Stockertim Dictionary · Artificial intelligence
Routing
Also called request routing, model routing
In short
Routing means sending each incoming item to the step, model or person most suited to handle it, usually based on its type, difficulty or how confident a classifier is.
- Advanced
- Process
- Checked
In detail
Not every input needs the same treatment. Routing makes the choice item by item. A simple rule might send all refund requests to the refunds team; a model might classify each message and send routine ones to an automated reply, complex ones to a larger model and sensitive ones straight to a person. The router is a decision point at the front of, or inside, a workflow.
Routing saves cost and improves quality at once. Cheap, fast handling for the easy majority frees capacity for the hard cases, and sending sensitive items to people avoids letting automation near them at all. When a classifier is used, its confidence score is often part of the rule: low-confidence items go to a person rather than being guessed.
Misroutes are the main risk. An urgent complaint routed as routine can sit unanswered. Routing rules are therefore tested on real samples, misroutes are tracked, and there is a safe default, usually a person, for anything the router cannot place.
Worked example
Northwind Bikes routes 1,000 customer emails a week.
- Routine questions answered from templates: 1,000 x 0.7 = 700
- Complex questions drafted by a model for staff review: 1,000 x 0.25 = 250
- Sensitive or low-confidence emails sent straight to a person: 1,000 x 0.05 = 50
- Check: 700 + 250 + 50 = 1,000
Staff handle 300 emails a week with full attention instead of 1,000, and nothing sensitive is answered automatically.
In practice
Support desks, document pipelines and agent systems use routing to match work to the right handler, track misroutes weekly and keep a person as the default destination.
Watch out
- Routing decides who handles an item, not whether the handling is correct.
- Without a safe default, items the router cannot place get lost or guessed.
Learn it properly
- Designing agent workflows15 min lesson Free course: AI agents and automation basics
- Order routing and best execution in practice15 min lesson Free course: Brokerage operations
- Dark pools, venues and where orders go15 min lesson Free course: Market microstructure
- Execution and microstructure for algorithms: the order book, queues and latency15 min lesson Free course: Algorithmic trading
- Payment for order flow, and where it is banned15 min lesson Free course: Market making
Official sources
- Guidelines for secure AI system development · National Cyber Security Centre
- NIST AI RMF Playbook · National Institute of Standards and Technology
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.