The workbook
Dropdown answers in, segment out — with example profiles.
01 / Your segments
Start from the segments you already trust.
Map the segment label from your segmentation and the questions that might predict it.
The Typing Tool does not make new segments; it learns to tell yours apart from as few questions as possible. Need segments first?
02 / Screening
Every question, scored three ways.
ANOVA F, mutual information and SHAP each rank the candidates; their average decides which questions to try.
For the Excel workbook, the shortlist is also pruned of questions that say the same thing — any pair correlated above 0.90 keeps only one.
03 / Accuracy against length
Fewer questions, or more accuracy.
Each length is cross-validated; the smallest that clears 75% accuracy and 0.70 macro-F1 is suggested — you pick.
Every rung shows accuracy, macro-F1 and calibration with their spread across five folds, plus a confusion matrix and per-segment precision and recall.
The classifiers
The Excel workbook tries every length from 4 to 12 questions with multinomial logistic regression and LDA. A model bundle can train the full catalogue at the lengths you pick:
- Logistic
- LDA
- Ridge
- Lasso
- Elastic net
- Polynomial logistic
- Decision tree
- Random forest
- SVC
- Naive Bayes
- Gaussian process
- MLP
- XGBoost
- LightGBM
- CatBoost
04 / The answer
A classifier your teams can use.
An Excel typing workbook: pick each answer from a dropdown, and the predicted segment appears — no Python needed.
Or a model bundle for your own systems, with a README and a loader script. Either way, the contact centre, sales or your website can classify a new customer in seconds.