Media mix modeling software
Open-source libraries, subscription vendors, and a $249 Mac app. Who each one is for, what it costs, and what you have to take on faith.
01 / The short version
There are three ways to get a media mix model today. You can pay a measurement vendor a subscription, you can stand up an open-source library with a data scientist, or you can buy a finished instrument and run it on your own machine. Each one is the right answer for someone. This page lays out who each is for, what it costs, and what you have to trust.
SignalRig sits in the third lane. We built it, so read the table with that in mind, and check every claim against the vendors' own pages. We link the open-source projects directly.
02 / Side by side
| Option | What you pay | Where your data goes | Who has to run it | Can you inspect the model? | Best for |
|---|---|---|---|---|---|
| SignalRig for Mac | $249 once | Stays on your Mac. No account, no cloud, no telemetry in the store build. | A marketer. Drop in your ad platform exports, read the verdict. | Yes. The model core is open source (Apache-2.0) and the app shows calibration and holdout grades for every fit. | Solo marketers, consultants, and small agency teams who need a defensible number without a data team. |
| Meta Robyn | Free (open source, R) | Wherever you run R. Typically your own machine or cloud. | Someone comfortable in R with time to tune hyperparameters and read diagnostics. | Yes, fully. It is code. | Teams with an analyst who can own the pipeline and explain the output. |
| Google Meridian | Free (open source, Python) | Wherever you run Python. Google-provided notebooks assume Colab or your own environment. | A Python-literate analyst comfortable with Bayesian models. | Yes, fully. It is code. | Teams with data-science support, especially those wanting Google's reach and frequency priors. |
| PyMC-Marketing | Free (open source, Python) | Wherever you run Python. | A Bayesian-literate analyst. | Yes, fully. It is code. | Teams that already use PyMC and want the most flexible model specification. |
| Measurement SaaS vendors | Subscription, commonly quoted in the low thousands per month and up, usually annual contracts | Uploaded to the vendor's cloud. | The vendor. You get a dashboard and a customer success contact. | Usually no. The model is the product and is rarely exposed in full. | Mid-size and enterprise brands that want managed service and are comfortable with a black box. |
| Agency or consultancy build | Project fee, typically five figures and up, plus refresh fees | The consultancy's environment. | Them. You get a deck. | Sometimes, on request, at the level of a methodology appendix. | Brands that need one big decision defended once, with expert narration. |
Vendor pricing varies by contract and is rarely published. Where we give a range it reflects publicly discussed figures as of 2026; confirm with the vendor. Open-source projects: Robyn, Meridian, PyMC-Marketing, SignalRig model core.
03 / What actually separates them
Robyn, Meridian, and PyMC-Marketing are transparent in the sense that every line is readable. That is worth a lot. But readable code does not tell a marketer whether a particular fit can be trusted. For that you need the model to grade itself: recover planted truths on synthetic data, predict weeks it never saw, and say plainly when the chains did not converge. SignalRig treats those grades as the product. A fit that fails its exam locks the budget recommendations instead of printing them.
An open-source MMM is free the way a kit car is free. The library costs nothing; the analyst who assembles it, validates it, and explains it to a client costs real money every month. If you already have that person, use the libraries. If you do not, the honest comparison is between a subscription vendor and a finished instrument, and that comparison is mostly about price and where your data lives.
Every SaaS option requires uploading weekly spend and outcome data for every client to a third party. For an agency that is a contractual question as much as a technical one. SignalRig's store build makes no network connections and sends nothing anywhere. CSVs are read from disk and the model runs on the Mac in front of you.
It is a single-machine tool with an opinionated model: geometric adstock, Hill saturation, a Bayesian fit with hard convergence gates. It will not run a hundred brands overnight, it has no API, and it does not pretend a thin dataset supports a confident answer. If your program has fewer than about a year of weekly data across your channels, the validator will tell you that before you spend an hour on it.
04 / See it before you buy it
The live demo walks a fully worked model end to end in the browser, no install. The Mac app adds the data validator, the graded exam on your own CSVs, and the media-plan export. $249 once. Requires macOS 13 or later.