Predictive Intelligence
Quick answer
Train classification, similarity and clustering models on your own data and use them without eroding trust.
Key takeaways
- Precision and coverage matter more than raw accuracy
- Start as a suggestion, automate once trust is earned
- Set confidence thresholds explicitly
- Retrain on a schedule or accuracy decays
Three model types
Classification predicts a field value, for example assignment group from short description. Similarity finds records like this one, which powers agent assist. Clustering groups records to reveal patterns, often used to find candidate problems or automation targets.
Training and evaluation
Models train on historical records with a defined filter and target field. Read the precision and coverage numbers before you deploy: a model that is 90 percent precise on 20 percent of records is useful, a model that guesses on everything is not.
Deploying carefully
Introduce predictions as suggestions before automation.
- Show the prediction with its confidence and let the agent accept it
- Only auto assign above a confidence threshold you agreed with the service desk
- Retrain on a schedule, processes drift and accuracy decays
- Exclude records created by the model itself from the next training set
Want to learn this properly?
Our live, instructor-led ServiceNow Training covers this hands-on, with real projects and a certification path.
Check your understanding
Which model type suggests similar past incidents?
- A. Classification
- B. Similarity
- C. Clustering
- D. Regression
Show answer
B. Similarity
Similarity finds comparable records for agent assist.
What should gate automatic assignment?
- A. Record age
- B. Confidence threshold
- C. Assignment group size
- D. Priority
Show answer
B. Confidence threshold
Only predictions above an agreed confidence should act without a human.