Most owners open their assistant's dashboard for the first time, see "412 conversations this month" and feel vaguely pleased. Then a week later they wonder whether any of it turned into work, and nothing on the screen answers that.
Conversation volume is the number vendors like to show because it only goes up. It tells you people clicked a chat bubble. It says nothing about whether a plumber got a job, a clinic got a new patient enquiry, or a customer with a problem got to someone who could fix it.
These five AI receptionist metrics are the ones that answer the real question. Some come straight from your analytics, one you have to keep yourself, and all of them fit in a ten-minute weekly check.
1. Enquiry capture rate
What it is: of the conversations that were genuine enquiries, how many ended with contact details you can act on.
The key word is genuine. Plenty of conversations are someone checking opening hours or asking where you park, and they should end without a lead. Counting those drags the rate down and sends you chasing a problem that does not exist.
Worked example (replace with your own): 300 conversations in a month. You skim a sample and judge that roughly 40% were enquiries about doing business, so about 120. Leads captured: 54. Capture rate: 54 ÷ 120 = 45%.
If that rate drops, look at where people stop. The usual causes are asking for contact details too early, or asking five questions before offering anything useful. Asking for details politely covers the wording.
2. After-hours share of leads
What it is: the proportion of captured leads that arrived outside your staffed hours.
This is the number that justifies the whole exercise. A lead captured at 2pm might have reached you anyway by phone or email. A lead captured at 9:40pm on a Sunday is one that previously sat in a contact form until Monday, by which time the visitor had often booked someone else.
Worked example: 54 leads, of which 21 arrived between 6pm and 8am or at weekends. After-hours share: 21 ÷ 54 ≈ 39%.
You do not need to hit any particular figure. What matters is knowing this number when you decide whether the tool earns its keep, because those 21 leads are the clearest case for it.
3. Handover rate
What it is: the share of conversations passed to a person, whether by live takeover, a transfer to a human, or a support ticket.
This one is easy to misread. A rate of zero is not a triumph; it may mean handover is switched off or buried, and frustrated visitors are simply leaving. A rate that climbs month on month is a signal worth reading.
Look at the reasons, not just the total:
- Complaints and account issues. These should go to a person. A steady number here is correct.
- Questions the assistant could not answer. A rising count means your content has gaps.
- Visitors asking for a human immediately. Check your welcome message; it may be overpromising or sounding robotic.
- Handovers nobody picked up. The most expensive kind. See metric five.
For more on where the line sits, when a person should take over goes through the triggers.
4. Unanswered questions
What it is: the questions the assistant had no relevant content for. SpideyChat records these as knowledge gaps, which makes this the easiest metric to act on.
Do not track the raw count alone. Track how many distinct gaps appear, and how many you closed since last month.
Worked example: 38 unanswered questions logged. Grouped, they come to 7 distinct topics, and 3 of those cover 29 of the 38. Writing three Q&A pairs this week would deal with three-quarters of the problem.
This is also the metric that improves your website. If twelve people asked whether you work on listed buildings, your service page probably should have said so.
5. Time to human follow-up
What it is: the time between a lead being captured and a person from your business contacting that customer.
The assistant's response time is measured in seconds and tends not to vary much. Yours is the number that decides whether the lead is worth anything. A perfectly captured enquiry that waits two days for a callback has quietly become a lost one.
Your chat analytics will not know when someone rang the customer back, so keep this one yourself. A column in your CRM or a spreadsheet with "lead time" and "first contact time" is enough.
Worked example: over a week, 14 leads, with follow-up gaps of 1, 2, 2, 3, 3, 4, 5, 18, 19, 20, 22, 40, 44 and 60 hours. The median is between 5 and 18, around 11.5 hours, but the spread tells the story: weekend leads are waiting nearly two days.
A one-glance scorecard
| Metric | Where it comes from | Healthy sign | Worth investigating |
|---|---|---|---|
| Capture rate | Leads vs genuine enquiries | Steady or rising | Sudden drop after a content or greeting change |
| After-hours share | Lead timestamps | A consistent slice each week | Near zero, which may mean the widget is hidden on mobile |
| Handover rate | Inbox, tickets, transfers | Mostly complaints and account issues | Rising due to unanswered questions |
| Unanswered questions | Knowledge gaps | Distinct topics shrinking | The same topic appearing month after month |
| Time to follow-up | Your CRM or spreadsheet | Same day for weekday leads | Weekend leads waiting until Tuesday |
Setting up the weekly check
- Every Monday, note the week's leads and how many arrived after hours.
- Skim ten conversations and count how many were genuine enquiries, then estimate the capture rate.
- Open the unanswered questions, group them, and write answers for the top two.
- Check the follow-up column and chase any lead older than your own target.
- Once a month, compare the five numbers with the previous month and write one sentence on what changed.
That fifth step sounds trivial. It is the one that stops you from making the same content change twice or forgetting why the capture rate dropped in March.
Start with this week. Write down the five numbers as they stand today, even if some are rough estimates, because a baseline is the only thing that makes next month's figures mean anything. If you want to translate captured leads into money, the chatbot ROI calculator will do the arithmetic, and the features page lists what the analytics dashboard records.