Open-source MMM is giving marketers a credible alternative to black-box attribution. Here's what the shift means for measurement strategy in Southeast Asia.
Measurement has always been marketing’s uncomfortable back room — the place where assumptions go to become spreadsheets, and spreadsheets go to become budget decisions nobody can fully defend. For the past decade, multi-touch attribution filled that room with confident-sounding numbers that were, in hindsight, largely a function of whoever owned the last cookie before conversion. That era is closing. What’s replacing it is older, stranger, and considerably more honest: marketing mix modeling, now arriving in open-source form.
Why Open-Source MMM Is Having Its Moment Now
AdExchanger’s Allison Schiff put it plainly: MMM is a very old idea undergoing a very timely reinvention. The convergence driving this is structural. Cookie deprecation — even at its stuttering, delayed pace — has methodically eroded the user-level signal that MTA depended on. Clean rooms help, but they require bilateral trust and data maturity that most brand-agency relationships don’t yet have. Incrementality testing is rigorous but slow. Into this measurement vacuum, open-source MMM frameworks like Meta’s Robyn, Google’s Meridian, and PyMC-Marketing have stepped in, offering something the black-box vendor alternatives never could: auditability.
When your CMO asks why the model is recommending a 15% budget shift from paid social to offline, you can show your working. For marketing directors in markets like Indonesia or Thailand — where media mix genuinely spans TikTok Shop, Shopee Ads, LINE OA, and outdoor simultaneously — that transparency isn’t just intellectually satisfying. It’s operationally necessary.
The Acronym Problem Is Real, But Don’t Let It Distract You
The community is already generating sub-variants: LMMM (Lightweight MMM), BMMM (Bayesian MMM), and various proprietary flavors from consultancies who’ve spotted the opportunity. Schiff’s aside about apologizing for the new acronym is a knowing wink at an industry that loves to rebadge fundamentals as innovations. Don’t get lost in the taxonomy.
What matters strategically is the methodological shift underneath it: Bayesian MMM in particular introduces prior knowledge — seasonality curves, known media saturation thresholds, historical elasticity estimates — directly into the model structure, rather than treating every run as if history never happened. For Southeast Asian markets with sharp seasonal spikes (Harbolnas in Indonesia, 9.9 and 11.11 across the region, Songkran in Thailand), this is not a minor technical detail. It’s the difference between a model that understands your market and one that perpetually learns it from scratch.
Implementation note: Bayesian models are computationally heavier. Teams running on modest cloud infrastructure should budget for GPU-accelerated instances during model training runs, particularly with weekly granularity across 8–12 media channels.
What This Means for Your Attribution Stack Right Now
The honest answer for most marketing teams is: MMM doesn’t replace your existing attribution infrastructure immediately, it sits alongside it. The practical workflow that’s emerging looks like this — use last-touch or data-driven attribution for tactical, in-flight optimization decisions (bid adjustments, creative rotation, channel pacing), and run MMM quarterly or bi-monthly for strategic budget allocation and scenario planning.
Sportradar’s contribution to the Digiday conversation about sports marketing is a useful illustration of why this matters. Live sport is one of the few remaining contexts where attention is genuinely undivided and brand exposure carries real weight beyond a click. But the activation challenge is precisely that sports sponsorship ROI is notoriously difficult to attribute through standard digital measurement. MMM, with its ability to model offline media contributions alongside digital, is arguably the most credible methodology for a brand trying to understand what their jersey sponsorship at a BNI Indonesian Liga 1 match is actually worth relative to their Shopee Ads spend.
For teams starting from zero: Meta’s Robyn has the lowest barrier to entry given its R-based implementation and strong community documentation. Google’s Meridian is newer and Python-native, which may suit data science teams already embedded in Python workflows. Neither requires surrendering your data to a vendor.
The Organizational Shift Nobody’s Talking About
Open-source MMM creates a measurement capability that lives inside your organization rather than inside a vendor contract. That’s strategically significant in ways that go beyond cost. When the model lives with your team, institutional knowledge about why certain priors were set, which channels were excluded in a given market, how carryover effects were calibrated — all of that accumulates inside the organization rather than evaporating when you switch tools.
The friction point is talent. Running a credible MMM practice requires either a data scientist with econometrics exposure or a very capable analytics partner who will genuinely transfer knowledge rather than guard it. In Southeast Asia’s current talent market, the former is expensive and competitive; the latter requires careful vendor selection. Either way, the investment is front-loaded.
The longer-term payoff is a measurement posture that doesn’t require renegotiation every time a platform changes its attribution window — which, as the past three years have demonstrated, is roughly every eighteen months.
Key Takeaways
- Open-source MMM frameworks like Robyn and Meridian give marketing teams vendor-independent measurement they can interrogate, audit, and own — critical as cookie-based attribution continues to erode.
- Bayesian MMM approaches are particularly well-suited to Southeast Asian market conditions, where sharp seasonal peaks and complex multi-platform media mixes make historical priors genuinely valuable rather than just statistically convenient.
- The practical implementation path is MMM for strategic quarterly allocation, combined with existing digital attribution for tactical in-flight decisions — not a replacement, a complement.
The harder question underneath all of this: if your marketing mix model is now open, auditable, and internally owned, what does that do to the relationship between your team and your media agencies? When the model shows that a channel your agency buys heavily is underdelivering on contribution, the conversation that follows will be different from the one you’ve been having. Is your organization ready for measurement that doesn’t need to be negotiated?
At grzzly, we work with marketing and growth teams across Southeast Asia who are navigating exactly this transition — building measurement infrastructure that’s both technically credible and organizationally durable, without getting sold a black box in the process. If you’re trying to figure out where MMM fits in your current stack, or how to build the internal case for the investment, we’re happy to think through it with you. Let’s talk
Sources
Written by
Rogue GrizzlyOperating at the contested frontier of cookieless targeting, clean rooms, and identity resolution. Comfortable where the infrastructure is shifting and the playbooks have not yet been written.