A chatbot that nobody reviews slowly gets worse. Prices change, services are added, seasonal questions arrive and the knowledge base drifts out of date, while the assistant keeps answering with confidence. Analytics is how you notice. But most dashboards offer far more numbers than a small service business can use, and it is easy to stare at a total-conversations chart and learn nothing. This playbook picks the handful of measures that actually change decisions, shows how to read them, and gives a fifteen-minute weekly routine that turns them into improvements. It is written for owners of salons, clinics, trades, studios and similar businesses, and it avoids invented benchmarks: your own trend over time is the only comparison that matters.
Start with the question, not the chart
Every metric should answer a decision you might make. If a number cannot change what you do, ignore it.
| Decision | Metric that informs it |
|---|---|
| Is the assistant worth keeping? | Leads captured and bookings made through chat |
| Is the knowledge base good enough? | Questions the assistant could not answer |
| Do I need staff cover at certain times? | Conversations by day and hour, and handovers |
| Is the website converting visitors? | Share of chats that end in a booking or contact |
| Are answers trustworthy? | Sampled conversation quality and corrected answers |
The six measures worth tracking
- Conversations started. The top of the funnel. Useful as a trend, not a target.
- Leads captured. Conversations that ended with contact details. Divide by conversations to get a capture rate.
- Bookings made. Appointments confirmed through the chat, and the share of chats that produce one.
- Unanswered or low-confidence questions. The single most useful list for improving the assistant.
- Handovers to a person. How often the assistant passes a conversation to staff, and why.
- Busy times. Which days and hours visitors chat, especially outside opening hours.
Everything else is detail. If you track only these six, you will know whether the assistant earns its place and where to improve it.
Reading each measure
Conversations and capture rate
A rising conversation count with a flat capture rate means visitors are engaging but not converting. Look at the questions they ask before leaving: often the answer is missing pricing, unclear next steps or no prompt to leave details. A falling count after a website change may mean the widget is hidden or slower to load. Compare against the same weeks in previous months, because service demand is seasonal.
Unanswered questions
This is the gold list. Each unanswered question is a visitor who wanted something your documents did not contain. Group them: pricing gaps, policy gaps, service gaps, location gaps and things you genuinely do not offer. Then add a short, factual passage for each recurring theme. Ten minutes of writing here usually improves more conversations than any setting.
Handovers
Some handovers are healthy: complaints, complex cases and people who ask for a human. Too many suggests the assistant lacks information; none at all may mean it is holding on to conversations it should pass on. Read a handful each week and ask whether a person was really needed.
Busy times
If a quarter of chats arrive after closing, that is demand you were previously missing, and a reason to make sure the assistant can book or capture details out of hours. If chats spike on Monday mornings, plan staff attention for follow-up then.
A fifteen-minute weekly routine
- Open the list of unanswered questions and pick the top three themes. Write or update a passage for each.
- Read five conversations chosen at random and five that ended in handover. Note anything wrong, cold or confusing.
- Check leads and bookings from the week: did any need manual follow-up? Were details complete?
- Glance at counts against last month, only to spot a sudden change.
- Write down one change to make before next week.
The habit matters more than the tooling. See testing starter prompts for a question set you can reuse in this review.
Sampling conversations well
- Read a mix: short and long, booked and abandoned, day and night.
- Read the last message of abandoned chats: where did the visitor stop?
- Look for repeated phrasing you have not seen before; that is new demand.
- Note answers that were correct but poorly worded, and fix the source text.
- Keep a running list of corrections and check that they hold.
Connect chat to money
Put a value on outcomes so the numbers mean something to you. If an average appointment is worth $80 and chat produced twelve bookings this month, that is roughly $960 of business the assistant helped book, before subtracting the cost of the plan. Be modest in attribution: some of those visitors might have booked anyway. Measure honestly by asking new customers how they booked, and by comparing out-of-hours bookings, which the assistant is more clearly responsible for.
| Outcome | How to value it |
|---|---|
| Booking through chat | Typical appointment value, minus your margin of doubt |
| Lead captured | Your lead-to-customer rate times average customer value |
| Question answered without staff | Minutes saved times hourly cost, only if you actually reallocate the time |
Benchmarks: use with care
You will see published figures for chatbot capture rates and resolution rates. Treat them as anecdotes: definitions differ, industries differ and the sources are often vendors. Your own trend, measured the same way each month, is more reliable. If capture rate improves after you add pricing to the documents, that is a real signal.
Common mistakes
- Chasing total conversations instead of outcomes.
- Never reading actual conversations.
- Ignoring unanswered questions.
- Comparing week to week without accounting for seasons.
- Treating every handover as a failure.
- Making many changes at once so nothing can be attributed.
A worked monthly review
| Question | What to look at | Possible action |
|---|---|---|
| Did chat earn its place? | Leads and bookings this month against last | Keep, or improve the weakest step |
| What did visitors want that we lacked? | Top unanswered themes | Write three new passages |
| Were people passed on well? | A sample of handovers | Improve the details staff receive |
| When should we be ready? | Busiest days and hours | Adjust staffing or out-of-hours booking |
Write the outcome of each review in two lines and keep the notes. After a few months you will see which changes worked, which is more valuable than any single chart.
Frequently asked questions
Which chatbot metrics matter most?
Leads captured, bookings made, unanswered questions, handovers and busy times. Together they show value and what to fix.
How often should I review analytics?
A fifteen-minute weekly review is enough for most small businesses.
What is a good capture rate?
Compare against your own history. Published figures use different definitions and are not reliable targets.
Why do unanswered questions matter?
Each is a visitor whose need your documents did not cover, and each fix improves many future conversations.
Should I read individual conversations?
Yes. Reading a small sample weekly reveals problems that counts hide.
How do I measure return on investment?
Value the bookings and leads the assistant produced, subtract the plan and AI costs and be conservative about attribution.
See the human handover feature, try a sample in the playground and read how to reduce funnel drop-off in chat and first contact resolution.
Further reading
- Google Search Central: creating helpful, people-first content: What Google looks for in useful pages.
RISPONDI A OGNI VISITATORE. PRENOTA CHI È PRONTO.
Installa Crescelo sul tuo sito in 15 minuti
Addestra il tuo assistente sui tuoi listini, documenti e condizioni di servizio. Attiva risposte fondate sulle tue fonti 24 ore su 24 e prenotazioni in chat senza sovrapposizioni.