Reimagining the Support Tier Model for AI

I designed the original three-tier support model for this organization — a structure built around human skill levels, escalation paths, and handoff protocols. It worked well for its time.

But support tier models were designed before AI existed as a practical workflow tool. The assumptions baked into the traditional model — that every ticket needs a human at first contact, that triage requires lived experience, that routing decisions happen after the fact — no longer hold.

If you were designing a three-tier support organization from scratch today, knowing what AI can reliably do and what it still can’t, how would the model change?

This concept explores that question interactively — letting you compare how the same ticket flows through a traditional model versus an AI-augmented one, and demonstrating a counterintuitive insight along the way.

The AI model is not the differentiator. Two organizations using the same AI platform will get dramatically different outcomes — and the gap comes down to one thing: knowledge base quality.

High-quality documentation lets AI resolve tickets with confidence. Low-quality documentation produces deflection — tickets that didn’t reach a human, but weren’t actually resolved either. The simulator below makes this visible.

A second distinction worth making: deflection ratecounts any ticket that didn’t reach a human as a success. True resolution rate only counts tickets where the customer’s actual problem was fixed. Organizations optimizing for the wrong metric will over-invest in AI and under-invest in the documentation that makes AI work.

Throughout this model, freed agent capacity is described as redirected toward complex, judgment-driven work — not eliminated. AI-augmented support done well means humans spend less time on password resets and more time on the problems that require experience, empathy, and context.

The redesigned model has three core changes from the original:

• AI as a pre-tier layer — not a replacement for any tier, but a structured first pass that classifies, enriches, and routes before a human ever touches the ticket
• Bypass routing for known defects — confirmed engineering issues skip T1 and T2 entirely, reducing time-to-engineering by hours
• Knowledge base as infrastructure — treated with the same rigor as tooling or headcount, with ownership, quality metrics, and a regular review cadence