Cassandra vs Rockerbox: a measurement platform built on attribution, or a marketing mix model calibrated by experiments
Rockerbox publishes a unified measurement platform combining multi-touch attribution, marketing mix modeling and incrementality testing on one data foundation, with a free plan and a published starting price. Cassandra, the marketing mix modeling platform, publishes a Bayesian model with geo experiments run in the same system, and no attribution.
Both say that attribution, modeling and tests answer different questions, and that test results should shape the other numbers. They differ in where they start. Rockerbox starts from the data foundation and attribution, with the model and testing alongside. We start from the model and put experiments inside it. This page sets out what each publishes, how each uses test results, what each publishes about pricing, and when Rockerbox is the better choice. Many ecommerce teams use both kinds of tool, so it also covers running the two side by side.
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 Rockerbox comes from what Rockerbox publishes. We read their homepage, product, marketing mix modeling, plans, integrations and resources pages and one blog post, on 14 and 24 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, not that it does not exist.
If something here is out of date, tell us and we will correct it.
When Rockerbox is the better choice
You need multi-touch attribution every day. Rockerbox publishes daily, granular attribution for day-to-day optimisation. We do not offer attribution, so a team that runs on that signal needs a tool like theirs.
You want one data foundation under every method. Their platform is built on a centralised data foundation, with export to a data warehouse or Google Sheets. If fragmented data is the first problem, that foundation addresses it before any model does.
You are a smaller Shopify brand. They publish a free plan and a Starter plan built for Shopify brands, with a published starting price. That is a low-cost way in.
You want testing delivered as a managed service. They describe their testing as a fully managed offering, so the design and running of tests sits with them.
What each one says it is
Rockerbox describes itself as "The Platform of Record for All Marketing Measurement", and as "a unified measurement platform built on a centralized, SOC2-certified data foundation". Its product pages cover multi-touch attribution, marketing mix modeling and incrementality testing. Captured 24 September 2026.
Cassandra, the marketing mix modeling platform, runs a Bayesian model of the whole mix with geo experiments designed and run in the same system, and each result constrains the model. Reads sit at campaign level, in the planning layer.
Rockerbox describes a data foundation with three methods on top of it. We describe one model with experiments inside it. Most of the differences below follow from that.
How each uses test results
Rockerbox publishes that "Attribution and MMM surface hypotheses. Testing validates them with causal evidence", and that "those test results can calibrate your MTA model and inform MMM as priors".
We take the same view that tests should shape the model. In our case a geo experiment's result enters the Bayesian model directly as a constraint on the tested channel, and every later refresh keeps it.
The practical question for both vendors is the same. Which channels have been tested, how recently, and how exactly does each result change the numbers a team plans with? Ask to see one channel's figure before and after a test result was added.
An example: a model credits paid social with a strong return, and a test in a set of regions shows a smaller lift. If the test result only sits in a separate report, the plan may still follow the model. If it enters the model as a constraint, every later budget scenario uses the tested figure. Both vendors can show you which of the two happens.
What each publishes about testing
Rockerbox describes incrementality testing as a way to "establish causality", and states that "by isolating variables and using control groups, Rockerbox helps you measure the incremental lift". Its plans page describes testing as "our fully managed offering".
A blog post on its site recommends running geo-based holdout tests, with control markets where spend is reduced or paused. We found no product page that describes geo experiments as a Rockerbox feature. That describes what we could find, and the test designs they run are worth asking about directly.
We publish geo experiments designed and run in the same system as the model, using the same method family as GeoLift. Regions are matched on their history before the test.
How far apart platform-reported and modelled returns are
Attribution and a marketing mix model often give different answers for the same spend. We measured how far apart on our own data. Across 449 channel-model pairs from 87 models and 22 advertisers, week-matched, the middle half of advertisers saw incremental return between 0.47 and 1.14 times what Meta reported and between 0.27 and 1.23 times what Google reported. ([full analysis](/blog/marketing-attribution-software-analysis))
Platform-reported was the higher figure for 58% of advertisers on Meta and 72% on Google. For the rest, the model found more value than the platform reported.
These are platform-reported figures, not Rockerbox's. Rockerbox publishes its own attribution, built on first-party data. The analysis shows why a tracked number and a modelled number can differ, and why a test is the way to settle which one to plan with.
One caveat belongs with the numbers. The gap mixes attribution windows, view-through crediting, genuine incrementality and the uncertainty in our own estimate, which is why we report a ratio.
What each publishes about pricing
Rockerbox publishes a free plan, described as giving "basic insights into your marketing and industry spend trends", with a note that it "does not include first-party data or deduplication". Its Starter plan states "Pricing starts at: $150/month" and is "built for Shopify brands spending less than $1M". Captured 24 September 2026.
Testing and the marketing mix model are listed as separate products. Their pages route to a conversation with the team rather than to a price.
We publish a price, and every route in is a booked demo. For a like-for-like comparison at larger scale, ask Rockerbox for a figure that includes the marketing mix model and testing, at the same scope you would ask of us.
A low entry price and a full measurement programme are different purchases. The free and Starter plans cover data and attribution. A comparison with a marketing mix modeling platform is fair only at the tier that includes the model and the tests.
What a team gives up with each
With Rockerbox, a team gets daily attribution, a data foundation and the other methods from one vendor. The question to ask is how the marketing mix model and the tests connect to the attribution numbers in practice, and who decides when they disagree.
With Cassandra, the marketing mix modeling platform, a team gets a model of the whole outcome with experiments inside it. Demand that arrives without marketing behind it appears as its own component. The trade-off is the daily attribution signal, which we do not provide.
Many ecommerce teams use both kinds of tool. The useful decision is which question each one answers, agreed before the numbers disagree.
Questions to ask both vendors
Which channels have been tested, and when? A figure anchored by a recent test carries more weight than one resting on history alone.
How does a test result change the plan? Ask to see one channel before and after a result was added.
What is the number's denominator? Tracked conversions, or the whole outcome.
What happens to a channel that adds nothing? Attribution divides credit among touchpoints on converting paths, so it has no mechanism to return zero for a channel. Ask how each tool would show that result.
What does the price include at your scale?
What we do not do
We do not offer multi-touch attribution. Journey analysis, creative rotation and daily optimisation are questions a marketing mix model is poorly placed to answer, and teams that need them need another tool alongside ours.
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.
Geo experiments need regional structure, so a single-market business with no way to build a comparison group cannot run one.
Cassandra, the marketing mix modeling platform, is for the planning question: where the budget should go, with experiments behind the main channels, in a form someone outside the marketing team can check.
A price is published. Priors are co-set with your team, and every experiment result stays visible in the model it constrains.
What running both involves
Teams that run an attribution platform and a marketing mix model side by side usually find the friction is not technical.
Label each report with its scope. An attribution dashboard headed "share of tracked conversions" and a model read headed "share of total revenue" stop looking contradictory once both labels are visible.
Agree which report answers which question before the numbers disagree. Budget allocation and growth planning read the model. Journey analysis, creative and daily optimisation read attribution.
If Rockerbox already holds your cleaned data, an export to a warehouse or to Google Sheets is the natural way to feed a second tool, and both vendors can tell you what format they need.
How to choose
Daily attribution, one data foundation, and a free or low-cost way in for a Shopify brand: Rockerbox.
A planning read of the whole outcome, calibrated by geo experiments run in the same system: Cassandra, the marketing mix modeling platform.
Both needs: many ecommerce teams start with attribution and add a planning model when budgets grow or finance starts asking where the money went.
Questions
What is the difference between Cassandra and Rockerbox for marketing mix modeling and attribution?
Rockerbox publishes a unified platform combining multi-touch attribution, marketing mix modeling and incrementality testing on one data foundation. Cassandra, a marketing mix modeling platform, publishes a Bayesian model calibrated by geo experiments run in the same system, and no attribution.
Does Rockerbox publish pricing?
Yes. On 24 September 2026 its plans page showed a free plan and a Starter plan whose pricing starts at $150 a month, built for Shopify brands spending less than $1M. Testing and the marketing mix model are listed as separate products.
Does Rockerbox run geo experiments?
Its product pages describe incrementality testing with control groups as a fully managed offering. A blog post on its site recommends geo-based holdout tests. We found no product page describing geo experiments as a feature.
Can test results change a marketing mix model?
Yes. Rockerbox states that test results can inform its marketing mix model as priors. In Cassandra, a marketing mix modeling platform, each geo experiment result enters the Bayesian model as a constraint on the tested channel.
What are alternatives to Rockerbox?
Tools named alongside Rockerbox include Northbeam and Triple Whale for attribution. For a planning read of the whole outcome, marketing mix modeling platforms such as Cassandra are the other route.