How to estimate token usage for client?
Hey folks! Quick help on Voiceflow credits for a web chat agent.
The estimator seems to assume ~1 credit/interaction and ~0.04 credits/message on GPT-5 Mini. I want to give my client a solid monthly estimate + clear cap. Can you sanity-check a few things?
Benchmarks: What’s a typical credits per conversation you’re seeing (e.g., ~4 user turns, ~4 bot turns, ~2 KB lookups)?
What counts: Besides user/bot turns, do system prompts, KB retrievals, tools, memory reads/writes add credits in practice? Any sneaky sinks to watch?
Cost control: Best quick wins—shorter outputs, max tokens, cheaper model for FAQs, escalate only when needed?
Caps/fallback: Easiest way you enforce a hard monthly cap and auto-switch to a lite/FAQ mode around 90%?
Seasonality: Do you swap plan tiers month-to-month for peak vs shoulder?
Working formula I’m using:
Monthly credits ≈ Visitors × Engagement% × Conversations/Visitor × Credits/Conversation
Would love real-world ranges (mini ≈ X–Y credits/convo) so I can share a clean estimate. Thanks!
0 Replies