How to Get Marketing Mix Answers Without a Data Team

The short answer: the free mix models cost the build instead of a licence. That matters when nobody on your team writes Python or R. Google's Meridian needs Python 3.11 or 3.12 and recommends at least one GPU, per its install page. Meta's analyst's guide to Robyn plans 12 to 22 weeks from scoping to recommendations. For a store whose two questions are where the next euro goes and what to cut, Causality Engine answers both per channel from the GA4 export you already have. It costs €99 once, per its pricing page, and the first finding is free on the start page. The steps: phrase the questions as budget moves and check your data against what a mix model needs. Then run the two free checks, read every channel and move money in steps.

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

What decides build or buy:

1. The calendar. Robyn's guide gives each phase a range. Defining the questions takes 1 to 2 weeks, data collection 4 to 6, data review 1 to 2, modelling 4 to 8 and analysis 2 to 4. Added up, that is 12 to 22 weeks (1 + 4 + 1 + 4 + 2 at the short end, 2 + 6 + 2 + 8 + 4 at the long end), per the analyst's guide. The budget decision usually comes sooner.
2. The history you hold. The same guide asks for a minimum of two years of weekly data. Causality Engine reads whatever GA4 history you export, with 12 months or more recommended on how it works.
3. How the answer shows its uncertainty. Meridian reports ROI with credible intervals, response curves and a budget optimisation, per Google's introduction. Causality Engine's read gives each channel a data-health score from 0 to 100 instead of an interval, as its Shopify guide states. Know which one you are buying before you compare them.

Neither route is wrong. They suit different teams.

What goes in, what comes out

• In: your GA4 property with purchase revenue, monthly spend per channel, your gross margin and average order value, and one budget decision that has a date on it.
• Out: a caused share per channel next to last-click, a break-even line to judge it against, and a next step per channel (raise, cut or hold).

Step 1: Turn "the mix" into two budget questions

Meridian's introduction names the three questions a mix model is built for: historical ROI and contribution per channel, the response curve for each channel, and how to allocate the next budget. Write yours in that shape, with your own numbers. For example: "If €2,000 a month leaves Paid Social, what happens to revenue?" and "Which channel takes the next €2,000?" A question with a figure in it can be answered. "How is the mix doing" cannot.

Step 2: Check your data against what a mix model needs

Robyn's analyst's guide lists the conditions. Two years or more of weekly data. Enough variation, since a channel whose spend never changed leaves the model unable to tell what it did. And enough volume for a channel to move the total. Meridian asks for a KPI, media data per channel and any control variables, and prefers geo-level data, per its introduction. If you fail the two-year test, the build route is closed for now, whatever the team. No amount of talent adds a second year of history.

Step 3: If an analyst is available, run the model yourself

Meridian installs from PyPI and runs a Bayesian model sampled with NUTS. That is compute-heavy, hence the GPU advice on the install page. Robyn installs from CRAN in R and needs a one-time install of the Python library Nevergrad, per its installation page. Both are free. Robyn's guide is direct that its automation does not remove the analyst, who still tunes parameters and interprets results. The software is free. The analyst is not.

Step 4: Without an analyst, run the two free checks

First, the bar. Causality Engine's free break-even ROAS calculator works out the contribution per order from order value, margin, returns, fees and shipping. It says to compare the break-even figure with your MER (total revenue over total ad spend), not with one platform's ROAS. Second, one anchor: a holdout on the channel with the most spend, set out in the geo holdout how-to. Together they tell you whether the marketing pays, and whether the biggest channel earns its place in the side.

Step 5: Read every channel from the GA4 export

Export Key event attribution paths from GA4 and drop the CSV on the start page. The privacy policy says the free preview reads it in your browser and does not upload it. The €99 read returns, per channel, its share of credit after Direct is split back, with last-click beside it. Then a data-health score and a next step to raise, cut or hold, tested in one region first, per how it works. Causality Engine's own mix modelling page describes it as causal attribution software, the fast alternative to a mix model. That is the right frame: one export in, a verdict per channel out.

Step 6: Ask follow-ups with an agent, and verify each answer

Pro adds a chatbot for ad-hoc questions of your attribution data, plus budget alerts when a channel's performance shifts, at €299 a month, billed monthly, per the pricing page. Use it for the questions from step 1, phrased against the read ("which channels sit more than 25% above last-click?"). Each answer that changes a budget line gets checked by a person against the channel table in the read. A chatbot answer is a pointer to a figure. The figure is what you defend.

Step 7: Turn the answers into budget moves

The agents page sets the rules. Within 25% of last-click, hold. Under 1% on both methods, too small to act on. Under 90 data health, move in steps. Then repeat: four quarterly reads cost €396 a year, per the checkup page.

Where each route stops

• The read does not model offline spend. Causality Engine's cost page lists no offline, retail or in-store measurement and no analyst on a call.
• Response curves are not in the read. The documented output on the agents page is shares, data health, next steps, paths and receipts. For saturation and the shape of each channel's returns, Meridian and Robyn are the tools built for it.
• A mix model with flat spend learns nothing. Robyn's guide says variation is what the model reads.

Both free frameworks come from companies that sell the media they measure. Meridian comes from Google, whose Meridian home page offers to gather your Google Ads and Display & Video 360 data for the model. Robyn comes from Meta's Marketing Science team, calibrated against ground truth that includes Facebook lift studies. The open code is the check on that, and it only works if someone on your side can read it.

Free checks first, then the read

Steps 1, 2 and 4 cost nothing and settle whether the marketing pays at all. Causality Engine, which I founded, answers the per-channel part without a model to build. The first finding is free on the start page. The read is €99 once and refundable within 30 days on the pricing page (checked on 27 October 2026), and the method is on how it works. The Causality Engine entry summarises the product.

FAQ

Is Causality Engine a marketing mix model?

Not in the Meridian or Robyn sense. Its mix modelling page calls it causal attribution software and the fast alternative to a mix model. The terms define the €99 read as one analysis of one GA4 attribution paths export.

When should I still build a mix model?

When TV, retail or other offline spend is a large part of the budget, when you need response curves, and when someone can maintain the model. Meridian's introduction lists those outputs.

Can I start with Robyn and a year of data?

Robyn's analyst's guide asks for two years of weekly data at minimum, and more if only monthly data exists. With one year, a read of the GA4 export plus a holdout is the realistic route.