Causality Engine vs Recast: Bayesian MMM or GA4 Paths

The short answer: Recast's own phase list adds up to twelve weeks or more of implementation. A Causality Engine read takes 1 to 2 minutes. If a DTC brand's spend is all digital and shows up in GA4, and the per-channel answer is due this month rather than next quarter, Causality Engine is the pick. The first finding is free, and a read of every channel is €99 once (pricing, checked on 27 October 2026). Recast fits a brand with TV, retail, Amazon or heavy promotions to untangle. That brand also needs a team that can send daily spend data every week, and the patience to sit through the build.

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

Three things decide it:

1. The weeks to a first number. Recast's implementation guide lists nine phases. Kickoff takes one week, data collection two, validation one, model discovery one, model build three to four, then benchmark delivery, gap review, acceptance and handover one each. At the short end that is 1 + 2 + 1 + 1 + 3 + 1 + 1 + 1 + 1 = 12 weeks, before any model iterations. A Causality Engine read finishes in 1 to 2 minutes, per its how-it-works page.
2. The data already on your side. Recast's data guide asks for one row per day, with spend for every channel and your KPI. Ideally about two and a half years of it, plus a promotional calendar. Causality Engine asks for the GA4 Attribution paths export with twelve months or more. Hold the second and not the first, and the choice is made.
3. How you will know it was right. Every Recast customer receives weekly out-of-sample forecast accuracy reports, per its MMM page. A read checks every number against your own export and grades each channel's data from 0 to 100. Pick the test you trust before you pick the tool.

Same budget question, two lenses. A mix model reads spend against sales over time, with saturation and delay built in. A causal read follows the sequence of channels each buyer touched in GA4 and estimates what each one caused, beside its last-click credit.

Recast and Causality Engine, compared

Where they differ

Spend in, or paths in

Recast works from how spend moved against sales, day by day. That is why it wants years of history. Its FAQ names the catch: spend held steady in the past leaves the model little to learn about what happens at higher levels. Its answer to the branded-search puzzle is a neat one. Branded search and affiliates are treated as outcomes of upper-funnel spend, so awareness gets credit for the searches it creates. Causality Engine works the same puzzle from the other end. The buyers recorded under Direct get split back to the channels that sent them, because Direct, as the start page puts it, is where buyers arrive, not why they came.

What each one sees

A mix model sees any spend you can put in a column. TV by week, out-of-home, catalog, retail promotions, Amazon. The read sees what GA4 records and nothing else, and Causality Engine's cost page says so plainly: no offline, retail or in-store measurement, no analyst on a call. Put a third of the budget into channels GA4 never sees, and a read cannot carry the whole allocation.

Who keeps it fed

Recast refreshes weekly, which is much of its value. It also asks for the dataset again every week, through S3, SFTP, BigQuery, Snowflake or a sheet, with alerts when data looks wrong. That is a pipeline, and someone's name is on it. Causality Engine's free and €99 tiers need one export by hand. Pro takes over the GA4 ingestion.

At Twinkels, last-click ROAS on Meta held steady while profit fell. Nobody misread the dashboard; it reported what it was built to report. The beta causal read found Meta-acquired customers returning more product, and the case-studies page reports a 60% revenue lift for the brand in 2024 after acting on it. That is the company's own published result, not an independent study.

Asking the model through an agent

Both let an AI assistant query the results. Recast's MCP post describes a Claude plugin and MCP connection for all customers, reaching 20 reports. One job it suggests: join Recast's incremental ROI with your own Meta export and flag where the two diverge. Causality Engine's Pro plan has a chatbot over your own attribution data. Either way, the agent drafts. A person checks the figure against the dashboard or the read's receipts before it enters a plan.

Cost, as published

Checked on 27 October 2026. Causality Engine's pricing page: €99 per read, one-time, refundable within 30 days. Pro runs €299 a month, billed monthly, and Enterprise is a custom 90-day pilot; prices exclude VAT. Recast publishes no price on getrecast.com. The site offers a demo, and certified Operators may bring their own pricing models, per its home page.

Recast's data guide asks for the whole 27-month dataset again every week, and its implementation builds a programmatic feed to send it. Once that feed exists, the easy Monday is the one where the refresh keeps running. It runs whether or not this week's question needs a model.

The free check first

Before either, a check that costs nothing. Add up last month's revenue as each ad platform reports it, and set the sum against the store's revenue. If the platforms together sold more than the shop did, which of them do you believe? Causality Engine's home page treats a sum larger than what you sold as one of five signs a read is worth it. Then drop the export on the start page for the free first finding, computed in your browser. If both look calm, stop there. If not, the €99 read prices every channel, and Pro suits eight or more channels with weekly budget moves.

Choose Causality Engine if

• Your job is splitting a digital budget across Meta, Google, TikTok, email and organic, and every one of them shows up in GA4.
• You want the answer this month, from history you already collected.
• Nobody on the team will own a weekly data feed.
• You want Direct's credit split back to the channels behind it.

Choose Recast if

• Your job is planning a mix that includes TV, out-of-home, retail or Amazon, where GA4 sees little.
• You want forecasts and budget scenarios refreshed weekly, with a model that absorbs your lift tests.
• You run promotions often and need each one's true lift, net of pulled-forward sales.
• You have a data owner and the months an implementation takes.

The two also stack. Causality Engine's Recast alternatives page, the company's own view, suggests keeping Recast and adding the read as a second opinion.

Start with the read

Causality Engine, which I founded, takes the GA4 export you already have: the first finding is free on the start page, and the full read costs €99, with each step of the method set out on the how-it-works page. The Causality Engine entry in the glossary covers what it is in brief.

FAQ

How much history does Recast need?

Ideally at least 27 months, says its implementation FAQ, and the data guide asks for daily rows. With less, Recast says to discuss it.

Is a Bayesian MMM more accurate than a causal read?

They are accurate about different things. Recast scores each model's forecasts against what then happened. Causality Engine grades each channel's data and checks every number against the export. For a digital-only store, the read answers the per-channel question sooner. For a mix with offline media, the model covers ground the read cannot.

Can Causality Engine use Recast's lift tests?

The pages do not say so. The read takes the GA4 export as its only input, per the how-it-works page. Recast, for its part, takes lift test results as optional inputs to its model.