Key Driver Analysis

Key Driver Analysis helps you discover which columns in your data most influence a chosen outcome. It works well for survey and behavioral datasets that mix numbers, opinion scales, and categories.

Open the tool

1. Click the More menu.
2. Choose Find key drivers.

This opens a panel where you pick an outcome, choose a model, and select candidate drivers. AddMaple will test likely relationships and run your chosen model to estimate each factor's impact.

How it works

AddMaple offers two models:

Random Forest — Trains many small decision trees that handle mixed data types and non‑linear patterns. Each factor is scored by measuring how much model accuracy drops when that factor is temporarily scrambled. Works best for categorical outcomes (e.g., Yes/No, satisfaction buckets, NPS groups).

Elastic Net — A regression model that finds linear relationships between drivers and numeric outcomes. It assigns a coefficient to each driver showing direction and strength of impact. Works best for numeric outcomes (e.g., satisfaction scores, revenue, completion time).

Choosing a model

Use Random Forest when:

• Your outcome is categorical (Yes/No, NPS bucket, satisfaction level)
• You have mixed data types (categories, scales, numbers)
• You want to capture non‑linear patterns and interactions automatically
• You don't need to interpret exact coefficients

Use Elastic Net when:

• Your outcome is numeric (satisfaction score, revenue, duration)
• You want interpretable coefficients showing direction and magnitude
• Relationships are roughly linear
• You have primarily numeric or ordinal drivers

AddMaple will automatically suggest compatible models based on your outcome type. You can switch models in the Configure step.

Step 1 — Choose an outcome

Pick one column you want to explain better. The outcome type determines which models are available:

For Random Forest:

• Opinion scales (Likert)
• Single categories (e.g., NPS bucket, Yes/No, churned vs active)

For Elastic Net:

• Numbers (e.g., satisfaction score, time on task, revenue)
• Opinion scales (Likert — treated as numeric)

Avoid choosing an outcome that will be duplicated among the candidate drivers (data leakage).

Step 2 — Pick possible drivers

AddMaple auto‑suggests columns that look related to your outcome. You can add or remove candidates freely. Useful drivers often include demographics, usage/behavior, attitudes, and experience ratings.

Supported driver types include numbers, opinion scales, single categories, and multi‑select categories.

Tip: Very similar columns can share credit and split their importance. If you see near‑duplicates, consider choosing just one.

Step 3 — Configure and run

Choose your model and adjust advanced options if needed. AddMaple will suggest compatible models based on your outcome type.

Advanced options (Random Forest)

• Number of trees: More trees stabilize rankings but take a bit longer.
• Tree depth: Limits how detailed patterns can get. Lower is simpler and safer.
• Features per split: Fraction of drivers tried at each split to add variety and reduce bias.
• Minimum rows to split: Smallest group size before the tree splits (helps smooth out noise).

Advanced options (Elastic Net)

• Alpha: Controls the balance between Lasso (alpha=1, variable selection) and Ridge (alpha=0, shrinkage). Default 0.5 mixes both.
• Lambda: Regularization strength. Higher values create simpler models with smaller coefficients.
• Standardize: Whether to scale all inputs to the same range before fitting.
• Loss function: How the model measures error. "Auto" picks based on your outcome type.

You can reset to recommended settings anytime.

Reading the results

After you run the analysis, you'll see:

• Top drivers and summary: A short description of what seems to move your outcome most
• Importance table: Higher scores mean a factor explains more of the differences you see
• Model fit metric:
  • For Random Forest: Out‑of‑bag balanced accuracy (how well the model predicts unseen rows)
  • For Elastic Net: R² score (how much variation in the outcome is explained)
• AI explanation: An automatically generated interpretation of your results

Interpreting importance scores

Random Forest — Importance scores show how much accuracy drops when each driver is scrambled. Higher values mean removing that driver hurts predictions more.

Elastic Net — Importance scores are the sum of absolute coefficient values. They show total linear effect size. Positive coefficients increase the outcome; negative coefficients decrease it.

General tips

• Results show associations, not strict causation. Use as directional guidance.
• Very small categories can be noisy — consider combining rare groups or filtering.
• Similar drivers may split importance between them.

Good to know

• Random Forest works well with mixed data types and captures non‑linear patterns automatically. It's robust to outliers and missing values.
• Elastic Net provides interpretable coefficients showing direction and magnitude of each driver's impact. Best for numeric outcomes with roughly linear relationships.
• You can switch models and rerun to compare results
• AddMaple automatically suggests columns that are statistically related to your outcome
• AI-generated explanations help you understand what the results mean

Availability: Key Driver Analysis is limited to certain plans.