Text Classifier and Smart Routing

AI as a router

The Text Classifier node takes text and a list of categories (each with a short description) and sends the item down the output for the best match. This replaces long chains of keyword rules for messy input such as support emails. Provide a clear "other" category so unclear items go to a person.

Rules vs AI routing

This keyword router ran in Node.js and routes three test messages correctly. It is cheap and predictable but brittle: "my card was charged twice" would fall through to general. AI classification handles such phrasing but costs tokens.

const route = t =>
  /refund|money back/i.test(t) ? "billing" :
  /error|crash|bug/i.test(t)   ? "technical" : "general";

console.log(["I want a refund", "App crash on login", "Hello"].map(route).join(","));

Output:

billing,technical,general

Use rules first, AI second

Cheap deterministic checks (sender domain, a known ID format) can handle obvious cases for free. Send only the remaining ambiguous items to the classifier to save cost.

Quick check: Why include an "other" category in a text classifier?

• It hides errors
• It speeds up the model
• It is required by the licence
• So unclear items can be sent to a person instead of being forced into a wrong category