Yandex.Direct for an online store: product campaigns, smart banners and feed 2026

Product campaigns in Direct are not “regular Direct, only with pictures.” The logic is different: the algorithm reads your feed and builds ads for each product and each user. I’m not analyzing “how a feed works in theory,” but specifically: the structure, errors that break learning, and real CPOs from my niche projects.

In e-commerce, a product campaign works with a product feed and requires a separate data quality check. My materials on the ID Store indicate +25% sales in nine months, and on Eighteeth - a 20% decrease in CPC; Available sources do not link these results specifically to smart banners. Below is a setup diagram, not a reconstruction of an unconfirmed campaign.

Not “set up a product campaign and wait for conversions” - but “make a normal feed, give the algorithm 14 days and don’t touch the bids.” Without the first, the second does not work at all.

1. How product campaigns differ from regular campaigns: the logic of working through a feed

In a classic search campaign, a marketer manually sets queries, writes ad texts, and sets bids. The algorithm is bargained, but the decisions are up to the person. In a product campaign, everything is different. You provide a feed - a structured list of products with prices, pictures and attributes. Direct itself generates ads for each product and each user.

Smart banners go further: they show a specific user those products that he has already viewed on your site, or similar ones, based on his behavior in the Yandex ecosystem. This is not retargeting in the old sense of the word. The algorithm knows that the user looked at sneakers on the Market, read about running in Zen, and visited your “Sports” category several times. A banner with your sneakers will appear at the right moment.

The product gallery is a separate format; it lives in search results above organics. Looks like Google Shopping: product photo, price, store name. The user searches for “Nike sneakers size 42” and sees cards from the gallery immediately on the first screen, before organic results. The CTR for the gallery is on average 3-7% versus 2-4% for regular text ads for the same queries.

Read more about the structure of the account and betting strategies in my general guide to Yandex Direct 2026.

2. Requirements for a feed in 2026: required and desirable attributes

A feed is an XML or CSV file that Direct reads and turns into ads. Without a correct feed, the campaign will either not pass moderation, or it will pass, but will work at half capacity.

Required fields for YML feed:

• id — unique identifier of the product. Never change after launch - otherwise the conversion history for the product will be reset to zero
• name - name. The more specific the better: “Sneakers Nike Air Max 270 black 42” instead of “Sneakers for men”
• url — link to the product page with UTM tags
• price - current price. Direct checks the compliance of the feed and the page - discrepancies lead to ad rejection
• currencyId — RUB for rubles
• categoryId — a category from your category tree. Campaign segmentation is then built on it
• picture — URL of the image. Minimum 450×450 px, no watermarks, no text on the image

Desirable fields that really affect CTR and conversion: description (200-500 characters, non-advertising text - product description), vendor, model, typePrefix (type: “sneakers”, “jacket”), sales_notes (for example “Free shipping”), available (true/false - Direct hides unavailable products automatically).

The main mistake when preparing a feed is to fill out only the minimum and think that this is enough. The algorithm builds look-alikes based on converting users and compares attributes. The more attributes, the more precise the targeting.

3. Types of product formats: gallery, smart banners, dynamic ads

Three formats that work with the feed in Direct, and each has its own role in the funnel.

Product gallery — search format, first search results screen. Audience: users with clear commercial intent right now. CPO in e-com clothing - 250-500 ₽, in electronics - 700-1500 ₽. Requires good pictures and current prices: the user sees the price before clicking.

Smart Banners — YAN format, shown on partner platforms. They work in two modes: retargeting (site visitors, especially those who added to the cart) and look-alike (users similar to buyers). Retargeting CPO is 40-60% lower than a cold audience. If you don’t separate, you mix up statistics and lose the ability to manage bets.

Dynamic Ads — automatically generated text ads for search based on a feed or site. They close a long tail of requests, which is impossible to collect manually. Especially useful for catalogs of 1000 SKUs. The downside is less control over texts.

I use all three formats in conjunction: gallery - for hot demand, smart banners - for returning departed and similar ones, dynamics - for covering the tail. I divide the budget approximately 50/30/20. About how to build retargeting in 2026 - separate article.

4. Setting up smart banners: goals, bidding strategies and retargeting audiences

Smart banners require correctly configured goals in Metrica before launch. Without goals, the algorithm is optimized for clicks - and you already have clicks from the gallery and search, why overpay for them in YAN.

Minimum set of goals for smart banners:

• Viewing the product card (not the main one, but a specific SKU)
• Add to cart is the main micro-conversion for e-com
• Start placing an order
• Payment/order completion is the main conversion for the CPA strategy

Payment strategies in smart banners. At the start - “Maximum conversions within the budget” with the target conversion “add to cart”. This gives the algorithm enough data to learn even with a small sales volume. After 2-4 weeks and 30-50 conversions accumulated, switch to “Pay for conversions” with a target CPO for the main goal.

I divide retargeting audiences into three segments with different rates. Cart abandoners are the highest bidders; they are a hot audience with a conversion rate of 8-15%. Looked at the product, did not add it to the cart - average rates, conversion 2-5%. We visited the site, did not look at the products - minimum bids or even a separate campaign for awareness.

During the campaign training period, you should not change settings every day. Check expenses and technical errors, and make a conclusion based on CPO after accumulating a sufficient number of conversions; The period depends on the volume of traffic.

5. How to segment your feed for different campaigns: categories, price, margin

Mixing an iPhone for 120K and a case for 300 rubles into one campaign is a classic mistake. The algorithm optimizes for conversions, not margins. Cases are sold 50 times more than iPhones - the algorithm will focus on cases and will not show expensive products.

The minimum feed segmentation that I use on all e-com projects:

• Top sales are 15-20% of SKUs, which give 70-80% of turnover. Separate campaign with maximum budget and aggressive bids
• High-margin products are not always top sellers. A separate campaign with a CPA strategy for the desired CPO, taking into account the margin
• New items - you need to quickly collect your conversion history, I set higher rates for the first 2-3 weeks
• Remains and sales are a separate campaign with the message “-30%, only until the end of the month”
• Basic catalog - the rest in one campaign with a minimum budget

For segmentation, I use the feed attributes: categoryId, price (I set a filter by price range), and the custom attribute custom_label if the CMS allows it. Direct supports filtering by all standard YML attributes.

At Eighteeth Russia we had two segments: consumables (low receipts, high frequency) and equipment (high receipts, rare conversions). Different campaigns, different strategies, different target CPOs. It was this division that gave +25% of leads with a 20% reduction in CPC.

6. Common errors in the feed and why the campaign does not train

The campaign “doesn’t work” in 60% of cases is due to errors in the feed, not in the account settings. I check the feed first, even if the client is sure that “everything is fine.”

Duplicates id. If two products have the same id, the algorithm doesn’t know which conversion history to take. Check the uniqueness of the id every time you update your feed. This especially often breaks down when changing the CMS or moving a catalog.

Blank or incorrect pictures. Direct accepts feeds with incorrect image URLs, but does not display such ads. In the report you see “the ad is active”, there are zero real impressions. Check images separately: go through 20-30 random URLs from the feed.

The price in the feed is different from the price on the page. Direct periodically checks compliance. A discrepancy of more than 5-10% leads to ad rejection. If you have promotional prices, update the feed immediately, not every other day.

Feed updates too frequently. Updating the feed during active learning partially resets the accumulated statistics on changed products. If you change prices on 30% of positions at the same time, expect a drawdown in the first 3-5 days.

All products in one category. If the categoryId is the same for all SKUs, segmentation by category does not work. Direct sees a “store without structure” and selects audiences worse. Even a basic tree of 5-10 categories is better than one.

No products available=false if out of stock. The algorithm continues to show ads with unavailable products. The user clicks and gets to the “out of stock” page - the CTR is good, the conversion is zero. The algorithm reads this as bad traffic and downgrades the quality of the campaign.

About the logic of auto-targeting and how Campaign Wizard uses site data to select audiences - discussed in detail separately.

7. Product campaign analytics: what metrics to look at in Metrica and Direct

Reports in Direct and in Metrica will differ - this is normal. Direct counts conversions based on its attribution (last click by default), Metrica may count differently depending on your settings. The source of truth is Metrica.

What I see in the Direct account for product campaigns:

• CPO by campaign and feed segment - once a week after the training period
• ROAS (income/expense) is the main metric for e-com. The rate depends on the margin: with a margin of 30%, the minimum ROAS is 3.5-4x to break even
• Percentage of impressions by type: search vs YAN in smart banners. If YAN is more than 80%, check if the audiences are mixed
• CTR for individual products in the gallery - if the CTR for a product is below 1%, check the image and price

In Metrica I look:

• E-com report: transactions, income, conversion from session to purchase - by sources and campaigns
• Audit report on the segment “came from smart banners” - viewing depth, failure rate. If the failure rate is above 70%, the problem is in the landing or in compliance of the banner and the product
• Cohort analysis: how buyers from different campaigns behave after 30-60 days - LTV, repeat purchases

How to calculate ROAS, CPL and normal values for niches - in a separate article about ROAS and CPL formulas 2026.

8. Real CPO numbers by niche: what is considered the norm in 2026

Data from my projects and communication with colleagues on 15-20 projects in different niches, 2025-2026. Product campaigns (gallery + smart banners) vs classic search campaigns:

On average, product campaigns produce a CPO that is 20-35% lower than classic search campaigns with comparable traffic volumes. But it works with the right feed and enough training period. Bad feed = results are worse than search campaigns, because the algorithm has nothing to base targeting on.

A separate story is retargeting through smart banners. In the “cart abandonment” segment, I saw a CPO of 120-200 rubles in clothing and 400-600 rubles in electronics for the same niches as in the table above. This is a comparison between a cold audience and a hot one - but it shows why retargeting needs to be considered separately.

How to build a full-fledged retargeting strategy for an online store - in the article about retargeting 2026. It’s also about look-alike and working with YA audiences.

Bottom line

Product campaigns are not an addition to the main performance mix, but its core for an online store. Smart banners bring back those who left. The gallery meets hot demand. Dynamics cover the tail. All this works through the feed - and if the feed is bad, not a single account setting will save you.

Three things that break most product campaigns: duplicate ids in the feed, edits in the account in the first 14 days, and mixing expensive and cheap products in one campaign. Eliminating them means getting results.

Related articles: complete guide to Yandex Direct 2026, Campaign Wizard: when it works, when it fails, ROAS and CPL formulas, retargeting 2026.

If you want to analyze the structure of your feed or see why product campaigns do not reach the planned CPO, write to Telegram or through form. Starting consultation - 0 ₽.