September 17, 2019
Building a better chatbot with chat analytics
Use FAQ Bot’s powerful analytics features to continually improve your bot.
Use FAQ Bot’s powerful analytics features to continually improve your bot
FAQ Bot’s chat analytics features make your customer’s interactions with your chatbot visible, measurable and actionable. You can see the actual questions your customers asked and how the bot responded. This provides rich voice of customer insights, and makes it easy to identify gaps in your knowledge base – areas where you could add new question variants or entirely new Q&A pairs you hadn’t previously considered. Finding gaps in your FAQ Bot knowledge baseHere are some quick ways you can use FAQ Bot’s built in chat analytics to identify gaps in your knowledge base.
Top up your knowledge base by reviewing questions answered by web search
This chart on the Analytics Overview screen provides a quick, visual way to review your bot’s message handling. In the example above, 61 answers were provided by web search.
Any questions answered by web search are likely to already have an answer on your website – so adding them to your knowledge base is quick and easy.
You can also monitor the proportion of questions answered by web search over time – if your bot’s improving, you should see the proportion decrease.
>>FAQ Bot tip Click on the different segments of the chart to quickly see the messages in a category:
Review any questions your FAQ Bot couldn’t answer
In the knowledge base insights tab, you can segment answers based on confidence – confident answers, uncertain answers or unanswered questions.
Unanswered questions are those that the bot couldn’t match with an answer in the knowledge base. You can review the exact wording as typed by your customer, question by question. You can then decide whether to add those questions: as a wording variant to an existing question (search your knowledge base first to see if a relevant question already exists); or as a brand new question and answer pair. Using this quick review mechanism, you see what your customers are actually asking, in their own words, and build your knowledge base accordingly.
Review uncertain answers to see if you can improve the matched results
If your bot isn’t confident the answer is a close match to the question, it will ask “did you mean?” and present possible answers to your user.
You can review these “did you mean?” type responses in the Uncertain Answers tab, to see why the bot isn’t replying with a confident answer.
It may be a wording issue – your customer asks a question in a different way or using different language than you predicted when you created your knowledge base. You can address this by adding a variant to an existing question in the knowledge base.
Uncertain answers can also highlight knowledge base gaps, which you can address by adding a new question and answer pair to your knowledge base.
How much time should I spend reviewing my chatbot’s analytics?
We reckon any time spent building a better understanding of what your customers are asking, searching for and saying, in their own words, is time well spent 😀
Generally, you might want to plan to keep a closer eye on your analytics immediately following the launch of the bot, to help you quickly spot any issues you hadn’t foreseen.
Then, as you make improvements, your knowledge base will be more complete, your bot will be better trained and you’ll need to spend less time making adjustments.
For those of you who like graphs, it looks a bit like this:
No time to review your chat analytics? Contact us to discuss our FAQ Bot monitoring service.