Last-click attribution stopped being useful a decade ago. We build Marketing Mix Models — open-source Robyn or custom Bayesian — that quantify the real incremental contribution of each channel, so your budget decisions stop being based on platform-reported ROAS and start being based on truth.
Scope this out →Channel-level incremental ROI, not platform-reported ROAS
A defensible answer to "what happens if we cut Meta by 30%"
Budget decisions tied to real contribution, not last-click noise
Minimum 18 months of weekly data across all channels, ideally 24 months. Fewer data points than that and the model can't reliably separate channel effects from seasonality and external variables. If your account is younger, we may recommend a hybrid approach — directional MMM plus geo holdout tests — instead of a full model.
Depends on the account. Robyn (Meta's open-source MMM) is our default — it's battle-tested, well-documented, and integrates cleanly with BigQuery. LightweightMMM (Google's) for specific cases where Bayesian uncertainty intervals matter. Custom Bayesian models only for very large accounts with unusual structural requirements.
Calibration against real experiments. We run geo holdout tests — pausing a channel in one set of regions while keeping it live in controls — and compare the measured lift against the model's prediction. If they disagree meaningfully, the model needs retuning. Without calibration, MMM is just expensive correlation.
Three main things. First, reallocate budget to channels with higher true incrementality. Second, set defensible ROAS or CPA targets that account for each channel's incrementality, not just platform-reported performance. Third, model scenarios — "what happens to revenue if we cut Meta by 30% and reallocate to Google Shopping?" — before making the decision.
Quarterly at minimum, monthly for fast-moving accounts. Market conditions, creative, seasonality, and competitor behavior all shift channel coefficients over time. A stale model quietly becomes wrong — and MMM users often don't notice until a budget decision underperforms badly.
No — complement. Platform attribution answers "which campaigns and creatives drove attributed conversions?" at a tactical level. MMM answers "how much of the company's revenue is actually caused by each marketing channel?" at a strategic level. You need both, and you need to know which question each is answering.
Tell us about the account. We'll tell you honestly what we'd do.
Start a project →