How to measure whether your chatbot is working

You don't need a dashboard full of charts to know whether a website assistant is worth keeping. You need a baseline, four numbers and half an hour a month. This guide uses no industry benchmarks, because the only comparison that matters is your business before and after.

Take a baseline before launch

Before the assistant goes live, write down what a normal month looks like. Without this, every later number floats in the air.

• Inquiries per month from each source: contact form, e-mail, phone, walk-ins.
• How many of those inquiries became customers.
• Roughly how much time someone spends answering routine questions each week.
• How many inquiries arrive outside business hours, if you can tell.

If you don't have records, keep a simple tally for two weeks before you launch. Rough numbers are better than none.

Conversations and inquiries

The first two numbers come straight from your Allinaa cabinet:

• Conversations: how many visitors used the assistant. A very low number usually means the chat is hard to find or the greeting doesn't say what it can help with.
• Inquiries: how many visitors left their details with consent. Each one lands in the inquiry inbox with the visitor's request, and you receive an e-mail notification.

Standard includes a monthly performance report e-mailed to you, so you don't have to collect these numbers by hand.

Answer quality: transcripts and corrections

Numbers don't tell you whether answers were right. Reading does. Each month, read a sample of conversations, for example ten picked at random plus every conversation that ended without an inquiry after a long exchange. For each one, ask:

1. Did the answer match your approved knowledge?
2. Was it complete, or did the visitor have to ask again?
3. Did the assistant offer a person or the inquiry form when it should have?

Keep a count of wrong answers per month. When you find one, request a correction from the cabinet. Corrections are reviewed by the Allinaa team and applied as a new version of your knowledge, never automatically, so the count should fall over time.

Outcomes: which inquiries became customers

This is the number that pays the bill. The inquiry inbox has statuses (New, In progress, Completed), private notes, search and a CSV export. Once a month:

1. Export the month's inquiries to CSV and open the file in a spreadsheet.
2. Add a column "Became a customer" and mark yes or no for each row, using your booking system, invoices or CRM.
3. Add a column "Value" with the first sale amount for each yes.
4. Add a column "After hours" and mark inquiries that arrived when you were closed.

Total the yes rows and the value column. Compare with your baseline and with the break-even number for your plan. Be strict: a customer who would have called anyway is not new business. The after-hours column helps you judge that.

Questions your website doesn't answer

Transcripts are also free market research. Note every question the assistant couldn't answer from your knowledge. If the same question appears three times in a month, it belongs in two places: in the assistant's knowledge and on your website. Typical finds are parking, payment plans, weekend availability and whether you serve a particular town.

A monthly review routine

This takes about half an hour. After three months you will have a clear answer, built from your own data, about whether the assistant earns its place. If you are not live yet, the 14-day free trial is a good time to take your baseline, and the pricing section lists what each plan reports.

Frequently asked questions

What is a good number of inquiries from a chatbot?

There is no universal number. Compare with your own baseline from before launch, and with the break-even number for your plan.

Can I see every conversation?

Yes. Conversations are available in your cabinet, and inquiries can be exported to CSV.

How do I fix a wrong answer?

Request a correction from the cabinet. The Allinaa team reviews it and applies it as a new version of your knowledge.