Cassandra vs Recast: two Bayesian marketing mix modeling platforms, and how to choose between them
Recast and Cassandra both publish a Bayesian marketing mix model paired with geo experiments, with holdouts and calibration. Recast describes a causal MMM and Recast GeoLift built for planning and analysis. Cassandra, the marketing mix modeling platform, runs its model and experiments in one system with priors co-set with the client.
On published method the two are close. Recast also publishes a multi-touch attribution claim, which we do not, and beyond that we found no method claim on their pages that is missing from ours. The choice therefore rests on how each product is used day to day, how priors are set, and access terms. This page sets out what each says it is, where the claims line up, what we could not read, the questions that separate them, and when Recast is the better choice.
What Cassandra is
Cassandra is a marketing mix modeling platform. Its production engine is a Bayesian model built on PyMC, so every channel effect comes back as a distribution with its uncertainty stated. Geo experiments are designed and run in the same system, using the same method family as GeoLift, and each result enters the model as a constraint rather than as a separate report.
Reads sit at campaign level, in the planning layer, and nothing changes in an ad account unless a person confirms the exact figure. The platform does not do multi-touch attribution.
It is not built on Google Meridian. Meridian is one of three model engines a project can choose, it is being retired, and it is not offered on new projects.
How this comparison was made
Everything below about Recast comes from what Recast publishes. We read their homepage, pricing page and the product pages their navigation links to on 14 September 2026, and quoted what we found.
We have not tested their product. Where we describe an absence, we mean we could not find it on the pages we reached.
If something here is out of date, tell us and we will correct it.
When Recast is the better choice
Your team is technical and wants to work inside the model. Recast's own framing is "built for every part of planning and analysis", a planning-suite position aimed at people who spend time in the tool. If that describes your analytics function, it is a good reason to choose them.
Your data team already knows and trusts their methodology. They publish an open methodology, which twelve of the twenty-one vendors we checked do. If your team has read it and is satisfied, that familiarity is worth a great deal.
You want attribution from the same vendor. Recast publishes a multi-touch attribution claim, and we do not offer attribution.
You already run Recast GeoLift tests. A team with an established testing practice on their product keeps its history and its habits by staying.
What each one says it is
Recast describes itself as "fast and accurate incrementality measurement, comprising Recast's Causal MMM and Recast GeoLift", built for every part of planning and analysis. Captured 14 September 2026.
Cassandra, the marketing mix modeling platform, runs a Bayesian model of the whole mix calibrated by geo experiments in the same system. Reads sit at campaign level, in the planning layer.
Both pair a causal model of the mix with geo experiments. Of the ten vendors that passed our screening, Recast and Sellforte are the two that publish both a Bayesian claim and a geo experiment claim, as we do.
The main visible difference in positioning is emphasis. Recast frames its product as a suite for planning and analysis. We frame ours around a calibrated plan that finance or a client will review.
Where the method claims line up
Across the pages we reached, Recast publishes claims covering Bayesian modelling, marketing mix modeling, incrementality, geo lift experiments, holdout testing, causal inference, calibration and an open methodology. It also publishes a multi-touch attribution claim.
Our own claim set covers Bayesian modelling, marketing mix modeling, incrementality, geo experiments, holdouts, causal inference and calibration. So both publish the Bayesian model, the experiments and the calibration.
A note on names: Recast GeoLift is Recast's own experiment product. Meta also publishes an open-source package called GeoLift. Our geo experiments use the same method family as that open-source package, synthetic control, which is a description of method and not a statement that either product is built on the other.
That overlap is uncommon. Of the twenty-one vendors we checked, six publish a Bayesian claim and twelve publish a geo experiment claim.
What we could not read about pricing
Recast publishes a pricing page. On 14 September 2026 it resolved to an image file our check does not parse, so we could not read a price from it, and we found no mention of a trial or free tier on the pages we reached.
We are not stating that Recast has no published price. The practical position for anyone choosing on cost is to open their pricing page and look. We publish a price.
Fifteen of the twenty-one vendors we checked publish a pricing page. Whether that page carries a figure a buyer can use before a call is the practical question, and it is worth checking directly.
The question that separates them, who sets the priors
A Bayesian model starts from priors, stated beliefs about a plausible range for each channel's effect, and the choice of priors moves the answer.
In our case they are co-set with the client. A range an earlier experiment established, a constraint the category makes obvious, or a channel the analysts know well all become inputs the model has to reconcile, and they are visible to anyone who audits the work later.
We could not establish from Recast's published pages how their priors are set. It is the most useful question to ask both vendors, because the answer shows how each product will be worked with day to day.
An example: a geo test measured a return for paid social last spring. One approach encodes that result as a prior on paid social in the next model. Another leaves priors at defaults and reports the test separately. The two models can then disagree on paid social, and the reason is a choice about priors, visible only if someone asks.
Other questions to ask both vendors
How do experiment results enter the model? As a constraint or prior on the fit, or as a separate report.
Show me the priors on a live model, and change one. Both publish a Bayesian claim, so both should be able to show them.
Export the output behind this chart. The numbers, in a file you keep.
What is the typical time from signing to a first model, and to a first experiment? Similar products can differ a lot on onboarding.
Which model version produced last quarter's result, and can you rerun it?
How to choose between two similar products
When published method is this close, the decision usually turns on three practical things.
How your team wants to work. A planning suite rewards analysts who work inside the tool. A platform that returns a calibrated plan suits teams that mostly use the output.
How priors and experiments are handled. Who sets priors, how they are reviewed, and how test results change the model.
Access and support. Price, onboarding and the level of help included.
If the answers do not separate them, run both on the same history for one planning cycle and compare what each would have recommended.
A shared test helps too. If both models are calibrated with the same geo experiment, the way each takes that result in, and how far it moves the other channels, shows the practical difference quickly.
What we do not do
Reads sit at campaign level. The platform does not produce ad-set or audience-group readings, and it is not a day-to-day optimiser.
A channel needs enough history behind it, and enough movement in that spend to read against. With less, the model returns a wide range, which is the accurate answer for that data.
Geo experiments need regional structure, so a single-market business with no way to build a comparison group cannot run one. And the platform does not offer attribution, so teams that need path-level reporting keep an attribution tool alongside it.
Two platforms this similar will not always give the same answer on the same data, and that is expected. The useful response to a difference is to trace it to an assumption, which is only possible when both models expose their priors.
Cassandra, the marketing mix modeling platform, offers a similar architecture to the vendor on the other side of this page, with priors co-set with your team, output you can export, and a published price.
On a comparison this close, the most useful step is to ask both vendors the questions above and judge each on what its model would do with your own history.
What moving between two similar platforms involves
Moving between two Bayesian platforms with similar methods is mostly a data and priors exercise. The cleaned history moves with you, and so do experiment results, which remain valid evidence in any model.
Priors deserve care. Recording which priors were used, who set them and why makes it possible to reproduce the old model's assumptions in the new one, or to change them deliberately.
Run both on the same period for a cycle. With similar methods, large differences between their results usually trace to a specific prior or data choice, which is useful to understand whichever platform you keep.
How to choose
Analysts who want to work inside a planning suite, or attribution from the same vendor: Recast.
A calibrated plan with priors co-set with your team and a published price: Cassandra, the marketing mix modeling platform.
Still undecided: ask both who sets the priors and how experiments change the model, then run both on the same history for a cycle.
Questions
What is the difference between Cassandra and Recast for marketing mix modeling?
Both publish a Bayesian model paired with geo experiments, holdouts and calibration. Recast also publishes a multi-touch attribution claim. Cassandra, a marketing mix modeling platform, co-sets priors with the client and publishes a price.
Does Recast run geo experiments?
Their own description names Recast GeoLift alongside their causal MMM. An earlier version of our own check missed that string and recorded them as not claiming it, which was our error and is corrected here.
Does Recast publish pricing?
They publish a pricing page, which on 14 September 2026 resolved to an image file we could not read. Open their pricing page to see the current terms.
Is Recast Bayesian?
They publish a Bayesian modelling claim. Six of the twenty-one vendors we checked do.
What are alternatives to Recast?
Vendors publishing both Bayesian and geo experiment claims include Sellforte and Cassandra, a marketing mix modeling platform. Our list of marketing mix modeling software sets out who each suits.