The deck
Opens on your question and its answer; ends on what to do.
01 / Your question
It starts with your question.
Agents use your business question to recommend the models to fit, and write every finding around it.
You name the outcome — NPS, satisfaction, intent, churn — and the attributes that might move it; the deck opens on your question and answers it.
Before any model runs
- Ask for it. “What’s driving NPS in our business banking segment?” The platform recognises the intent and selects Driver Analysis.
- Map the variables — the outcome, the potential drivers, and any control variables, which are held constant in the model but left out of the importance table.
- Reverse-coded items. Any scale worded the opposite way is flipped, so a high score always means the same thing.
- The outcome scheme — NPS (Detractor 0–6, Passive 7–8, Promoter 9–10), top-2-box, a custom mapping or the raw values, suggested from the variable’s label for you to confirm.
02 / The model plan
Several models, one honest comparison.
You confirm a recommended model plan; every model is cross-validated against the others.
- Random Forest
- Gradient Boosting
- XGBoost
- LightGBM
- CatBoost
- Linear
- Logistic
- PLS
- Shapley Value Regression
- MARS
- Bayesian Linear
- Neural Net (MLP)
The primary model sets the importance ranking. If another fits better on cross-validation, the report says so rather than hiding it.
How the plan is chosen
Without a usable question the plan falls back to rules: logistic regression and a random forest for a yes/no outcome, random forest and gradient boosting for NPS, PLS when there are many drivers for the sample size. Tree models can add SHAP explanations; linear models report p-values; Shapley value regression is exact up to 12 drivers.
03 / What matters
Every driver, ranked by what it moves.
Importance sums to 100% across the drivers, and each is tiered: primary, secondary or low priority.
Tiers come from where a driver falls in the spread of importance — the top quarter primary, the 40th to 75th percentile secondary. Drivers that are missing, constant or not numeric are left out, with the reason stated.
For NPS, two stages
Separate models for Detractor → Passive and Passive → Promoter, because what creates promoters is rarely the inverse of what creates detractors. Each stage needs at least 20 respondents.
04 / Where to act
Importance, against how you perform.
High importance and below-median performance is where to concentrate.
The map splits at the medians into four quadrants — Concentrate here, Keep up, Possible overkill and Low priority — with every driver labelled and its numbers in a table beside it.
05 / The answer
Get the answer, in a deck.
The PowerPoint opens on your research question and its answer, then shows the drivers and what to do.
Executive summary · Modeling approach · Driver findings and quadrant maps · Insights · Strategic recommendations — with the model comparison and an editable quadrant chart in the appendix.