-1.png?width=1132&height=292&name=Asset%201@2x%20(12)-1.png "Odins.ai Logo")
Build vs buy
Google Meridian: build or buy?
Meridian is free, Bayesian and good. So should you build marketing mix modeling in-house on it, or buy a managed platform? The math is rarely the hard part. The work sits in collecting the data, integrating it and setting the priors. Here is what each of those takes, both ways.
Reviewed September 2026 · sources cited on this page
Odins
Buy: a managed marketing investment platform
Build on Meridian
Open-source framework, run by your team
At a glance
Build or buy, side by side.
The short version of the whole page. Details and sources below.
Framework facts: Google Meridian documentation and changelog (version 2.0, September 2026), and Runge and Pauwels (2026), Open-Source Media and Marketing Mix Modeling, SSRN 6317979, packages as of November 2025.
The model is the small part. The data is the project.
Meridian starts when the data is ready. Getting it ready is most of the work in an in-house build, and it never ends. This is what your team takes on when you build.
- A pipeline per ad platform, kept alive as APIs and account structures change
- Offline spend collected from agencies and broadcasters: invoices for cost, spot reports for timing, at least two years back
- One consistent hierarchy of markets, products and channels, so last year's campaign names still map
- Sales, pricing and promotion data joined to the same weekly grain
- When you buy, that is the service: managed integrations plus an offline pipeline, delivered as one structured dataset you can also send to your own warehouse
Priors decide the model. Defaults do not know your business.
In a single-country market you have one weekly observation against ten or more channels. The data cannot identify everything on its own, so the priors carry much of the load. Meridian ships uniform defaults, and an academic review rates the misuse risk from them as high. Setting good priors is a judgment call someone has to own.
- Build: your team decides every prior, or accepts defaults that treat all channels alike
- Build: Meridian 2.0 can calibrate priors from geo experiments, if your market is large enough to run them
- Buy: priors set in 2 to 3 working sessions, from your spend history, saturation signals in the digital channels and structured interviews
- Buy: what you learned from earlier MMM work and tests goes into those sessions
- Either way, ask the same question of every prior: where did it come from?
Channel contribution, 80% ranges
from the posteriorA tight range is a decision. A wide range is a test plan, with cost boundaries.
The choice
When to build, when to buy.
Both are honest answers to the same problem. The deciding factor is your team and your data, not the math.
Build on Meridian if
- You have 2 to 3 data scientists who can own the model long term, and data engineering to feed it
- You have geo-level data and a mix that is mostly digital
- Model building is a capability your company wants in-house
Buy if
- You want the data collected, structured and kept right for you, offline included
- Your market is a single country, where disciplined priors matter more than data volume
- The marketing team should act on results in weeks, without a data science layer in between
The honest overlap
Teams like Bolt and HelloFresh run open-source MMM in-house with large data science organizations. If that is you, building can be the right call. And if you have already started, nothing is wasted: the data work transfers, and what you learned goes into the prior sessions.
Frequently asked questions.
Should we build MMM in-house with Google Meridian or buy a platform?
Build if you have 2 to 3 data scientists who can own the model for years, data engineers to feed it, and geo-level data. Buy if you want results in weeks, have offline media, or work from one national time series where priors decide the outcome. The license price is the smallest number in the calculation. Salaries, months of build time and permanent maintenance are the real cost.
What are the options besides building on Meridian?
Three groups. Other open frameworks: PyMC-Marketing for flexibility, Meta Robyn for a fast first look. Managed platforms such as Odins, where data, model and recommendations are run for you. Enterprise vendors such as Analytic Partners and Kantar for global programs. Our guide to the best MMM providers sorts them by fit.
What does it take to collect the data for Meridian?
Meridian has no connectors. You build a pipeline for each ad platform, collect offline spend from agencies and broadcasters, join it to sales, pricing and promotions, and keep all of it consistent week after week. We have watched well-resourced companies build in-house and slip on timelines. With Odins, data integration is part of the service.
How are priors set in Google Meridian?
Meridian ships with default uniform priors. That makes it easy to start and easy to misuse: a 2026 academic review (Runge and Pauwels) rates the risk as high. You can and should set your own. Since version 2.0, Meridian can also calibrate priors from incrementality experiments such as GeoX. Either route puts the judgment calls on your team.
How does Odins set priors?
Before the model sees data, in 2 to 3 sessions with your team. We use three sources: your historical spend, saturation signals from the digital channels, and structured interviews about how your channels work. That is what keeps a single national time series stable.
Robyn, Meridian or a commercial MMM platform for a mid-sized company without a big data team?
Without a data team, a managed platform. Robyn is the fastest first look, but it is frequentist and no longer actively developed by Meta. Meridian is the most packaged Bayesian framework and still needs data scientists. A managed platform delivers the data, the model and the recommendation without the hires.
We already started building on Meridian. Is that wasted?
No. The data work transfers directly, and what your team learned makes the prior sessions better. Some teams also run Odins alongside an internal model as a benchmark.
Is Odins built on Google Meridian?
No. Odins runs its own Bayesian MMM stack, built on the same statistical principles Meridian uses. The feature-by-feature comparison is on Odins vs Google Meridian.
Marketing spend, modeled like an investment.
A 30-minute walkthrough of what the managed version would look like on your channels: the data, the priors and the verification.
Book a demo