MMM Isn't Just for Enterprises Anymore

Marketing mix modeling — a statistical method for estimating how much each marketing channel contributes to sales — has existed since the 1960s. Procter & Gamble, Coca-Cola, banks, and telecoms all used it. Always large companies, because MMM required teams of statisticians, millions of rows of historical data, and six-figure consulting budgets.

In 2025, Google released its own MMM framework, Meridian, as a free open-source tool available to anyone. In February 2026, Scenario Planner launched on top of it — a graphical interface that lets users work with the model without writing code. Meanwhile, the industry has largely abandoned the illusion that last-click attribution is measurement, and MMM has returned to the center of attention.

This is a meaningful shift. But for the average service business in Kazakhstan, there is a caveat: the fact that MMM has become free does not mean it has become applicable. Here is what actually happened, and what to take from it right now.

What MMM is in one paragraph

MMM is a statistical model that takes historical data on marketing spend by channel — Google, Meta, TikTok, TV, radio, outdoor, email, and so on — alongside sales data for the same period, and estimates how much each channel contributed. Unlike attribution, MMM does not track individual users. It works on aggregated data, week by week, month by month. The output is an estimate of each channel's contribution to sales, accounting for contextual factors (seasonality, prices, promotions, weather, competitors), saturation curves (where doubling the budget does not double the result), and the long-term effects of brand building.

The main practical advantage is that MMM does not depend on ATT, ad blockers, cookies, or consent flows. It operates on aggregates that none of those affect. In an era of privacy, that is precisely what made MMM relevant again — after two decades during which the industry was absorbed with click-based attribution.

What Meridian offers

In January 2025, Google announced that Meridian would become an open-source MMM framework available to all — marketers and data scientists alike. Harikesh Nair, Senior Director of Data Science & Engineering at Google, framed the idea in the announcement: 'Meridian is a modern approach to MMM, built on Bayesian causal inference, that lets you blend your prior knowledge with real data and surfaces the true incremental impact of marketing.'

Several technical decisions in the release make Meridian distinct from classical MMM and are worth noting.

Reach and frequency, not impressions. Older MMM frameworks measured video advertising by impressions. But 10 impressions could mean 10 different people each seeing the ad once, or one person seeing it 10 times — those are different business effects. Meridian distinguishes reach from frequency, which improves measurement for YouTube and video formats.

Experiments as prior estimates. This is technically significant: results from incrementality tests (covered in the previous guide) can be loaded into the model as 'prior knowledge.' A business that runs a geo-experiment on one channel can embed those results into the broader model — so the model calibrates against reality rather than operating in a statistical vacuum.

Non-media variables. Prices, promotions, and assortment changes can be added to the model. This transforms MMM from an 'advertising model' into a 'business growth model' — one that can separate what marketing drove from what a price reduction or seasonal demand drove.

Open source. Meridian is published on GitHub. Anyone can read it, modify it, and adapt it to their business. It is not a Google black box — it is a framework that can be studied and controlled.

Google cited a client example — Australian financial company Finder, which implemented Meridian. From their case study in the announcement: 'The insights we received confirmed the incremental value that YouTube drives beyond what is visible through standard conversion tracking.' In other words, Meridian detected a channel's contribution that classical attribution was undercounting.

Scenario Planner: the next step toward accessibility

In February 2026, Google added Scenario Planner on top of Meridian — a graphical interface for modeling budget scenarios without programming. According to MarTech's description, it is an interactive environment for non-technical marketers: the model is already built and trained, and the user moves budget sliders by channel and sees a sales forecast.

This is an important shift. Before 2026, using MMM required a data scientist who could read Bayesian statistics and interpret model outputs. Scenario Planner removes that barrier. A CMO or marketing director can run a scenario — 'what if we move 30% of the budget from performance to brand?' — and get a forecast without writing a line of code.

In parallel, Google announced the integration of Meridian into Google Analytics 360 — the enterprise version of its analytics product. For businesses already on 360, MMM becomes part of the analytics stack rather than a separate undertaking.

The reality: which businesses can use this right now

The main filter is data. MMM does not just require historical data — it requires data of a specific quality and volume:

• At least two to three years of weekly history. Less than that, and the model cannot distinguish seasonality from trend from noise.
• Several active channels with meaningful budgets. If 90% of spend goes to one channel, there is nothing to model — there is no variation.
• Clean, labeled data on spend and sales. 'How much did we spend on Instagram in the week of February 12, 2024?' needs an answer accurate to the tenge.
• Sufficient sales volume in each week. On small numbers, the model captures noise.
• Contextual variables, ideally. A promotions calendar, prices, public holidays, weather (for offline retail).

For a large retailer or e-commerce business with millions of customers, all of this exists by default. For a dental clinic in Almaty with a budget of ₸1.5 million per month and a dozen new patients per week, MMM will currently return noise rather than answers — not because Meridian is a poor tool, but because the model lacks enough signal to separate channel contributions from random variation.

This needs to be stated plainly: for the average service business in Kazakhstan, MMM is too early. And the fact that Meridian is free does not change that — the cost of the tool was never the main barrier. The barrier has always been data volume.

What to do while MMM is still too early

Some of what MMM gives large companies can be approximated with simpler means right now. The logic is the same — estimate each channel's relative contribution and make budget decisions without relying solely on attribution.

A simple weekly channel log. Build a table: week, budget per channel, leads from CRM, deals closed, revenue, key contextual events (promotions, holidays, competitor activity). After six to twelve months, patterns become visible — which channel correlates with lead volume, which does not. This is not a statistical model, but it is not 'going by feel' either.

On/off tests — the primary substitute. Covered in detail in the previous guide. They give the most direct answer to 'what does this channel contribute?' and require no historical data.

Comparison across locations. If a business has multiple branches, different channel mixes at different locations already provide a rough MMM analog: which combinations work, which do not. This is free and works with the volumes already available.

Context recorded alongside the numbers. What MMM calls 'non-media variables' is, for a mid-size business, simply a calendar: 'we ran a promotion — leads up 20%,' 'a competitor opened in our area — down 15%.' A simple document kept alongside the marketing report turns numbers into something explainable.

When to return to MMM. Once two or more years of weekly history have accumulated, four to five channels are active with comparable budgets, and there are at least 50 to 100 leads per week — the data is ready to run through Meridian. Until then, it is like building econometrics on two data points: formally it computes, practically it says nothing.

Checklist

What to do now to be ready for MMM in two to three years:

1. Start keeping a weekly log: budget by channel, CRM leads, deals closed, revenue, contextual notes. A simple table, updated once a week.
2. Clean up spend labeling: so you can answer unambiguously — 'how much did Instagram cost in week X?' UTM structure, campaign hierarchy, a consistent format.
3. Record a context calendar: promotions, public holidays, competitor openings, price changes — kept alongside the marketing data.
4. Do not overlook the brand channel: brand awareness is the hardest contribution to measure and the most underestimated. If you have a recognizable brand in Kazakhstan, that itself contributes to sales in ways that no ad platform dashboard will show.
5. In 12 months — look at the accumulated data and decide: is there enough signal to try a simplified MMM? Meta's robyn package is available in R; Google's Meridian runs in Python.
6. If the business is growing quickly — watch the Scenario Planner release notes. In 18 months, its interface may make MMM genuinely reachable for a business spending ₸5–10 million per month.

What to understand about this shift

The Meridian release is not a signal that 'MMM is now for everyone.' It is a signal that measurement is becoming more democratic — and in three to five years, the level of analytics available to mid-size businesses will be much closer to the enterprise level than it is today. Right now the barriers are data, time, and expertise. Tomorrow the first barrier will remain (no tool release removes it), while the other two will gradually weaken.

The practical conclusion for service businesses in Kazakhstan is not 'go implement Meridian.' It is to prepare for it: maintain data in a format that can be loaded into a model in two to three years. A business that starts collecting structured data now will have a working MMM at almost no cost in two years. One that did not — will be starting from scratch again.

At Alteora, we help service businesses build this data structure — not as a bet on MMM tomorrow, but as a condition for any meaningful measurement today. A clean format now is the foundation for serious analytics later.