Yandex.Direct 2026 Campaign Master: when it works, when it doesn’t, and how to prepare it correctly

The Campaign Master is not Direct with wheels. It is a separate tool with different optimization logic, different data requirements, and different points of failure. In my practice, a third of clients who come with “MK does not work” simply did not allow it to be trained properly.

I have been leading performance projects in Direct since 2016, of which the last three years have been mainly in the new neural network paradigm. Here I’ll look at how MK actually works, under what conditions it outperforms manual campaigns, and under what conditions it drains the budget. With a CPL table from real offices.

“Campaign Wizard is not working” - in 80% of cases it means “I didn’t let the algorithm learn properly” or “I have 3 conversions per week.”

1. What is Campaign Master and how does it differ from regular Direct?

Classic Direct is manual control: you set semantics, write ads, set bids for phrases. The algorithm negotiates for you, but you make the decisions. MK works differently: you give website, goal and budget, the algorithm itself decides everything else - queries, formats, audiences, bids.

This is not a simplification - it is a different model. MK reads the content of your website (H1, descriptions, prices, structure), looks at users who have already converted, and builds similar audiences. Without a sufficient number of conversions, there is nothing to build from.

The second key difference: MK works in all formats simultaneously (search + YAN + media), optimizing between them automatically. Manual campaigns - search separately, YAN separately, retargeting separately. It's more control, but also more manual work.

Read more about how the entire Direct office will be structured in 2026 in my step-by-step guide to Yandex Direct.

2. How MK learns - signals, conversions, time horizon

Learning MC is a statistical process. The algorithm needs examples: “here are the users who bought”, “here are those who left.” To understand the difference, a minimum sample size is needed.

Minimum: 10 conversions per week. With this volume, MK begins training. Optimally - 20-30 per week. If there are less than 10 conversions, instead of the main goal (purchase, application), use more frequent micro-conversions: adding to cart, viewing the page for more than 60 seconds, scrolling up to 75%.

Time horizon of training - 7–14 days. At this time, the CPL will be higher than normal - the algorithm “tests” different audiences and requests. It's not a bug, it's a process. On days 10–14, the strategy stabilizes. This is why touching a campaign in the first two weeks is a major mistake.

Signals that MK uses for training:

• Conversions from Y.Metrica (events, goals)
• Yandex behavioral signals: search history, Market, Maps, Zen
• Analogue of look-alike for converted users
• Landing page content (semantics, structure, prices)
• Characteristics of devices, time of day, geolocation

Correctly set goals are a necessary condition. Without them, MK is optimized for “clicks,” which is useless for most businesses. How to set goals - in my article about setting goals in Yandex Metrica 2026.

3. When MK wins over manual campaigns

Three conditions under which MK is consistently better than manual campaigns:

E-com with a wide range of products. An online store with 500+ SKUs - a manual campaign with separate ad groups for each product eats up a week of work. MK reads the entire catalog and determines which products to promote more actively, based on real conversions.

Mass market niches with a wide audience. Food delivery, fitness clubs, cosmetics, clothing - a large audience with different interests. MK finds converting segments better than manual targeting, especially if there is a sufficient history of conversions.

Launch without preliminary semantics. There is no time to assemble the kernel - the MK starts in 20 minutes. It works like a scout: for the first 2–3 weeks it shows which queries actually convert, then based on them manual campaigns are built for high-margin phrases.

An important factor is the visual component. MK actively uses pictures in YAN. If you have a beautiful product, high-quality photos - MK is revealed 100%. In the niches “complex service without visuals” (legal services, B2B consulting), the advantage of MK is reduced.

4. When MK loses - B2B, narrow semantics, long cycle

MK is not a universal tool. Four situations where he is consistently inferior:

A narrow B2B niche with rare conversions. Industrial equipment, enterprise software, specialized services - 3–7 applications per week. MK will never collect the required amount of data for training. There are only manual campaigns with targeted semantics here.

Transactions with a long cycle. If it takes 2–4 weeks from the first click to the application, MK does not see conversions in the training window and degrades. The solution is to transmit the signal as early as possible along the funnel or connect offline conversions via CRM.

Brand queries. Protecting your own brand from being intercepted by competitors is a manual campaign with a minimum bid. MK will mix branded traffic with general traffic and you will overpay.

Narrow commercial requests with high CPL. “Implantation All-on-4 Moscow price” is one request, targeted, expensive. The manual campaign hits it right. MK will blur into “dentistry Moscow” and “inexpensive implants” - and the CPL will fly away by 2 times.

5. MK vs manual niche campaigns - table with real CPL

Data from my projects and those of my colleagues, 2024–2025. Median values ​​of CPL, MK and manual campaigns under the same conditions.

To calculate planned CPL for your project, read about ROAS and CPL formulas 2026. There is also a table of normal values ​​for niches.

6. How to read MK analytics - metrics and training period

MK shows statistics differently than manual campaigns. What to look for in reports:

“Training period” in campaign status. This is a Direct information block that shows: “the algorithm is accumulating data.” As long as it exists, it’s normal that the CPL is high. Disappears approximately on the 7th–14th day with a sufficient number of conversions.

CPL in the first 7 days is not an indicator. Watch it from the 10th day. Before this, it’s just a waste of the budget for exploration. If you stop MK after 3 days because “CPL is 3 times higher than normal,” you are killing the campaign at the most uninformative time.

What metrics are important in the MK report:

• Conversions (not clicks) are the main indicator after training
• Share of impressions in search vs YAN - if YAN is more than 70%, check the quality of the audiences
• The percentage of failures for MK traffic in Metrica should be comparable to other campaigns
• Average viewing depth is a sign of traffic quality

Important: the Direct report and the Metrica report may differ. The source of truth is Metrica, because Direct counts “conversions” based on its attribution. This is discussed in detail in the article about setting up Y.Metrics and goals.

7. Common mistakes - and how to avoid them

Over three years of working with MK in different niches, I have collected a consistent list of errors that kill the result:

"Campaign pause = tutorial reset." The most common. We paused the campaign for the weekend and reset the accumulated data. MK does not “remember” training through pauses in the same way as manual campaigns. If you need to reduce expenses on the weekend, reduce your daily budget rather than pause it.

Changing strategy in the middle of training. We launched “maximum conversions”, a week later we switched to “CPA” - the training began again. Wait 14 days on the first strategy, then switch.

Budget too small. An MK with a budget of 500 ₽/day in a niche with a CPL of 2,000 ₽ cannot physically train - it has no data. Minimum budget = target CPL × 10–15 per month (for example, CPL 1,000 ₽ → budget from 30,000 ₽/month).

Inconsistency between the landing page and the offer in the ad. MK generates ads automatically from your website. If the site says “promotion -30%” and the promotion is over, the algorithm continues to pull this text and the CTR drops. Update your website and creatives synchronously.

Ignoring micro-conversions with low traffic volumes. If you make 5 applications a week, set up goals higher up the funnel (clicked on the phone, spent more than 90 seconds on the page), send them to MK as the main conversion. Then, when the volume increases, switch to the final goal.

If MK “doesn’t work” - here’s mine 6-step checklist for diagnosing performance campaigns.

8. Practical checklist before launching MK

Run through this list before clicking “run”. This is not a theory - this is what I do on every new project.

1. Goals in Metrica are set up and working. Go to Metrica → “Goals” → “Check”. Click the button/fill out the form on the website - the goal should work within 5 minutes. If not, launching MK is useless.
2. The metric is linked to Direct. “Campaign settings” → “Metrics” → the counter is connected, the goals are visible in the list. If not linked, the MK does not receive a conversion signal.
3. Budget not lower than CPL × 15 per month. It is considered simple: if the target CPL is 1,500 ₽, the minimum budget of the MK is 22,500 ₽/month. Otherwise, the training period stretches to infinity.
4. Minimum 5 pictures for YAN. At least one is a product on a white background, one is a product in the context of use, one is a team/office. Stock neutral images = high bounce rate.
5. UTM tags are configured. MK generates its own UTMs, but often not according to your scheme. Configure manually via “Additional UTM parameters” - otherwise Metrica will not understand where the traffic is coming from. More details in the article about UTM and Metrica.
6. A scenario of actions for the 14th day has been written. Which CPL is “good” for you, at which “bad” one do you restart, what exactly will you change. Without a plan, in 2 weeks you will make decisions based on emotions.
7. The landing page was checked against the CRO checklist. MK can bring good traffic, but a bad landing page will kill conversion. Diagnostic checklist includes a quick CRO check in 15 minutes.
8. A reminder has been set for the 14th day. Literally - add it to your calendar. The first real analysis of the MK results is two weeks after the launch, not earlier.

About A/B testing of hypotheses after the first two weeks of MK work - in my article about A/B tests in marketing 2026.

If you are setting up MK on a new project and want to check the logic, write to Telegram @dipustovalov. I analyze specific offices, showing what exactly kills the result. Or come to contact page — there is a discovery call format for 30 minutes.

Related materials: CPC benchmarks in Yandex Direct, CPA calculator, ROAS calculator.