the other.
The fix is boring: two campaigns per intent. One with only Search enabled, one with only Показы на поиске disabled and РСЯ enabled. Copy the keyword set once, then adapt each for its network — RSYA benefits from broader phrase-based groups and more aggressive image ads; Search wants tighter keyword sets and more responsive text ads.
Split 2 — Brand Traffic Is Not Marketing
Branded search converts at 8–15% for a decent brand. Generic converts at 0.8–3%. If both live in the same campaign under Оплата за конверсии, the autostrategy will happily allocate 80% of budget to branded terms because they hit the target CPA effortlessly. You'll show great numbers and acquire zero new customers.
Move brand into a dedicated Search campaign with its own goal, its own budget cap, and — critically — a hand-set target CPA below the generic one. Then move brand keywords out of the generic campaign with exact-match negatives (-!бренд). This one change usually reveals the true CPA of your generic marketing within a week.
The brand-as-negatives pattern
For every brand campaign you create, add the brand phrases as exact negatives in every other campaign that could match them. Yandex's phrase matching is aggressive; without exact negatives, generic campaigns will keep hoovering branded impressions. Use -!бренд (exclamation forces exact word form) or -[бренд] (square brackets for exact phrase) — the difference matters for Russian morphology.
Split 3 — One Goal Per Bidding Unit
Yandex Metrica lets you attach any number of goals to a campaign. Yandex Direct then lets you tell autostrategy: “target ≤ 500 ₽ per conversion” where “conversion” is a weighted mix of all those goals. In practice this averaging destroys learning: if you attach “form submit” (real lead, 500 ₽ target), “click to WhatsApp” (soft, worth maybe 100 ₽), and “5 pages viewed” (engagement, worth nothing), autostrategy will chase the cheapest of the three, which is almost always the noisiest signal.
| Setup | What autostrategy learns | Real business result |
|---|---|---|
| 3 goals, all at 500 ₽ target | Whichever goal is easiest to trigger at that price | 90% of “conversions” are the softest goal |
| 3 goals, weighted (500 / 100 / 20) | Weighted score, mostly driven by the largest goal | Better, but still noisy — autostrategy sees fewer clean signals |
| 1 goal per campaign, hard-set value | One binary conversion event per unit | Clean learning; CPA stable; real leads scale predictably |
The right pattern: split campaigns by what you actually want. One campaign targeting “form submit” on a real lead. Another targeting “click to call” if calls are a separate business flow. Never stack multiple heterogeneous goals in one bidding unit expecting the average to be meaningful.
Package Strategies (Пакетные стратегии) — Use With Care
Yandex introduced package strategies to let related campaigns share a learning pool. On paper it's the answer to fragmentation: split your campaigns for structural clarity but let them share signal. In practice it breaks the whole point of splitting if you package campaigns with different goals, different networks, or different intent.
Package strategies work when the campaigns inside are functionally the same — for example, 5 Search-only campaigns targeting the same lead goal across 5 regions, or 3 RSYA-only campaigns targeting the same purchase goal across 3 product categories. Don't package Search with RSYA. Don't package brand with generic. Don't package lead with purchase.
Negatives at Three Levels
Every mature Yandex account has negative keywords in three places, each with a different purpose. Miss any level and you leak budget in a way that's hard to spot in the standard reports.
Global junk that applies to every campaign: -бесплатно, -скачать, -своими руками, competitor brand names when you never want to bid on them, adult / piracy / mobile-game garbage from RSYA site names.
Intent boundaries between campaigns. Brand negatives in the generic campaign; generic negatives in the brand campaign; product-line negatives when you have multiple product-line campaigns. This is where the anti-cannibalisation work lives.
Tight intent shaping within a campaign. If you have купить велосипед and ремонт велосипеда as separate ad groups in one campaign, each needs the other's keywords as negatives to prevent Yandex's auction from picking the wrong ad-group creative for a matched query.
A separate check that gets missed often: RSYA site placements. Every 2–4 weeks, pull the Площадки report at account level, filter to zero-conversion spend, and add the top junk domains as negative platforms in the account-level exclusion list. In 2026 the RSYA network includes a lot of MFA and mobile-game inventory; a few hours a month of cleanup can shift 10–20% of RSYA budget onto real traffic.
What About Master Campaigns (Мастер кампаний)?
Master campaigns are Yandex's answer to Performance Max — one campaign type that automatically picks from Search, RSYA, and auto-targeting based on a landing URL and business goal. They're a real tool for very small accounts (single-store SMBs with under 30k ₽/month) where structural splitting is more overhead than benefit.
For anything larger, treat Master campaigns like PMax — useful as an addition after your structured Search and RSYA are working, never as a replacement. Give them their own goal, their own budget cap under 20% of total, and monitor Search query overlap with your structured Search campaigns (Yandex is worse than Google at preventing internal cannibalisation between Master and standard campaigns).
The Working Structure — What It Looks Like
Search — Generic — [product line 1]: Целевая доля рекламных расходов or Максимум конверсий, one primary goal.
Search — Generic — [product line 2]: Same, own budget cap.
RSYA — [product line 1]: Максимум конверсий, same primary goal as the matching Search campaign, broader phrase set.
RSYA — [product line 2]: Same.
Retargeting — cart abandoners: Separate RSYA campaign, audience segment from Metrica, own budget.
Master campaigns — Optional: Small budget under 20% total, product-page landing.
That's the pattern under nearly every well-run mid-market Yandex Direct account we've inherited. Below 100k ₽/month you can collapse product lines into single campaigns; above 500k ₽/month you can split further (by region, by device, by seasonality). The three fundamental splits stay the same.
Audit Checklist
-!бренд exact form).Why Structure Matters More Now Than in 2022
In 2022, Yandex's autostrategies still respected manual overrides and were forgiving of imperfect structure. In 2026, autostrategy is the whole product — Оплата за конверсии, Максимум конверсий, and Целевая доля рекламных расходов run 80%+ of active budget. The bidding engine is powerful, but it only reflects the signal you give it. Feed it a coherent unit (one placement, one intent, one goal) and it optimises hard. Feed it a mess and it optimises for the loudest sub-signal — which is almost never your best customer.
Structural refactoring is unglamorous compared to bid-strategy tweaks and creative tests, but on every rebuild we've done it's the single biggest lever. Split first, then let the autostrategies work. The same logic applies to any auto-bidded platform — see the Google Ads Smart Bidding piece on the equivalent primary/secondary lever, or how we approach these rebuilds inside Yandex Direct management.
