Google has made Meridian, its open-source marketing mix model, available to everyone. That matters for B2B SaaS because marketing mix modeling (MMM) answers a question your Google Ads attribution report structurally cannot: not “which click gets credit for this lead,” but “how much revenue would we lose if we cut this channel — and where should the next dollar go?” In a world of long sales cycles, consent-driven data loss, and blended spend across Google, LinkedIn, and demand gen, that is exactly the question a founder or head of growth needs answered before approving next quarter’s budget.
But “free and open-source” is doing a lot of work in the announcement, and MMM is not a drop-in upgrade for every SaaS account. This guide defines what Meridian actually is, explains how MMM differs from the attribution you already run, sets an honest bar for when it is worth the investment, and shows where it fits alongside your existing Google Ads measurement rather than replacing it. If your starting point is click-path measurement, our primer on Google Ads attribution models is the right companion read — MMM lives one altitude above it.
What Meridian actually is
Meridian is a Python-based, fully Bayesian marketing mix modeling package from Google, now free for any advertiser to download and run. In plain terms, it estimates the incremental revenue each marketing channel produces — the sales that would not have happened without that channel’s spend — using aggregate, geo-level data rather than individual clicks or cookies. Because it never touches user-level identifiers, it is privacy-safe by construction and immune to the consent and signal loss that is quietly degrading click-based measurement. It replaces Google’s earlier Lightweight MMM library and is built to slot into a modern data-science stack.
Two modeling ideas make it more than a spreadsheet regression. First, adstock: the effect of ad spend on revenue is not instantaneous but accumulates and decays over days or weeks, which matters enormously for B2B SaaS where a click today can become a closed deal ninety days later. Second, saturation: the return on an extra dollar is not linear — every channel eventually hits diminishing returns, and MMM estimates where that curve bends so you stop over-funding a saturated channel. Being Bayesian, Meridian returns full probability distributions rather than a single confident-looking number, so instead of “Google Ads drove $412,000,” you get an honest range with stated uncertainty. For budget decisions, that honesty is the feature, not a limitation.
MMM vs attribution: two different questions
The most common mistake is treating MMM and attribution as competitors when they answer different questions and belong together. Attribution — including Google’s data-driven attribution — operates at the click and user level, tracing the touchpoints that preceded a specific conversion. It is the right tool for tactical, in-channel decisions: which keywords to bid up, which ad copy converts, which campaign to scale. MMM operates top-down on aggregate spend and revenue, estimating each channel’s contribution over time without tracking anyone. It is the right tool for strategic decisions: how to split the budget across Google, LinkedIn, review sites, and demand gen for the coming quarter.
Put concretely: attribution tells you which click to credit; MMM tells you whether the channel as a whole is pulling its weight. This distinction is why the 2026 consensus among B2B measurement teams is “method stacking,” not picking a winner — attribution for tactical optimization, MMM for strategic allocation, and incrementality tests as ground truth. MMM is also the corrective for a failure mode B2B SaaS teams know well: a Google Ads account that looks efficient on last-click but looks profitable while producing no real pipeline. A mix model that reads revenue rather than form-fills exposes that gap because it cannot be fooled by a cheap conversion that never becomes a customer.
Why long SaaS sales cycles break last-click — and where MMM helps
Last-click attribution is at its weakest precisely where B2B SaaS lives: long, multi-touch buying journeys with heavy assist from channels that rarely get the final click. A prospect might discover you through a demand-gen video, revisit via a review-site listing, and only later search your brand and click a Google ad before booking a demo — at which point last-click hands Google 100% of the credit and zero to the channels that created the demand. Over a quarter, that mis-attribution compounds into real money moved to the wrong places. MMM corrects it because it reads the whole system: it can attribute lift to upper-funnel spend that never earned a click, and it captures the delayed, adstocked payoff of a channel whose deals close months after the impression.
This is the same structural problem behind blended CAC — the recognition that per-channel numbers lie when channels assist each other, and that the honest denominator is total spend against total new revenue. MMM is, in effect, the quantitative engine that decomposes blended performance back into per-channel incrementality without pretending each conversion had a single cause. For a head of growth defending a LinkedIn or demand-gen line item that Google Ads reporting makes look unprofitable, an MMM estimate of that channel’s assisted, incremental contribution is the single most useful artifact you can bring to a budget meeting.
When it is worth it — and the honest cost
MMM starts to pay off for B2B SaaS above roughly $30,000–$50,000 a month in blended spend across at least three channels, with enough conversion volume and history for the Bayesian model to produce stable, usefully narrow estimates. Below that threshold, or for a company running Google Ads only, the data is too sparse and the channel mix too simple; the model will hand you wide, uncertain ranges that do not justify the effort. MMM learns from variation — different spend levels across geographies and time — so a small, single-channel account simply does not give it enough to work with. Smaller SaaS advertisers should invest first in clean conversion tracking and cheap geo-based incrementality experiments, then graduate to MMM as spend and channel count grow.
And the “free” needs an asterisk. Meridian’s software licence is genuinely zero, but a working program requires structured spend-and-revenue data by channel and region, a data scientist who can specify and validate the model, and experiments to calibrate it against reality. Industry estimates put an in-house open-source MMM at roughly $100,000–$200,000 a year in headcount, against $2,000–$10,000 a month for a managed MMM platform and $200,000–$500,000 for legacy consulting. So Meridian removes vendor lock-in, not the operational cost. The decision is not “free tool, why not” — it is whether your spend is large enough that a five- or six-figure measurement program pays for itself by reallocating budget more accurately. Above a few hundred thousand a month in blended spend, it usually does; below the threshold above, it usually does not.
How to run it alongside Google Ads
Treat Meridian as the top layer of a three-tier stack, not a replacement for anything you already run. Keep your Google Ads conversion tracking and enhanced conversions solid — they still drive Smart Bidding and day-to-day optimization — and let MMM consume aggregate outputs to answer the budget-allocation question those tools cannot. The practical sequence is: get conversion tracking clean and outcome-based first, ideally already optimizing for SQLs rather than raw leads; assemble at least a year of weekly spend-and-revenue data by channel and geography; run experiments you can use to calibrate the model; then fit Meridian and use its response curves to rebalance spend where returns are highest.
The reason the tracking layer has to be right first is that MMM is only as trustworthy as the revenue signal you feed it. If Google Ads is optimizing toward cheap conversions, your aggregate revenue series will be noisy and the model will inherit that noise. Feeding real downstream value into bidding — the discipline behind a SaaS conversion value ladder and value-based bidding — is what makes both the in-channel optimization and the mix model honest. Done in order, the layers reinforce each other: clean value signals improve bidding, better bidding produces cleaner revenue data, and cleaner data makes Meridian’s allocation recommendations something you can actually act on.
What Meridian will not do
MMM is a strategic instrument, not a real-time dashboard, and expecting the wrong thing from it is the fastest way to be disappointed. It will not tell you which keyword converted this morning, it will not optimize a bid, and it will not give you a fresh read every day — mix models are refit on weekly or monthly cadences and are built for allocation decisions measured in quarters, not intraday tweaks. It also demands genuine statistical variation to identify effects; if you have spent a flat amount on every channel for two years, the model has little to learn from and its estimates will be correspondingly vague. And because it works on aggregate data, it cannot answer user-level questions — that is attribution’s job.
There is also a discipline cost that is easy to underestimate: a Bayesian MMM requires priors, validation, and calibration against experiments to be trusted, and an uncalibrated model can produce confident-looking numbers that are simply wrong. This is why the mature answer is to combine methods rather than crown one. Use MMM for annual and quarterly budgeting, attribution and conversion tracking for tactical in-channel work, and incrementality tests to keep the model honest. For B2B SaaS teams whose spend clears the threshold, Meridian being free removes the last easy excuse not to measure channel contribution properly — but the work of measuring it well is still yours to fund, and it is that work, not the software, that turns a mix model into better budget decisions. Pair the exercise with realistic B2B SaaS CAC benchmarks so the reallocations MMM suggests are judged against where your acquisition costs should actually land.