## TL;DR
Start with deterministic rules: assignment rules route by category and location, keyword-based categorization fills category/subcategory, and priority follows urgency-plus-impact. Pilot on one category, measure misroutes, then expand. Rules beat ML for the first 80 percent of triage.

## The error
```text
(Automation task; no error.)
```

## Steps
1. Map the current triage: sample 100 recent tickets and note how each was categorized, prioritized, and assigned. Expected: baseline misroute rate. You need the baseline to prove improvement.
2. Build assignment rules: ServiceNow > System Policy > Assignment Rules, routing by category, location, or CI. Expected: rules created, ordered most-specific first. Rule order matters; put exceptions above generals.
3. Add auto-categorization: business rules or flow designer that set category/subcategory from keywords in the short description. Expected: configured. Start with your top 5 categories by volume.
4. Set priority from urgency and impact with a clear matrix. Expected: matrix documented. Let requesters set urgency; the system computes priority.
5. Pilot on one category for two weeks, measure misroute and recategorization rates, tune, then expand. Expected: misroutes down. Expand only what the data supports.

## When to use
- Slow or inconsistent manual triage
- High ticket volume with clear patterns

## When not to use
- Low volume (manual triage is fine)
- Highly variable tickets with no patterns (rules will misfire)

## Compatibility
- ServiceNow ITSM; Flow Designer or business rules

## Variants
### Predictive Intelligence
ServiceNow's ML categorization can layer on top of rules for the long tail; train it on your corrected data.
### Auto-assignment to individuals
Assign to groups, not people, unless you have true load balancing; people go on leave.

## Why it happens
Triage is pattern matching at scale, which is exactly what rules do well. The failure mode is over-automation: rules for patterns that do not exist yet.

## Edge cases
- Keep a human override path; automation confidence is never 100 percent.
- Review rules quarterly; org changes silently obsolete them.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_rocvSVJBLC-uWw-njO6UYQ