Marketing Channel Attribution 2026: From Last Click to MMM

Every time a client says “context works, SEO doesn’t work,” I ask what their attribution model is. The answer is almost always the same. Last Click. This means that he has been cutting the channels that feed his funnel for three years.

I have been working with analytics and attribution models since 2017 - first on the agency side, and for the last four years on the client side. In my current projects, the median for associated conversions in SEO is from 28 to 41% of all sales. Last Click doesn't see any of them. The numbers in the article come from practice and conversations with about twenty colleagues who work closely with attribution.

The main misunderstanding about attribution: they think that it is about an “honest” calculation of the contribution of channels. It’s more correct to think that this is about budget solutions. Incorrect attribution = incorrect decisions about where money is going next quarter.

1. What is attribution and why is it important?

Attribution is the rule by which credit for a conversion is distributed across channels. A person saw your post in TG, a week later came to the article from a search, then clicked on retargeting in VK and bought. Three touches, one conversion. Who should record the sale?

In Last Click - VK Ads gets 100%. SEO and TG - 0. The marketer looks at the report, sees “VK works, SEO doesn’t work” and cuts the SEO budget. After three quarters, organics dry up, retargeting loses its audience, and conversions drop for no apparent reason. This is not a hypothetical scenario. This is a standard trajectory for companies with a budget of 1 to 10 million rubles per month that have never seriously engaged in attribution.

According to Ruler Analytics (2024), 40% of marketers make budget decisions based on Last Click. Another 35% use “something in GA4”, not knowing exactly which model is the default (Data-Driven is there, but it requires 3,000+ conversions per month and silently degrades when there is a lack of data). The remaining 25% are either well versed or work with very large budgets, where MMM is already mandatory.

2. Attribution models: from simple to accurate

Before the comparative table - briefly about each model, so that the numbers in the table are clear, and not just numbers.

Last Click — 100% credit to the last channel. Simple, but systematic lies about the upper levels of the funnel. First Click — 100% first touch. It’s also one-sided, but it shows which channel “opens” clients. Useful for evaluating outreach campaigns in isolation.

Linear — equal credit to all touches. If there are three channels, each gets 33%. It does not reflect the real contribution, but it does not break the coverage channels to zero. A working option for teams without resources is something more complicated. Time Decay — greater weight for touches closest to conversion. Logic: the fresher the contact, the more important it is. Suitable for short transaction cycles.

Data-Driven MTA — the algorithm calculates the real contribution of each channel based on statistics on thousands of user paths. Threshold - 3,000+ conversions per month. Below this, the algorithm begins to “guess.” MMM — econometrics on aggregates. Does not look at user paths at all, works with data on channels, sales and external factors. Minimum - $100K/month budget and a year of history.

3. Why Last Click lies: examples and figures

Specific case. EdTech project, online courses, budget 2 million rubles/month. In the Last Click report for 2024, the picture is as follows: context (Yandex Direct) - 58% of conversions, VK retargeting - 27%, email - 10%, SEO - 3%, Telegram placements - 2%.

The marketer cuts Telegram placements and reduces the SEO budget. A quarter later, context and retargeting “unexpectedly” rise in price by 35% according to CPL. What happened: Telegram placements and SEO fed the top level of the funnel - the audience, which then fell into retargeting. When the top level dried up, retargeting began to “process” colder audiences. CPL has increased. Last Click didn't show this - it only saw the last click.

If you look at associated conversions in the same GA4: SEO participated in 41% of all conversions (with 3% “by Last Click”), Telegram placements - in 28%. This is not “influence”, these are real user journeys. Last Click just didn't count them.

The numbers look different for offline businesses too. Clinic, CPL by context in Last Click - 1,800 ₽. When we connected offline conversions (recording through a call center + CRM integration) and switched to Linear, the real CPL of the context turned out to be 3,200 ₽, and SEO, which looked “dead,” was 900 ₽. The difference in decisions is plus 700K ₽ to the SEO budget and minus 400K ₽ to the context in the next quarter.

4. MTA: how it works and when it’s enough

Multi-Touch Attribution works at the level of individual user paths. The algorithm takes thousands of conversion and non-conversion paths, compares them - and calculates which channels actually increased the likelihood of a purchase, and which were simply present.

GA4 Data-Driven MTA - free and built-in. But there is a caveat: it only works in the Google ecosystem. Telegram placements, VK Ads, offline points - they get there only through the Conversions API and enhanced conversions. Without this, GA4 sees an incomplete picture. For the Russian market, where a significant part of the budget goes to VK and Yandex Direct, this is critical.

What's needed for a functioning MTA in 2026:

• UTM markup on all channels - direct traffic no more than 15% (checked in GA4 through the “Attraction” report).
• First-party data: login, email hash, phone - for linking cross-device paths.
• Conversions API in VK Ads and Yandex.Audience - for offline conversions and events from CRM.
• GTM with correct triggers - so that GA4 sees events, and not just pageview.
• 3,000+ conversions per month - otherwise Data-Driven will degrade to Linear.

If the conversions are less than 3,000, don’t try to pull Data-Driven. Linear through GA4 + manual analysis of associated conversions will give a more honest picture than a “smart” algorithm based on insufficient data.

5. MMM: when you need it and how to start

Marketing Mix Modeling - This is a different class of instruments. MMM doesn't look at the user journey. It takes aggregate data: how much was spent on each channel, what was the sales volume, plus external factors (seasonality, competitor prices, promotions) - and builds a regression that explains how much sales are explained by each channel.

The main advantage is cookieless by nature. MMM does not need user tracking. This makes it the future of attribution in a world without third-party cookies. The main drawback is the entry threshold. Needed: minimum $100K per month total media budget, a year of data history (24+ points), and either an in-house analyst with econometrics experience or a commercial tool.

Open-source options: Robyn from Meta (Python/R, free) and LightweightMMM from Google (Python). Both require an understanding of regression analysis. Commercial: Ekimetrics, Nielsen, Analytic Partners - from €50K per year for implementation and support.

On what budget should I switch to MMM? I focus on a threshold of about 9–10 million rubles per month for a media budget. Below - the ROI from implementing MMM does not reduce. Above - without MMM you are making decisions on an incomplete picture because MTA does not see TV, OOH and offline promotions.

6. Cross-device: attribution blind spots

Cross-device - one of the main holes in any MTA model. According to Google, in e-com, 35-45% of conversions go through more than one device. A person sees an advertisement on a phone, researches a product on a laptop, and buys from a tablet. Without an ID, these are three different users for GA4.

How to close the hole: (1) login is the cleanest method if you have a personal account; (2) email hash via enhanced conversions - GA4 stitches sessions together by hash if the user entered an email; (3) probabilistic matching - statistical matching by IP, User Agent, behavior, accuracy 60–70%.

Without solving the cross-device problem, Data-Driven MTA will consistently underestimate mobile channels. The user clicked from a mobile phone, bought from a desktop - the mobile click was “lost” without stitching. In niches with a high share of mobile traffic (gaming, EdTech, lifestyle), this distortion can be 20–30% of the real contribution of mobile.

7. Practical choice: what to use for what budget

Without filler - a specific table for budgets.

One caveat: third-party MTA tools (Northbeam, Rockerbox) cost from $2,000 to $10,000 per month. They are justified if you have several channels with significant budgets and there is no way to build a normal first-party infrastructure through GA4. In the Russian market, most companies in the range of 1–9 million rubles make do with GA4 Data-Driven + manual analysis of associated conversions.

8. What to do about it right now

Attribution is not a one-time task. This is infrastructure. Once you set it up poorly and forget it, a year later you make decisions based on data that does not describe reality.

Specifically: (1) Check the attribution model in GA4 - Settings → Attribution → Reporting attribution model. If you haven’t touched it, it’s Data-Driven; make sure you have 3,000+ conversions per month. (2) Open the Assisted Conversions report and compare the contribution of SEO, email and reach channels with Last Click. The difference will show who you are “robbing.” (3) If the UTM markup is crooked, first fix it through GTM, then you think about attribution.

Related topics: ROAS/ROMI calculator, KPI dashboard template, what is MTA, cookieless attribution.

If you want to analyze your specific situation, write to Telegram or through form. Starting consultation - 0 ₽.