1. Booking rate
Of the conversations where the customer was a fit, how many ended with a booked appointment? This is the bot's reason to exist. We tag every conversation the bot handles and every booking it creates, so GoHighLevel can report the ratio. Healthy is 30 to 50 percent. Below that, look at where conversations stall: usually the bot is asking too many questions before offering a time.
2. Hand-off rate
How often the bot passed the conversation to a human. Around one in five is healthy. Near zero is a warning sign: the bot is probably handling things it shouldn't, like price questions or complaints. Over half means the brief is too narrow or the knowledge base is missing common questions. Read the hand-off reasons; they are a to-do list for the knowledge section.
3. Time to human
After a hand-off, how long until a person replied? The bot promised "shortly", and the customer is holding it to that. Under fifteen minutes in business hours is the standard; longer than that and the hand-off feels like being put on hold. If this number drifts, the fix is usually in the alert routing, not the bot.
4. Correction rate
How often a human had to correct something the bot said: a wrong hour, a service it doesn't offer, a price. Reps flag these with a reaction in the conversation view, and the count goes into the weekly review. Each correction becomes a knowledge-base edit. This number should trend toward zero within a couple of months; if it doesn't, the knowledge section is being neglected.
A flagged correction is a customer telling you exactly what the bot gets wrong. Fixing it takes two minutes and prevents the same mistake for every future caller. Treat the flag as the most valuable output the team produces.
5. Sentiment
Numbers miss tone. Once a week someone reads twenty random transcripts and rates the customer's mood at the end: better, same, or worse than at the start. A bot can hit every metric above while quietly annoying people. If more than one in ten conversations ends worse, the examples in the prompt need work, or the bot is too persistent.
The weekly ritual
Fifteen minutes, every Monday: pull the five numbers, read twenty transcripts, list the knowledge edits and prompt tweaks, make them, bump the version. Bots that get this attention keep improving. Bots that don't get quietly switched off within six months.
Key takeaways
- Booking rate: bookings divided by qualified conversations. Target 30 to 50 percent.
- Hand-off rate: aim for 15 to 25 percent. Near zero means overreach; over half means the brief is too narrow.
- Time to human after hand-off: under fifteen minutes in business hours.
- Correction rate: how often a human had to fix a bot statement. Should trend to zero.
- Sentiment: read a sample of transcripts. Numbers miss tone.