Managing chatbot costs

What you actually pay for

Three operations consume AI provider tokens:

1. Indexing — only changed content, hashed against a 30-day cache. A full reindex of a typical author's corpus costs fractions of a cent.
2. Query embedding — every chatbot question gets vectorised once. Negligible.
3. Chat completion — the actual answer. ~$0.001 per turn on gpt-4o-mini.

Worked example

For an author with 20 KB articles, 50 FAQs, 100 resolved tickets, and 500 chatbot conversations per month on OpenAI gpt-4o-mini:

• First-ever full reindex: ~$0.002
• Each subsequent reindex (changes only): under $0.0001
• 500 chat turns: ~$0.40
• Total: ~$0.40 / month

Cost controls built into EnvaTicket

• 24-hour response cache keyed by query + license set. The same FAQ asked 50 times bills you once.
• 30-day embedding cache keyed by model + text. Re-running reindex is free.
• Hash-skip reindex — unchanged records never re-embed.
• Per-user daily hard cap (default 100). Prevents a single buyer running up bills.
• Confidence threshold — low-confidence questions escalate to tickets instead of burning more tokens.