If you've spent any time looking into Marketing Mix Modelling, you've probably noticed the comparisons are a bit of a mess. Some lists put a free open-source library next to a £150k consulting engagement and call them alternatives. They aren't, and pretending otherwise is the reason a lot of buyers end up with the wrong shape of solution.
This is a buyer's view of the MMM tooling landscape in 2026, written by someone who runs MMM for a living. I'll be honest about what each option does well, where it falls down, and the question to answer before you pick anything.
TL;DR
The MMM market splits into four real categories: open-source frameworks, modern commercial platforms, MMP-led modules, and traditional consultancies. The choice has less to do with software and more to do with how much internal capability you have, how much you spend on media, and who owns the model after the project ends.
A note up front. Hanya, the programme I run, is built on top of Meta's open-source Robyn framework. So when I talk about Robyn I'm talking about software I work with every week. I've kept the comparison even-handed, but you should know that going in.
Two things people call “MMM tools”
Most of the confusion in this space comes from collapsing two very different things into one category.
The first is the modelling engine. This is the statistical software that takes clean, prepared data and produces an estimate of how each marketing channel contributed to sales. Meta's Robyn, Google's Meridian, LightweightMMM, PyMC Marketing, Recast, and the modelling layers inside platforms like Mass Analytics all sit here. Some are free. Some are commercial. They all do roughly the same job at the core.
The second is the delivery model. This is who does the work of getting the data ready, building the model, interpreting it, presenting it, and updating it. Open-source frameworks deliver none of this on their own. Commercial platforms deliver some. Full-service consultancies deliver all of it but charge accordingly. Productised offers (Hanya included) sit between those poles.
Confuse the two and you'll buy a tool when you needed a partner, or hire a partner when you wanted control. Get the distinction right and the rest of this guide will make more sense.
Open-source frameworks
Open-source MMM has matured a lot in the last five years. The frameworks below are genuinely capable. The catch is rarely the modelling itself, it's everything around it.
Meta's Robyn
Robyn was released by Meta in 2020 and has become the most widely-used open-source MMM framework in the world. It's written in R, with a Python wrapper. It uses ridge regression with hyperparameter tuning via Nevergrad, and it's geared toward digital-heavy media plans where you have weekly granularity across multiple channels. The repo lives at github.com/facebookexperimental/Robyn.
What makes Robyn good is the community and the maturity. The documentation is solid, the examples are realistic, and the GitHub repository is genuinely active. If you have a competent data scientist on staff you can get a first model running in a week or two of focused work, assuming the data is already in shape.
The honest catch: that “assuming the data is already in shape” phrase is doing a lot of work. The framework will not help you stitch together spend exports from six ad platforms, reconcile naming conventions with your CRM, or back-fill missing weeks of TV impressions. That work is yours, and on a real client problem it tends to be eight to ten weeks of effort before Robyn does anything useful.
Best for: teams with strong data engineering and a data scientist comfortable in R.
Google's Meridian
Meridian is Google's Bayesian MMM framework, released in 2024. It's written in Python and TensorFlow Probability, supports geo-level modelling out of the box, and integrates more easily than Robyn does with Google Ads and Search Ads 360 data.
The Bayesian approach gives you proper uncertainty intervals on every channel coefficient, which matters when you're presenting results to a sceptical board. The flip side is that Bayesian models are slower to fit and demand more statistical fluency to set up, validate and explain. If your data scientist is comfortable with priors, posteriors and convergence diagnostics, great. If not, expect a longer ramp.
Best for: Google-heavy media mixes, teams that need geo-level reads, and analysts comfortable with Bayesian methods.
LightweightMMM
LightweightMMM is also from Google, released earlier than Meridian and aimed at being, as the name suggests, lighter. It's a Bayesian library built in JAX and NumPyro. Google has effectively positioned Meridian as its successor for new builds, but Lightweight is still in use, particularly where teams want a simpler library to extend themselves.
The codebase is smaller and easier to read than Meridian's, which some practitioners prefer. Documentation and active development have slowed since Meridian launched.
Best for: teams who want a minimal library to fork and customise, or who started on it and have no reason to migrate.
PyMC Marketing
PyMC Marketing is a marketing-focused layer on top of the PyMC Bayesian library. It's the most flexible of the open-source options, in the sense that you can model almost anything, but that flexibility comes at a cost in setup time and required expertise.
If you have an in-house Bayesian statistician and an unusual modelling problem, PyMC Marketing is hard to beat. For everyone else it's overkill.
Best for: research-grade work, custom problem shapes, teams with serious Bayesian chops.
Commercial MMM platforms
Commercial platforms are where the picture gets more varied. Some are full-stack tools you log into. Some are wrappers around open-source frameworks. Some are essentially software-enabled consultancies. The label “platform” hides a lot of variation, so look closely at what's actually being delivered.
Recast
Recast is one of the better-known modern MMM platforms, founded by ex-data-science leads from larger brands. It's a SaaS product with a Bayesian modelling engine and a UI for exploring results. They market themselves on speed of refreshes and integration with marketing data sources.
Recast suits in-house marketing analytics teams who want a vendor-supported model without commissioning a full consulting engagement. Pricing typically starts in the low six figures annually.
Mass Analytics
Mass Analytics has been in MMM for longer than most. They sell software (MassTer) and services. The software is genuinely capable in the hands of experienced practitioners but has a steeper learning curve than newer platforms. They're a reasonable fit for in-house analytics teams in larger organisations who want to bring MMM in-house and still have vendor backup.
Mutiny
Mutiny is newer, with an emphasis on automation and ease of use. The trade-off, as with most easy-to-use tools in this space, is reduced visibility into what the model is actually doing. For teams without a data scientist this can be the right trade. For teams with one, it can grate.
Analytic Partners, Nielsen, Ekimetrics
These three sit in the traditional consultancy bracket. They're proprietary, end-to-end, and priced accordingly (usually £100k to £250k+ per engagement). The methodology lives inside the firm, which means you depend on them for refreshes and interpretation. They're strong on stakeholder management, which sounds woolly until you've watched a finance director try to swallow an MMM result that contradicts their priors.
Their main weakness is the same as their main strength: you don't own the model, and you can't run it without them. When the engagement ends, the asset stops working.
DIY and spreadsheet approaches
A surprising number of brands still try to do MMM in Excel. With enough patience you can fit a basic linear regression in a spreadsheet, and for very small media budgets that may even be defensible.
The problems show up quickly though. You can't model adstock or saturation curves without leaving the spreadsheet. You can't get uncertainty estimates. Validating against holdout data is fiddly, and every refresh is hand-crafted, which means models drift fast.
For brands spending under £500k a year on media and running two or three channels, a spreadsheet model alongside a sensible attribution view might be enough. Above that, the tooling gap shows.
MMP plus MMM hybrids
The Mobile Measurement Partners (MMPs), Appsflyer, Adjust, Singular and Branch, have all added MMM modules to their platforms in the last couple of years. The pitch is that you can do attribution and MMM in one place.
For app-first businesses where the MMP is already wired into every channel, this is genuinely convenient. The MMM module reuses the spend and event data the MMP is already collecting, which removes a lot of the data-prep pain. For brands without an existing MMP relationship, or with significant offline channels, the case is weaker. The MMM module is usually competent but not best-in-class, and the focus on app channels can leave gaps for brands with a mixed media mix.
How the categories compare
| Category | Cost | Technical bar | Time to first model | Transparency | Support model |
|---|---|---|---|---|---|
| Open-source (Robyn, Meridian, etc.) | Free software, plus internal time | High; in-house data scientist required | 8–12 weeks (mostly data prep) | Full code visibility | Community forums |
| Modern SaaS platforms (Recast, Mutiny) | Low six figures / year | Moderate; analyst can run it | 4–8 weeks | Partial; methodology documented | Vendor support team |
| Established platforms (Mass Analytics) | Mid six figures / year | High; trained user required | 6–10 weeks | Partial; documented methodology | Dedicated account team |
| Productised offers (Hanya) | £25k–£150k per build | Low; programme handles delivery | 12 weeks, fixed | Built on open-source; code shareable | Named delivery lead |
| Full-service consultancies | £100k–£250k per engagement | None required from client | 4–6 months | Proprietary; hard to inspect | Project team for the duration |
| MMP+MMM hybrids (Appsflyer, Adjust) | Add-on to MMP contract | Low to moderate | Fast for app-first brands | Partial; tied to MMP data | Vendor support team |
| DIY spreadsheet | Cost of your time | Statistical literacy required | Variable; refreshes are manual | Full | None |
A couple of patterns are worth pulling out of that table.
Time to first model is the variable buyers most often underestimate. With open-source frameworks the modelling itself is the fast bit. The data work is the slow bit, and it's almost always six to ten weeks of effort regardless of which framework you pick. Commercial platforms can shave that down with pre-built data connectors. Productised offers compress it further by automating the data preparation step. Traditional consultancies take the longest in calendar time because of how their engagements are paced, even though they have the most capacity to throw at the problem.
Transparency is where the open-source options pull ahead. You can read the code. You can audit what the model is doing. You can show a data scientist exactly which assumption you'd like to challenge. With proprietary platforms and consultancies you're trusting the methodology, often without a meaningful way to inspect it. The Marketing Accountability Standards Board (MASB) has been pushing for more transparency in marketing measurement for years, and open-source frameworks are partly a response to that pressure.
How to choose
The honest answer is that the right tool depends on a few things: how much you're spending on media, how much internal capability you have, and how much certainty you need before you act.
If you're spending under £500k a year on media, MMM probably isn't the right call yet. Get attribution clean, run some incrementality tests, and revisit when your media budget grows. There's a fuller breakdown of when MMM starts to pay back in how much does MMM cost.
If you're spending between £1m and £5m and you have a data team, an open-source framework with a partner to handle the data layer is usually the best balance of cost and control. This is the gap Hanya is built for.
If you're spending over £5m and you want a fully outsourced engagement, a traditional consultancy is the path of least resistance, with the trade-offs flagged above.
If you're spending over £5m and you want to bring MMM in-house, a commercial platform like Recast or Mass Analytics is worth a look, ideally with a one-off engagement to set the model up and train the team.
Questions worth asking any vendor
- Is the methodology open or proprietary, and can we see what's under the hood?
- Who owns the model after the engagement, and can we run it without you?
- How long does the data preparation step really take in weeks, not in marketing copy?
- What does a refresh cost and how long does it take, and what's included in the price?
- What's your definition of incrementality, and can you prove it on our data with a holdout test?
If a vendor can't give you straight answers to those, walk away.
The bigger picture
Choosing a tool is a smaller decision than people make it out to be. The bigger questions are about your data, your team, and what you'll do with the answers once you have them. A perfect Robyn model that no one acts on is worth less than a rough Excel model that changes a budget meeting.
Whichever tool you pick, plan for the soft stuff. Get stakeholders aligned on what MMM can and can't tell them. Decide in advance what level of evidence will move budget, and pick someone who'll own the model internally once the build is done. The tool is the easy bit. There's more on this side of MMM in Beyond Attribution and the complete guide to Marketing Mix Modelling.