Metrics & KPIs

Time-to-First-Token (TTFT)

/ time-to-first-token-ttft /

Delay from prompt submission to the model’s first output token — the LLM’s share of response latency.

Why it matters

TTFT isolates the model’s share of latency. Tracking it separately makes it obvious when a model change is what slowed everything down.

Related — Metrics & KPIs

After-Call Work (ACW)

Average Handle Time (AHT)

Average Speed of Answer (ASA)

Call Abandonment Rate

Containment Rate

CSAT (Customer Satisfaction Score)

Customer Effort Score (CES)

Deflection Rate

End-to-End Latency

First Call Resolution (FCR)

Time-to-First-Token (TTFT)

Discover our world