01
Start with categories your team already uses
Train on requests labeled High, Medium or Low. StayCharted predicts a priority category and confidence; it does not calculate lead scores or rank a queue.
Choose one recurring classification task. Keep labels consistent and provide examples of each category from the sources you expect to use.
02
Prepare reviewed examples
Use the information available when a new row arrives and the category your team assigned. These examples are illustrative.
| Request description | Reviewed priority category |
|---|---|
| Existing customer needs an expansion quote today | High |
| General information request with no deadline | Low |
Avoid unique IDs as categories unless you have a genuine, repeatable classification task with enough examples for each label.
03
What this model does and does not do
High, Medium and Low are discrete labels learned from your examples. AMT does not generate a continuous lead score, forecast revenue or sort your queue. Prediction confidence measures certainty about the category, not the importance of a request.
04
Test before putting it to work
Hold back representative examples to measure accuracy overall and by category. Check the mistakes that matter to your team, including confident wrong answers.
Choose a confidence cutoff from your evaluation results. Review uncertain rows and keep a way for a person to correct any answer. New inputs can behave differently from the test set.
05
Use categories in your existing workflow
- New input description
- Trained category model
- Predicted category and confidence
- Human review where needed
- Your application applies the approved result
Fill a file or call the API on an eligible plan. Your software owns queue ordering, routing, imports and updates; the classifier supplies the label.
06
Keep examples current
Save reviewed corrections, add examples when labels or inputs change, and test the next version before publishing it. The model does not learn a new category merely because it encounters an unfamiliar row.
07