MMM Pilot™

Reading the "ROAS Convergence" Chart: Is Your Model Stable?

The “Seismograph” of Your Analysis

When MMM Pilot (via Robyn) analyzes your data, it doesn’t just calculate the answer once. It runs thousands of trials, constantly tweaking the math to see what works best.

The ROAS_convergence1.png chart documents this journey. It shows you how the model’s opinion about your channels changed over time.

Think of this chart as a monitor of stability. You use it to answer one critical question: “Did the model finish thinking, or did I interrupt it while it was still guessing?”

1. The Axes: Time vs. Value

  • X-Axis (Trials/Iterations): This represents the timeline of the analysis.
    • Left: The beginning of the run (The model is guessing wildly).
    • Right: The end of the run (The model should be confident).
  • Y-Axis (Mean ROAS): This is the calculated Return on Ad Spend for your channels.

2. How to Read the Lines

Each colored line represents a specific marketing channel (e.g., Facebook, Google Ads, TV).

  • The “Chaos” Phase (The Left Side): You will notice the lines jumping up and down aggressively at the beginning. This is normal. The AI is exploring extreme possibilities—testing what happens if Facebook ROAS is 0.5 and what happens if it is 10.0.
  • The “Convergence” Phase (The Right Side): As you move to the right, the lines should start to flatten out. This is what data scientists call Convergence. It means the model has stopped guessing and has “settled” on a specific number.

3. Good vs. Bad Signals

✅ The Good Signal: Flat Lines

If the lines on the far right side of the chart are horizontal (flat), it means the model has reached a stable conclusion.

  • Verdict: You can trust the results. Even if you ran the model for another hour, the answer likely wouldn’t change.

⚠️ The Warning Signal: “Fuzzy” or Zig-Zag Lines

If the lines at the far right are still jumping up and down significantly, the model has not converged.

  • What it means: The model was still confused when the timer ran out. It found multiple different answers that all looked “kind of okay,” but it couldn’t pick a winner.
  • The Risk: The ROI numbers in your final report might be unstable. If you ran the exact same project again, you might get different results.

4. Why is there a “1” in the filename?

If you have many marketing channels, Robyn splits them into multiple images (ROAS_convergence1.png, ROAS_convergence2.png, etc.) to keep the chart readable. If you have fewer than roughly 6 channels, you will only see this one file.

Summary

Check the right-hand side of this chart.

  • Stable/Flat? The model is solid.
  • Jittery/Moving? The model might need more data or a longer runtime to find the truth.

More in Plot reference

  • Decoding the Cluster "Elbow": The Grouping Diagnostic

    The Elbow Plot serves as a grouping diagnostic, identifying the optimal number of model "families" by measuring their mathematical similarity. By finding the "sweet spot" where the curve bends, MMM Pilot distinguishes between truly different strategic options and mere variations, ensuring the AI presents you with a diverse shortlist of high-quality model candidates.

  • Decoding the Hypersampling Plot

    The Hypersampling Plot acts as a diagnostic tool to determine if your model is converging on a "sweet spot" or simply guessing. By visualizing thousands of trial runs, it reveals whether specific settings like Adstock or Saturation produce consistent, low-error results or if your data is too sparse to provide a clear signal for certain channels.

  • Decoding the Pareto Front: The Trade-off Map

    Understand the trade-offs between mathematical accuracy (NRMSE) and business logic (DECOMP.RSSD) using the Pareto Front. This guide explains how MMM Pilot filters thousands of models to find the "efficient" survivors, helping you avoid the trap of overfitting while selecting a plausible, actionable narrative for your marketing strategy.

  • Decoding the Robyn "Model One-Pagers"

    Why does MMM Pilot generate multiple image files for a single analysis? Learn how to read the Model One-Pagers—the detailed "resumes" for the top-performing AI models. This guide breaks down the 8 essential diagnostic charts, from Waterfalls to Response Curves, helping you understand the "why" behind the algorithm's decisions.

  • How to Read the "Actual vs. Predicted" Chart

    Is your marketing model telling the truth? The Actual vs. Predicted Response chart is your primary tool for validating accuracy in MMM Pilot. Learn how to compare your historical data against the AI's predictions and interpret the R-Squared metric to ensure your budget decisions are based on reliable insights.

  • How to Read the "Pareto Clusters Detail" Plot

    Don't just pick the model with the lowest error. The Pareto Clusters Detail plot is your map to finding the perfect balance between statistical accuracy and business reality. Learn how to navigate the trade-offs and select the right "Champion" model for your marketing mix analysis.

  • The Payoff: Reading the Budget Optimization Results

    Stop guessing how to split your marketing budget. The Total Budget Optimization Result chart compares your historical allocation against the AI's "Bounded" recommendation. Learn how to read the difference between Average ROAS and Marginal ROAS (mROAS) to uncover "efficiency revenue" -extra growth you can unlock without spending a single extra dollar.

  • The ROI Report Card: Media Efficiency Summary

    Where are you making money, and where are you burning it? The Media Efficiency Summary is your marketing ROI report card. Learn how to compare "Media Share" (what you spent) against "Media Contribution" (what you got) to identify which channels are punching above their weight and which ones are dragging down your profitability.

  • Visualizing Efficiency: Response Curves & Saturation

    Are you overspending on ads that have stopped working? The Response Curves chart visualizes the "Law of Diminishing Returns" for every channel in your mix. Learn how to read these saturation curves to identify which channels are "maxed out" and which ones are scalable "rocket ships" ready for more budget.

  • Visualizing the "Echo": Weibull PDF Adstock

    Does your marketing work instantly, or does it take weeks to peak? The Weibull PDF Adstock chart visualizes the "memory" of your ads. Learn how to distinguish between "Direct Response" channels (immediate decay) and "Brand Building" channels (lagged peaks) to perfectly time your campaign bursts and measure ROI over the correct timeframe.

  • Visualizing the "Unsung Heroes": Other Factors Decomposition

    Marketing isn't just about paid ads. The Other Factors Decomposition chart isolates the revenue impact of your organic efforts (like newsletters) and external economic conditions. Learn how to prove the value of your "free" channels and understand how macroeconomic forces are helping or hurting your bottom line.

  • Visualizing the Battlefield: Competition Impact over Time

    Is your competitor stealing your customers or helping you build the category? The Competition Impact over Time chart isolates external competitive forces to show their direct effect on your revenue. Learn how to interpret positive impact (Category Growth) versus negative impact (Cannibalization) to adjust your defensive marketing strategy.

  • Visualizing the Calendar Effect: Holiday Impact Decomposition

    Did you sell more because of your ads, or because it was Black Friday? The Holiday Impact Decomposition chart isolates the specific revenue impact of calendar events. Learn how to distinguish between "earned" sales and "holiday" spikes, and identify which dates naturally suppress your revenue so you can save budget during dead zones.

  • Visualizing the Natural Rhythm: Trend & Seasonality Decomposition

    Are your sales growing because of your ads, or just because it's December? The Trend & Seasonality Decomposition chart separates your marketing performance from natural business cycles. Learn how to distinguish between your long-term Brand Trend (the baseline) and recurring Seasonal patterns to avoid misattributing "lucky timing" to "marketing genius."

  • Visualizing the Optimization: The Allocator Response Curves

    Why is the AI telling you to cut spend on your favorite channel? The Simulated Response Curve reveals the math behind the Budget Allocator's decisions. Learn how to read the saturation curves to visualize the shift from your "Initial" spend to the "Optimal" recommendation, ensuring you aren't wasting budget on channels that have already hit their limit.

  • Visualizing the Pulse: Paid Channels Decomposition

    It’s not just about how much you sold, but when. The Paid Channels Decomposition over Time chart visualizes the "heartbeat" of your media mix, separating the baseline from the marketing impact. Learn how to distinguish between "always-on" consistency and "flighted" campaign spikes to diagnose ad fatigue and timing effectiveness.

  • Visualizing the Scoreboard: Total Response Decomposition

    What really drove your total sales this year? The Total Response Decomposition chart acts as your marketing balance sheet, ranking every factor from Paid Media to Competitors by their total contribution. Learn how to distinguish between your "Baseline" revenue (brand equity) and "Incremental" revenue (marketing impact) to see the true picture of your growth drivers.