When a run completes, the Run Results page brings together an AI evaluation report, every model output, and your options for next steps. All outputs are generated by Robyn, Meta’s open-source MMM framework — giving you full confidence in the methodology behind each number. This guide explains what you are looking at and how to get the most from the results.
The AI Evaluation Report
The first thing you see at the top of the results page is the AI Evaluation Report — a structured analysis generated automatically by MMM Pilot at the end of every successful run.
What the Report Covers
| Section | What It Tells You |
|---|---|
| Model Fit Assessment | How well the model fits your data. Covers NRMSE and R² statistics. Flags overfitting or underfitting concerns in plain language. |
| Contribution Breakdown Review | Whether each channel’s attributed contribution is plausible given real spend levels. Unusual results are flagged explicitly. |
| Spend Efficiency Analysis | Assessment of curve shape and diminishing returns per channel. Identifies channels that may be over- or under-saturated. |
| Budget Recommendations | Summary of the budget allocator output — which channels have the strongest marginal returns and where reallocation could improve efficiency. |
| Overall Assessment | A plain-language verdict on the model’s reliability and readiness for decision-making. |
Using the Evaluation
- A strong Model Fit Assessment (low NRMSE, high R²) means the model explains your data well.
- Review the Contribution Breakdown Review for any flagged implausibilities — these are your first signal to investigate or re-run with adjusted configuration.
- The Overall Assessment is your quick read — if it’s positive, you can proceed to presenting and acting on the results.
Run Metadata
Below the AI evaluation report, the sidebar and header show:
- Run Name — The name you assigned when configuring the run.
- Status — Completed, Failed, Processing, or Queued.
- Date & Duration — When the run completed and how long it took.
- Credits Used — Credits consumed for this run, based on actual processing time (1 credit = 15 minutes).
- Configuration — Summary of key settings (trials, iterations, time window, number of paid media channels).
Artifact Categories
Model Evaluation
High-level outputs for assessing model quality and selecting the best solution.
| Artifact | Description |
|---|---|
| Pareto Front | The trade-off between model decomposition accuracy and Business Outcome (NRMSE). Use this to identify the best-fit solutions. |
| One-Pager Summary | A concise executive summary of the model’s key findings — channel contributions, spend efficiency, and budget recommendations. |
| Model Fit Statistics | Numerical summary of NRMSE, R², and decomposition quality for each solution. |
Plots
Every plot includes an AI-generated description in plain language — readable by non-technical stakeholders without statistical background.
| Plot | What It Shows |
|---|---|
| Contribution Breakdown | Attribution of your outcome variable by channel — paid media, organic, context, trend, and seasonality. |
| Waterfall Chart | How each component builds up to explain total outcome for a selected period. |
| Spend Efficiency Curves | Diminishing-returns curves per paid media channel, showing how incremental spend translates to incremental outcome. |
| Fitted vs. Actual | How closely the model’s predictions match real historical data — a visual check of model fit quality. |
| Carry-Over Effect Transformations | Visual representation of the modelled decay effect for each paid media channel. |
Solution Files
Downloadable CSV files for further analysis in spreadsheet tools or BI platforms.
| File | What It Contains |
|---|---|
| Contribution Breakdown CSV | Daily or weekly contribution values per channel across the full time window. |
| Budget Allocator Output | Recommended spend allocation across channels for your specified budget. |
| Raw Model Output | Full model solution data including hyperparameter values, fit statistics, and decomposition figures. |
Available Actions
| Action | Description |
|---|---|
| Download Artifacts | Download individual artifacts or all outputs as a zip archive. |
| Duplicate Run | Open the Run Configuration dialog pre-populated with this run’s settings. Useful for small iterative adjustments. |
| Run with Suggested Hyperparameters | If the AI evaluation generated hyperparameter refinements, this button appears here. Pre-populates a new run with AI-recommended bounds for your next iteration. |
Comparing Across Runs
The runs sidebar on the project page lists every run in chronological order. Use it to:
- Switch between runs and compare AI evaluation summaries side by side.
- Track improvement from iteration to iteration.
- Return to an earlier configuration if a later iteration performs worse.
All run configurations and results are permanently stored — no runs are automatically purged.