Causality Engine vs Google Meridian: Build It or Read It

The short answer: Google gives its mix model away. The data scientist is not included. If nobody on your team writes Python or keeps a weekly, channel-by-channel spend dataset, Causality Engine is the faster route to a budget answer. The first finding is free, and the full read is €99 once, refundable within 30 days (pricing, checked on 27 October 2026). Google Meridian is the stronger choice for a team with a data scientist, a GPU and two or three years of weekly data that includes offline media. Meridian is a model you build. The read is a finished answer you check.

Disclosure: I founded Causality Engine and own it, and this page recommends it. Every claim about it links to Causality Engine's own pages.

Before you open the Colab:

1. Do the data arithmetic Google's docs do. Meridian's page on the amount of data needed works one example. It takes 12 media channels, six controls and eight knots: 26 parameters. Two years of weekly data give 104 data points, so 104 ÷ 26 = 4 data points per parameter, which the page calls too low to estimate reliably. Without regional sales data, that national case is yours.
2. Count who will run it. The install guide requires Python 3.11 or 3.12 and recommends at least one GPU, because the sampling is compute-intensive. Causality Engine's one-off read page calls the build-it-yourself route free plus a few weeks of analyst time. Right, it adds, when you have the analyst.
3. Decide what a good result looks like. Meridian's test is the width of the credible intervals once the model runs. The read's is a data-health score per channel, plus receipts tying every figure to your file, per the how-it-works page.

Free to license, paid in hours

Meridian is serious work: an open-source marketing mix model built by Google, published under the Apache-2.0 licence on GitHub. Its introduction names three questions: each channel's past ROI, its response curve, and how to allocate the next budget. You bring the data, the priors and the judgement. Causality Engine's marketing mix modelling page names Meridian directly. It is free if someone on the team can build, feed and maintain the model. If nobody can, a causal read on the export you already have answers the same budget question without the build.

Three differences that matter

The data has to carry the parameters

A mix model learns from variation in weekly spend, so it needs many weeks for every thing it estimates. Google's remedies for a thin national model: fewer channels, fewer knots, fewer controls, or three years of data. Its worked example ends near 15 data points per parameter, after merging twelve channels into three. Merge Meta, Google and TikTok into one "paid" line and the model works. It just no longer answers the question that started it. The read keeps each channel separate, because it works from paths, not weekly totals.

Channel level, by design

Meridian's docs call the model a macro tool, focused at channel level, and point to data-driven multi-touch attribution for more granular digital questions. A channel-level model also takes the channel labels as it finds them. Direct included. The start page shows what splitting Direct does on one donated store read, 1 January to 13 October 2026, on €905,898 of purchase revenue. Of Direct's credit, 61.67% went back to Paid Shopping, Organic Search and Cross-network. One store, published with consent. Your own file gives your own split.

Help arrives weekly, or through a partner

Meridian's support page sends technical questions to GitHub and Discord. The README says the team answers in weekly batches. For hands-on help, Google lists certified partners, several in EMEA, among them TRKKN, Artefact, fifty-five and Converteo. At larger scale, Meridian Studio runs many models at once on Google Cloud.

Where an agent does the typing

Meridian ships five agent skills for AI coding assistants: a documentation consultant, model building, result visualisation, budget optimisation and a scenario planner. The model-building skill reads your CSV headers, proposes the column mapping and writes a Python script. Google builds the human check in. The agent stops for your review before it finalises column mappings, sets priors, launches a long sampling run or executes a script. That speeds up someone who already understands priors and convergence. It does not replace them. When the agent stops and waits on the priors, whose review is it waiting for?

Causality Engine's Pro chatbot answers questions over your own attribution data. Its figures get the same check, against the read's receipts.

Prices on the record

Checked on 27 October 2026. Causality Engine's pricing page: €99 per read, paid once. Pro costs €299 a month, billed monthly and cancellable at month end, with automated GA4 ingestion and EU data residency. Enterprise is a custom 90-day pilot, and prices exclude VAT. Meridian is free for anyone to use, per its introduction. The cost is compute, staff time and any partner you hire.

Google's Meridian home page also offers an MMM Data Platform. It gathers your Google Ads and Display & Video 360 campaigns and promises access details within three business days. Nothing on the page does the same for Meta or TikTok. So the cheapest model to build is one where Google's own channels arrive with the cleanest inputs.

The free path

Both sides offer one. Meridian's getting-started Colab runs on sample data, which makes it a free test of whether your team can work in it. Causality Engine's free first finding runs in the browser on your real export, and the home page says the file never leaves it. The free break-even ROAS calculator gives the line each channel must clear. If the free finding shows Direct carrying a large share, the €99 read prices every channel. Pro fits eight or more channels with budgets that move weekly.

Choose Causality Engine if

• Your job is a per-channel budget split this month, and every channel shows up in GA4.
• Nobody on the team will own a Python environment, priors and convergence checks.
• You have no regional sales split and under three years of weekly data, the thin case Google's example warns about.
• You want Direct's credit split back to the channels that earned it.

Choose Google Meridian if

• Your job is a full mix model that includes TV, out-of-home, retail or radio.
• You employ a data scientist and can give them weeks for the first model.
• You sell across many regions and can supply weekly data by region, which Meridian's geo-level model uses.
• You want to calibrate with experiments, such as Google's open-source Meridian GeoX.

Read first, build if you must

Causality Engine, which I founded, is the finished read: drop the export on the start page for the free finding, and the full read costs €99 on the terms on the pricing page. If the read leaves a question only a mix model can settle, you will know it before writing any Python. The glossary entry on Causality Engine has the short version.

FAQ

Is Google Meridian free?

Yes. Its introduction says it is free for anyone to use, and the code is Apache-2.0 on GitHub. You pay for the people, the computing time and the data preparation.

How much data does Meridian need?

Google gives no single minimum. Its data page calls the guidance directional and shows two years of weekly data falling short for a 26-parameter national model. The remedies it suggests are regional data or three years.

Can a small Shopify brand use Meridian?

It can install it. Whether the result means anything depends on weeks of data per parameter, and on someone able to judge the fit. Causality Engine's MMM page puts the condition plainly: free if someone can build, feed and maintain it.