CPA calculator 2026 - cost of target action

CPA formula - what and where to substitute

CPA stands for Cost Per Action - the cost of one target actions in advertising. The formula seems elementary: I divide the advertising budget for the number of actions and get the cost of one action.

CPA = Channel budget ÷ Number of actions

Example:
80,000 ₽ ÷ 55 applications = 1,455 ₽ per application

Then things get interesting. The word “action” in this formula is the most slippery part. In Yandex Direct, a “conversion” is any target event in Metrica: click on a button, scroll to the end of the page, send forms. In VK Ads, a “lead” is an entry in the lead form of the social network itself, which doesn't always reach my CRM. In Metrics, “achieving a goal” is This is actually a third entity that considers everything that I myself have configured.

The situation on my projects is normal: there are 120 “applications” in Yandex Metrica, in Yandex Direct there are 95 “conversions”, in CRM there are 71 real leads from this channel. A discrepancy of 40% is not a bug, it is the norm. So when I count CPA for a report or for deciding whether to continue the campaign or not, I take figure from CRM, and not from the advertising account. Cabinet makes mistakes in his favor.

If you want more details, I have it CPA term in glossary with analysis of variations (CPL, CPS, eCPA, target CPA) and how they are between connected.

CPA benchmarks for Russian niches 2026

The numbers in the table are my working niche range for 2026: a guideline, with which I compare the campaign to understand whether its result is normal or not. This is not an industry standard or the result of a market survey, and a working comparison tool for Russian-language campaigns in Yandex Direct, VK Ads and MyTarget. Auction in the Russian Federation over the past two years has risen noticeably in price, and the range takes this into account.

The main thing is that these numbers mean: “one CPA for all” is a myth. If compare the median CPA for FMCG (550 ₽) and for B2B SaaS (9,000 ₽) - this is not “FMCG is 16 times more effective.” These are just different niches with different economy of one sale: in FMCG I pay 550 ₽ for the first purchase with by check 800 ₽, and in SaaS - 9,000 ₽ for a trial, which then pays off in year through subscription 3,000 ₽/month. It's pointless to compare head-on.

What is considered a "good" CPA - and why it's a trick question

When a client asks “I have a CPA of 1,200 ₽ - is that good or bad?” I don't answer "good" or "bad". I ask a counter question: what you LTV and what is the margin? Without these two numbers, the answer does not exist.

A simple example from my practice. Two projects, both educational.

• Project A. CPA 1,500 ₽ for registration for free lesson. Conversion to a paid course is 8%. The average bill is 35,000 rubles. CAC= 1,500 / 0.08 = 18,750 ₽. LTV (with repeat purchases) - 65,000 ₽. A ratio of 3.5:1 is an excellent CPA, you can pour more.
• Project B. CPA 800 ₽ per registration. Sounds better than A. But the conversion to sale is 2.5%, the receipt is 12,000 ₽. CAC = 800 / 0.025 = 32,000 ₽. LTV - 14,000 ₽. Ratio 0.4:1 - project loses money on every client, and the low CPA masks this.

Therefore, when I talk about a “good” CPA, I keep two things in mind: calculator nearby: LTV/CAC calculator and unit-economics of the project. First I look at what maximum CPA I can afford (this is derivative of LTV and margin). And only then I compare it with the actual CPA from the calculator above.

Not “CPA should be low”, but “CPA should be below unit-margin”. A low CPA in an unprofitable niche is a first class ticket to "Titanic".

5 mistakes when calculating CPA

I see these errors in approximately every second audit. They all lead to one: the actual CPA is 30-80% higher than the “drawn” one, and the team doesn't understand why the unit economics don't add up for months.

1. They count based on all clicks, not just conversions from this one channel. The user came from Yandex Direct, left, returned through direct access and left a request. In Metrica it is attributed to “direct”, it is not in Direct. CPA by Direct artificially is overestimated, according to “direct” – it is underestimated to zero. Solution: take attribution from the advertising account (there is a “last click” model channel"), and not from Metrica (the “last source” there).

2. Returns and cancellations are not deducted. In e-com 8-15% applications fall off at the payment stage, another 5-8% of returns. If in CPA considers “placed orders” and not “paid and delivered” real CPA is 15-25% higher than reported. In B2B the analogue is “lead” qualified" vs "lead reached the meeting."

3. They measure CPA in Metrica and spend the budget in Yandex Direct and VK Ads. These are three different pinpoints with three different attribution models. Metrica loses events on iOS and Safari, VK Ads is underperforming in offline conversions; Direct is overestimating its share. A real CPA is (sum of spend from all accounts) ÷ (sum of leads from CRM). I don't believe in any other formula.

4. They ignore “shadow” attribution. The user clicked on advertising, did not leave a request, three days later wrote in Telegram “I I saw your advertisement - tell me more.” If this lead is in CRM marked as “Telegram organic”, CPA for advertising will be displayed above, what he is. Solution: ask “how did you know” at the first contact and add it to the lead card.

5. Compare the CPA with a competitor in another niche. I I regularly hear: “my neighbor’s CPA is 600 ₽, and I have 2,200, which is not right? And your neighbor goes to FMCG with a check of 1,200 ₽, and you go to MedTech with a check 45,000 ₽. You need to compare with the median for your niche (table above), and not with an arbitrary project from someone else’s case.

CPA vs CPL vs CAC - what's the difference in 30 seconds

These three abbreviations are most often confused, despite the fact that the difference between they are actually three different stages of the funnel.

To be really rude: CPL < CPA < C.A.C. Lead becomes action action becomes a paying customer - at every step of the funnel loses people, so the cost rises. In my practice, typical ratios: CPA = CPL × 2-3, CAC = CPA × 2-5 depending on the niche. In B2B SaaS the difference is even more dramatic: CPL 800 ₽, CPA 4,000 ₽, CAC 25,000 ₽.

How to reduce CPA by 30% - three levers

When the CPA goes beyond the benchmark, the first thing I do is not go into the betting. Raise/lower bet is the latest lever. First I I go through three levels of the funnel from top to bottom: creatives, landing page, audience.

Lever 1: landing page - conversion from 1.5% to 2.5% = −40% CPA

The most underrated lever. If the landing page CR increases from 1.5% to 2.5%, the number of applications with the same budget and the same clicks will increase by 67%. And the CPA will drop by 40%. This is mathematics, not magic.

What I do first on the landing page: I reduce the form to 2-3 fields, put the offer on the first screen, add social proof above fold, checking mobile speed (if LCP >3 sec - half of the mobile traffic falls off). More details in my article “Landing Page 2026: What Works, What Doesn’t”.

Lever 2: creatives - 8x difference in CTR

In my campaigns, the difference between the best and worst creative by CTR consistently 5-8x. Not by 20%, not twice, but eight times. Worst: CTR 0.4%, auction bid is 25 ₽ per click. Best: CTR 3.2%, same rate, but CPC drops to 4-6 ₽ due to the high quality score. CPA along with him.

What works in 2026 in the Russian Federation: video 9-15 seconds with the problem in the first frame; static with a specific figure and price (and not “the best in its class"); UGC format with a real client, not a studio booklet. I looked into it in more detail in “Advertising creatives 2026: what went well with my campaigns”.

Lever 3: audience - LAL reduces CPA by 1.5-2x

LAL (look-alike, similar audience) to a quality base is usually the fastest way to cut CPA. Condition: base look-alike should be “hot” - paying customers, not everyone subscribers. For 200-300 paying clients, the Yandex algorithm or VK Ads builds reasonable LAL, CPA on such an audience on my projects are on average 35-50% lower than the broad target.

What exactly not works: LAL for “everyone who left a request” - there is too much garbage, and the algorithm learns from people who don't buy. And LAL for a small base (less than 100 people) - there There are not enough statistics, and the audience is almost the same as the general public.

Frequently asked questions about CPA