How can we improve the relevancy of articles suggested by the Suggest Help Articles Task Bot? | Community
Skip to main content
Answered

How can we improve the relevancy of articles suggested by the Suggest Help Articles Task Bot?


Forum|alt.badge.img+1

Hi all, we just turned on the Suggest Help Articles Task Bot for both users and leads and in the first week of use we are finding its article suggestions to be way off. It seems to be suggesting just our oldest/highest viewed articles without taking into account keywords (see below for an example).

 

I've already read Intercom's "Let users help themselves with smart suggestions" and "Share articles to answer questions faster". I've done the suggestions about writing relevant articles with the words the customers are searching for.

 

The suggestion about -"When a customer gets in touch, you should send them a relevant article in your first reply to them. If they send you a second message before you reply Operator won’t learn from this." - really doesn't work for my company. When our users actually get to talk to a live agent they expect targeted, step-by-step support, not just guidance on where to find an article and especially not on the first reply.

 

Is there any other way I could improve the task bot's accuracy? As it is now, I'm not sure we'll keep it turned on since its suggestions are not getting the content that is on our help center to our users.

 

Example: One of our top inquiries is about "unfinalized tracks". We have one main article that is titled "unfinalized tracks" and mentions that term in the article body about 20 times. Then we have 2-3 other articles that have that same keyword about 1-4 times in the body or title. Well the Task Bot has not suggested the "Unfinalized Track" article once in multiple instances where a user has chatted us asking about "unfinalized tracks" even when they write that keyword several times in their message. But it is suggesting the articles where the keyword appears just once or ones where the keyword doesn't appear at all!

Best answer by Eric Fitz

Hey @user32​, the primary method by which the Articles Suggestion bot improves the quality of its suggestions is from you and your team sharing articles manually with your customers. However, you've mentioned that this doesn't fit your core workflows, so I totally understand this may not be an option for you.

 

Have you considered letting your customers browse your Help Center directly from within the Messenger? Our internal research, when we tested this feature for our own customers, has found that this leads to a significant amount of deflection, which could save your support team time and effort!

View original
Did this topic help you find an answer to your question?

3 replies

Eric Fitz
Employee
Forum|alt.badge.img+5
  • Employee
  • 1630 replies
  • Answer
  • November 5, 2021

Hey @user32​, the primary method by which the Articles Suggestion bot improves the quality of its suggestions is from you and your team sharing articles manually with your customers. However, you've mentioned that this doesn't fit your core workflows, so I totally understand this may not be an option for you.

 

Have you considered letting your customers browse your Help Center directly from within the Messenger? Our internal research, when we tested this feature for our own customers, has found that this leads to a significant amount of deflection, which could save your support team time and effort!


Forum|alt.badge.img+1
  • Author
  • Active User
  • 10 replies
  • November 5, 2021

Hi @eric f11​ thanks for clarifying this. It is indeed disappointing because we have feedback from our users that being provided with an article upon first reply is not a good support experience. If the task bot could learn when we sent an article over the course of a reply, that would be better. But ideally the task bot would learn based on keywords in the article itself, or that we could give it. We'll most likely be turning this feature off.


Eric Fitz
Employee
Forum|alt.badge.img+5
  • Employee
  • 1630 replies
  • November 8, 2021

I can totally understand the decision to turn it off here, given the reasons you've outlined.


Reply