Bot Live Chat Weekly Review: Read Last Week’s Chats and Decide What the Bot Learns Next (Free Outcome Tagger)

Bot live chat gets better in one way only: someone reads what happened and changes something. Most small teams set up a bot, answer the chats it hands over, and never look back at the whole week. This guide gives you a 30-minute weekly review instead. You sort last week’s chats into six outcomes, pick the questions your team keeps answering by hand, and teach the bot those first. A free tool does the sorting for you.

6outcomes every chat falls into
30 minfor the weekly review
20+chats make a fair sample
1change to make each week

⚑ In short

  • Every chat that involves a bot ends in one of six outcomes. Counting them tells you where to work.
  • “Bot missed, a person stepped in” is the most valuable pile: those agent answers are what the bot should learn.
  • “Asked for a person, nobody came” is the most urgent pile: fix it before anything else.
  • Change one thing a week, then check next week whether the numbers moved.
Bot live chat in LiveChatBot: the LiveChatBot shared inbox with a conversation open
How it looks in LiveChatBot: the LiveChatBot shared inbox with a conversation open.

Why review

The bot can’t tell you what it got wrong

A bot doesn’t know when it failed. It answered, the visitor left, and as far as the software is concerned, that was a conversation. Your team knows a bit more, because they see the chats that came to them. But they see them one at a time, in the middle of a busy day, and nobody adds them up.

That’s why so many website bots stay exactly as good as they were on launch day. The same five questions keep reaching a person. The same awkward answer keeps going out. Nobody is ignoring it; there’s just no moment in the week when someone looks at all the chats together.

A weekly review creates that moment. It’s not an analytics project. It’s one person, thirty minutes, a sample of last week’s chats and a simple question for each one: how did this end?

The answers fall into a small number of piles, and each pile has its own fix. A pile of chats where the bot answered and the visitor said thanks needs nothing. A pile where the visitor asked for a person and nobody replied needs urgent attention. A pile where a team member stepped in and answered something the bot couldn’t is a ready-made list of what to teach next.

If you’ve never reviewed your chats, the first session is the most useful one you’ll ever do. Expect to find two or three things that surprise you. After a few weeks the review gets quicker, because you’re checking whether last week’s change worked rather than starting from scratch.

This guide assumes your bot and your team share one inbox, so a chat can move from the bot to a person. If you’re still deciding how that handover works, read our handover guide first, and our turn-taking rules for what happens once both are in the conversation.

Six outcomes

How every bot live chat ends

Sort each chat into exactly one of these. The order below is the order to fix them in.

πŸŽ“

1. Bot missed, a person stepped in

The bot replied, then someone from your team answered without being asked. Your teaching list.

🚨

2. Asked for a person, nobody came

The visitor asked for a human and no one replied. Fix first.

❓

3. Left after the bot

The bot answered and the visitor went quiet. Some were helped, some gave up. Read a few.

🀝

4. Asked for a person, got one

The handover worked. Check the wait time.

πŸ‘€

5. Person only

The bot never spoke. Check whether it should have.

βœ…

6. Bot solved it

The bot answered and the visitor said thanks or OK. Leave it alone.

Collecting the chats

Pick a fair sample in ten minutes

You don’t need every chat. You need enough to see patterns, chosen so they aren’t all from one kind of day.

How many. Twenty to forty chats is plenty for a small business. Fewer than twenty and one odd conversation can swing the numbers. More than forty and the review stops being a thirty-minute job.

Which ones. Take chats from across the week, not just Monday. Include some from busy hours and some from quiet ones, some from after hours if your bot runs at night, and some that ended in a ticket. If you only look at chats your team handled, you’ll miss the ones where the bot answered alone, and those are half the story.

How to copy them. Open each chat in your inbox and copy the messages into a plain text file. Mark each line with who said it: `V:` for the visitor, `B:` for the bot, `A:` for a person from your team. Put a line with three dashes between chats. That’s the format the free tool below reads.

Remove personal details. Before you paste anything anywhere, take out names, phone numbers, email addresses and order numbers. Replace them with a placeholder like [name] or [order]. You don’t need them for the review, and it’s a good habit under India’s DPDP Act 2023 and privacy rules elsewhere.

Keep the file. Save each week’s sample with the date in the file name. In a month you’ll want to compare, and having the raw chats beats relying on memory.

In LiveChatBot every conversation sits in the team inbox, whether the AI Agent, the keyword bot or a person handled it, and offline messages that became tickets are on the tickets list. The ticketing guide shows how chats and tickets link up.

Free tool

Tag last week’s chats in one go

Paste your chats in the V: / B: / A: format. The tool sorts each one into an outcome, counts them and lists the questions your team answered that the bot didn’t.

Paste last week’s chats. Start each line with V: (visitor), B: (bot) or A: (a person from your team). Put a line with --- between chats.

Runs only in your browser; nothing is sent or saved. Remove customer names, phone numbers and order numbers before pasting.

Reading the results

What each share is telling you

Once your chats are tagged, look at the shares before you look at individual chats. They tell you where to spend the next week.

A big "Bot missed" share

If more than about a quarter of chats end with a person answering something the bot couldn't, the bot doesn't know enough. That's good news, because the fix is clear: the tool's "Teach the bot next" list shows the exact questions and the answers your team gave. Questions that appear twice or more go to the top.

Any "Nobody came" chats at all

Even one is worth fixing. A visitor who asks for a person and gets silence is the most likely to leave and not return. Check three things: did anyone get an alert, were opening hours set correctly, and did the away form appear to collect a phone number or email? Our guide to opening hours covers the away form.

A big "Left after the bot" share

This pile is ambiguous on purpose. A visitor who reads a good answer and closes the tab looks the same as one who gave up. Read five of these chats in full. If the bot's answers were right and complete, most were probably helped. If you see short, vague or off-topic answers, that's a second teaching list.

Lots of "Person only"

If many chats never got a bot reply, check why. Maybe they came in while the bot was switched off, or the topics were ones you've deliberately kept for people, like payments or complaints. That's fine. But if routine questions are in this pile, the bot isn't reaching them.

Frustration flags

The tool marks chats containing phrases like "not what I asked" or "??". Read every one of them, whatever pile they're in. They're the conversations a customer is most likely to remember.

A healthy bot live chat setup, after a few weeks of reviews, usually shows a growing "Bot solved it" share, a shrinking teaching list and no "Nobody came" chats at all. The exact numbers depend on your business; what matters is the direction.

Teaching the bot

Turn your team's answers into the bot's answers

The teaching list is the heart of the review. Each item is a real question a real visitor asked, with the answer a person on your team gave. That's exactly what a bot needs.

For each item on the list, decide which of three things to do:

  1. Add it to the bot's knowledge. If the answer is general and stable (stock rules, delivery areas, sizing, how a service works), add it to whatever the bot answers from. In LiveChatBot the AI Agent answers from the knowledge you give it, so a clear paragraph or FAQ entry is enough. Our guide to training your AI chatbot covers how to write it.
  2. Add it as a saved reply. If you also run the free keyword bot, or want your team to send the answer in one tap, save it. Saved replies support {name}, so they don't sound canned. See saved replies.
  3. Keep it for people. Some answers depend on a specific order, account or situation. "Is XL in stock?" changes daily; "Can I change my address?" needs someone to edit the order. For these, teach the bot to hand over quickly instead of guessing.

Write the answer the way your best team member said it, then tidy it. Keep it short. Add the one detail visitors usually ask next. If an answer includes a price, date or limit, double-check it before it goes in; our number check guide explains why wrong numbers do the most damage.

Only take three to five items per week. Teaching ten things at once makes it hard to see which change helped, and some will be wrong the first time.

Decision table

From outcome to action

OutcomeFirst thing to checkTypical fix
Bot missed, a person stepped inIs the answer general or case-specific?Add general answers to the bot's knowledge or saved replies
Asked for a person, nobody cameDid an alert go out? Was it after hours?Fix hours, turn on app alerts, set the away form
Left after the botWas the bot's answer correct and complete?Rewrite weak answers; leave good ones
Asked for a person, got oneHow long until the first human reply?Holding line within 2 minutes; backup person
Person onlyShould the bot have replied?Check bot settings, hours and topics kept for people
Bot solved itNothingLeave it alone

The urgent pile

When visitors ask for a person and nobody answers

Treat every chat in the "nobody came" pile as a small incident. Go through it the way you'd go through a missed phone call from a good customer.

Find the time. Was it inside your opening hours? If yes, someone should have been alerted. If no, the visitor should have seen a message saying when the team is back and a way to leave contact details.

Check the alert. If your team relies on phones, make sure the Android app is installed, logged in and allowed to send notifications. A phone on battery saver can silently hold back alerts. Our Android app guide has the settings to check.

Check the fallback. LiveChatBot's Flows include a chat rescuer that can step in when a visitor is waiting, and the away form collects an email or phone number outside hours. Offline messages can become tickets, so a person follows up the next morning instead of the visitor being forgotten.

Follow up. If the chat included a contact detail, reply to it today, even late. A short "Sorry we missed you, here's the answer" often wins the customer back.

Then note what you changed, so next week's review can confirm the pile is empty.

The meeting

A 30-minute weekly agenda

The review works best as a fixed slot in the week, with the same steps every time. Monday morning suits most businesses, because last week is complete and fresh.

Minutes 0–10: collect. Copy 20 to 40 chats from last week into a text file, marked V:, B: and A:, with personal details removed.

Minutes 10–15: tag. Paste them into the tool above. Note the share for each outcome in a simple spreadsheet, one row per week.

Minutes 15–25: act. Fix any "nobody came" chats first. Then pick three to five items from the teaching list and add them to the bot's knowledge or saved replies. Read any chats with frustration flags.

Minutes 25–30: record. Write one line about what you changed and what you expect to move. "Added XL stock answer and address-change handover; expect fewer 'bot missed' chats about kurtas." Next week, check.

If two people share the inbox, do it together once a month. The person who answers chats will spot things the owner won't, and the other way round. Over time this weekly habit is what turns a bot live chat setup from something you installed into something that actually knows your business.

Checklist

This week's review

Weekly review

Getting started

Run your first review this week

Start small. Pick twenty chats from last week, mark the lines, and run them through the tool. Fix anything in the "nobody came" pile today. Teach the bot three answers from the teaching list. Write down what you changed. Next Monday, do it again and see whether the numbers moved.

If you don't have a bot yet, the same review works on your team's chats alone: the "person only" pile becomes your list of what a bot could take on. Our audit of what to automate helps you sort it.

LiveChatBot's free plan includes the team inbox, an AI Agent with a small free monthly allowance, a free keyword bot that answers from saved replies, Flows with a chat rescuer, opening hours with an away form, tickets and the Android app. It covers two websites with 100 visitors per website each month, with no per-seat charges. See pricing and start at livechatbot.in.

Frequently asked questions

How often should I review my bot live chat conversations?

Once a week suits most small businesses. It's often enough to catch problems quickly and spaced out enough that each week's change has time to show an effect.

How many chats do I need for a useful review?

Twenty to forty is enough for a small business. With fewer than twenty, treat the shares as a rough guide, because one unusual chat can swing them.

What should I fix first?

Any chat where a visitor asked for a person and nobody replied. After that, work through the questions your team answered that the bot could not.

Does the free tool store or send my chats?

No. It runs entirely in your browser and nothing is sent or saved. Still remove names, phone numbers and order numbers before pasting.

Should the bot learn every answer my team gives?

No. Teach it general, stable answers such as delivery areas or return rules. Keep case-specific questions, like a particular order or payment, for a person and make sure the bot hands those over quickly.

Keep reading

Related guides

πŸ“˜

Bot in Chat With Your Team: Who Speaks When, So the Bot and Staff Never Talk Over Each Other

Read the guide β†’

🧭

Bot Companies and Lock-In: Score How Hard It Would Be to Leave Before You Sign

Read the guide β†’

πŸ› οΈ

Bot Chat Website Drop-Off Finder: Why Visitors Leave Mid-Chat and How to Fix Each Exit Point

Read the guide β†’

A better bot every Monday

Answer routine questions automatically, hand the rest to your team, and improve it a little each week. Start free on livechatbot.in.