01
Your team already knows what deserves attention
An existing customer needs an expansion quote before a deadline. Another inquiry asks for general product information. Both belong in the queue, but your team may decide to handle them differently.
Scoring and prioritization turn that experience into a repeatable prediction. StayCharted AMT learns from historical examples paired with your approved scores or priority levels, so new work can be assessed against your own criteria.
02
Teach the model what priority means in your business
Start with one workflow and examples showing how your team prioritized previous records. These illustrate a sales inquiry queue:
| Information available when the inquiry arrived | Approved priority |
|---|---|
| Existing customer needs an expansion quote before a deadline | High |
| Qualified prospect planning a purchase next quarter | Medium |
| General information request with no stated project or timeline | Low |
A service queue will use different examples, such as the reported impact of an issue or the urgency of a request. Keep each model’s task clear rather than combining unrelated definitions of importance.
03
Choose priority tiers or a defined scoring scale
Priority tiers such as High, Medium, and Low are useful when the next step is choosing a queue or review order. Numerical scores are useful when your process already has a defined scale and consistently scored examples.
Whichever you use, document what the labels or numbers mean. A score should express the business criterion you trained for; a number is not automatically a probability of purchase or a forecast of revenue.
Prediction confidence is a separate concept: it describes certainty about the prediction, not how important the lead or request is. Keep those meanings separate when deciding what action to take.
04
Prepare examples using what was known at the time
Pair historical inputs with reviewed scores or priority levels. Include only information that would be available when a new record needs to be assessed.
For example, a lead’s eventual purchase amount would not be available when an initial inquiry arrives. Including later information in the inputs can make a historical test look useful without reflecting how the model will work on new inquiries.
- Use consistent definitions for each priority or score.
- Include examples across the levels your team uses.
- Reserve separate examples for evaluation.
- Review older labels when business priorities have changed.
05
Use predictions to organize the next action
Review scores in a batch file or call the StayCharted AMT API from the application that manages your queue. Your application decides how the prediction affects the work.
- New lead or request
- Your application or batch workflow
- StayCharted AMT model
- Predicted score or priority
- Organize the queue and review exceptions
Keep explicit deadlines and urgent exceptions in your workflow rules. A model recommendation should not prevent your team from escalating a request or changing its priority when new information arrives.
06
Evaluate what happens to the important work
Check results across priority levels rather than relying only on an overall match rate. Pay particular attention to high-priority examples assigned a lower priority, and to routine work repeatedly placed at the top of the queue.
- Compare predictions with decisions your team has reviewed.
- Test the queue order on recent examples.
- Review urgent exceptions and disagreements.
- Update training examples as your criteria change.
Start with recommendations your team can inspect. Use the evaluation to decide where the model is useful and where a person should continue making the final decision.