Rules vs. ML for Support Ticket Routing

A support operations team routes incoming tickets today using a combination of structured intake forms, keyword-based routing rules, and a small manual triage group. The team handles about 2,000 tickets per day. Ticket categories are relatively stable (billing, technical, returns, account access) but occasionally a new category emerges from a product change.

A product manager wants to "add ML to make routing smarter." Should the team use ML now, later, or not at all? Walk through your reasoning in three parts: (1) what baseline you would compare against, (2) one specific reason ML might help, (3) one specific reason the current approach might still be the better decision.

Follow-up ladder

1. Rung 1: The team tells you 30% of tickets are free-text with no structured intake. Does this change your recommendation?
2. Rung 2: Volume triples to 6,000 tickets per day after a product launch. How does that affect the decision?
3. Rung 3: You recommend rules-first. Three months later, the team says routing accuracy is 78% and misrouted tickets cost the company $50 per ticket in wasted agent time. At what point does the math justify ML, and what would trigger a reassessment?