Chatbot analytics
Chatbot analytics is the record of what people asked your chatbot and how it did. It tells you how many questions came in, what share were answered, what they were about and where the bot had nothing to say. Deacon shows four figures against the period before, groups questions into topics, and keeps every conversation for you to read and search. It starts counting with your first conversation, on every plan, including the free one.
On every plan, including the free one.
At a glance
| The four figures | Messages, People, Answered and Waiting for a reply. |
|---|---|
| Answered | The share of the questions Deacon tried that it answered from your content. |
| Topics | Questions grouped by what they are about, most with a type such as How-to, Pricing or Bug report. A topic appears after two people have asked about the same thing. |
| Conversations | Every one, word for word, with search and filters for Unanswered, Wants a person and the rest. |
| Also on the page | The kinds of message people send, which sources answered them, the languages visitors wrote in and where in the world they are. |
| How far back | Seven, thirty or ninety days, or everything since you installed. |
| What it doesn’t have | Ratings, thumbs up or down, exports, scheduled reports or a link to Google Analytics. |
What Deacon’s dashboard shows you
Two pages hold it. Analytics has the totals and Conversations has the words.
Four figures against the period before
Messages is every question visitors put to Deacon, including the ones it declined. People is how many visitors sent them. Answered is the share of the questions Deacon tried that it answered from your content. Waiting for a reply is the people who left an email address and haven’t heard from your team, with a link to go and reply. The first three show their change on the period before and a small chart of the trend. Below them, a chart of messages per day marks the days you edited your knowledge, so you can see whether an edit moved anything. Chats in the Playground are left out.
Topics tell you what people ask about
Deacon groups questions that mean the same thing, across wordings and across languages, and names each group. Most topics get one of seven types, which are bug report, feature ask, how-to, pricing, account and billing, wants a person and feedback. The list can be sorted by how many people asked or by when it last came up, and filtered by type. Open a topic to read the latest questions in the visitors’ own words, with how many weren’t covered and how many people left straight after. Mark it done and it drops to the bottom. It comes back by itself if more people ask.
Every conversation, with search and tags
Conversations is the page you land on once you are set up. Its top line says what share of questions your agent is handling and how many gaps are left to answer. Each row carries the tag that fits its outcome, and the filter shows a count beside each. The tags are Unanswered, Wants a person, Failed, Over limit, Answered and Dismissed. Search looks through what visitors asked, and the dates can be set to today, this week or this month.
One conversation tells you why
Open a conversation and you see every turn, the source each answer came from, the page the visitor was on, whether they have chatted before and the topic the question joined. Anything your app has told Deacon about the visitor, such as their plan, is shown alongside. A conversation in another language can be translated with one click. Any question Deacon couldn’t answer has an Answer this button under it. Write the answer and the next visitor who asks gets it, and the gaps left to answer at the top of the page goes down by one.
Analytics starts with your first conversation
Install Deacon and the counting begins. There is nothing to set up and no report to build. You get 50 answers a month on the free plan, and it needs no card.
Which chatbot numbers are worth reading
Every tool has a dashboard. These are the figures to look for in any of them, and how to read each one.
The share answered
This is the share of questions the bot answered without a person. Vendors call it the resolution rate or the containment rate, and each counts it a little differently. Some count a visitor who gave up as contained. Read the definition before you compare your figure with anyone else’s.
What went unanswered
The list of questions the bot had no answer for is worth more than any percentage. Each one is either missing from your docs or something customers want that you don’t offer. Work through the list every week.
How often a person was needed
The escalation rate is the share of conversations passed to a person. It matters most when one question is behind most of the handovers, because that answer could be written down once.
What people ask about
Counts by topic tell you what to fix in the product and what to write in the docs. Thirty people asking how to export invoices means a page is missing, and thirty asking for an integration you don’t have belongs in your planning.
Be careful with deflection
The deflection rate counts the questions that never reached your team. A visitor who closed the chat in frustration never reached your team either. Read real conversations alongside it.
Read ten conversations a week
Reading is the only way to tell whether the answers were any good. Ten conversations a week is enough to catch a wrong answer, a missing page or a tone you don’t like.
Questions about chatbot analytics
No. Deacon has no link to Google Analytics and no export. Its own website analytics is on the next tab, fed by the same script, so you can see visitors and goals beside what people asked.
No. There is no rating, no thumbs up or down and no survey. What it has in their place is the Answered share and the list of questions it couldn’t answer.
A reply Deacon wrote for the visitor. One that said it didn’t have the answer counts as not answered. Questions that failed, and ones that arrived after the month’s answers ran out, are left out of the sum on both sides.
Not that month’s. Each figure records what the visitor got at the time, so it stays as it was. What changes is what happens next, and the gaps left to answer, which goes down as you work through them.
Each question is compared with the ones nearest to it in meaning, and a model judges whether the two ask the same thing. A question joins a topic only when it matches that topic’s typical questions. A topic is listed once two people have asked within the dates you chose, and its name is always in English.
Not today. There is no export and no scheduled email, so the dashboard is where the numbers are.
Yes. Every figure, topic and conversation is on every plan, and nothing about analytics is limited by plan. See the plans
Read your first conversations
It takes about five minutes to install. The free plan is 50 answers a month, for as long as you like.