MMM Pilot is powered by an industry-proven statistical modeling engine. The platform wraps it with a guided interface, AI-assisted configuration, and cloud-based execution — so you build rigorous Marketing Mix Models without writing code or managing infrastructure.
The Modeling Engine
MMM Pilot is powered by Robyn — Meta’s open-source Marketing Mix Modeling framework, widely used by practitioners and researchers across the MMM community. Robyn uses ridge regression with geometric Carry-Over Effect transformation and hill-function saturation to decompose the contribution of each marketing channel to your chosen business outcome.
What the Model Produces
- Contribution Breakdown — Attribution of your outcome variable across paid media, organic channels, context variables, trend, and seasonality. See exactly what drives your results.
- Spend Efficiency Curves — Diminishing-returns curves for each paid media channel, showing how incremental spend translates to incremental outcome.
- Budget Allocator — Optimal spend allocation recommendations across channels for any given total budget.
- Model Diagnostics — Fit statistics including NRMSE, R², and decomposition accuracy, plus Pareto front visualisation to help you select the best-fitting solution.
AI-Assisted Configuration
Setting up an MMM run typically requires deep statistical knowledge. MMM Pilot’s AI layer removes that barrier with automated suggestions at every step.
Field Mapping
When you open the New Run dialog:
- The platform presents all imported fields across your data sources.
- You assign each field a role:
- Dependent Variable — The outcome you want to explain (e.g., revenue, conversions).
- Paid Media — Channels with controllable spend (e.g., Google Ads, Meta Ads).
- Organic — Non-paid channels (e.g., SEO traffic, email opens).
- Context — External variables (e.g., economic indicators, seasonality indices).
- Click AI Suggest to let the platform automatically propose a role for every field — based on field names and patterns — so you do not have to classify each one manually.
Hyperparameter Configuration
Each paid media field requires hyperparameter bounds that control the shape of Carry-Over Effect and Spend Efficiency curves.
- Default values are provided out of the box and work for most analyses.
- AI Suggest Hyperparameters analyses your data characteristics and recommends tailored bounds.
- After a run completes, the AI evaluation can suggest refined hyperparameters for your next iteration.
Time Window
Narrow your analysis to a specific date range within the available data. The dialog shows:
- The earliest and latest available dates from your Organic Base data source.
- The number of data points within the selected window.
- A warning if the window is too short for reliable modeling.
Model Parameters
- Trials — The number of independent optimisation trials the engine runs. More trials means a broader search for the best solution, and more compute time.
- Iterations per Trial — The number of iterations within each trial. Higher values refine the search further.
- Increasing both improves model quality — and also increases estimated processing time and credit consumption.
Cloud-Based Execution
Your analysis runs on dedicated cloud infrastructure, not in your browser. Start a run, close the tab, and come back when it’s done.
How It Works
- You configure the run and click Start Run.
- The platform validates your configuration — sufficient data points, no field conflicts, enough credits.
- The estimated credit cost is calculated (1 credit = 15 minutes of estimated processing time, minimum 1 credit). The run is queued.
- A dedicated cloud environment loads your data, executes the model, generates all outputs, and uploads the results.
- The UI shows real-time status updates: Queued → Processing → Completed. You receive an in-app notification when the run completes.
Processing Time
- A typical run with 5,000 iterations across 5 trials takes approximately 5–15 minutes, depending on the number of channels and data points.
- Credits are deducted based on actual processing time upon successful completion. Failed runs are not charged.
- Close the browser and return later — your results will be waiting.
Output Artifacts
Every completed run produces a full set of downloadable artifacts.
| Category | What’s Included |
|---|---|
| Model Evaluation | Pareto front, one-pager summary, model fit statistics |
| Plots | Spend Efficiency curves, Contribution Breakdown, waterfall chart, fitted vs actual, Carry-Over Effect transformations |
| Solution Files | CSV exports of Contribution Breakdown data, channel contributions, and budget allocator outputs |
| Diagnostics | Additional diagnostic outputs for deeper investigation |
All artifacts are viewable in the browser and downloadable individually or in bulk.