Announcing the next phase of Automatic Answers Follow

Earlier in the year we launched the Early Access Program (EAP) for Automatic Answers - an initial alpha feature set. Today Automatic Answers has moved into an open beta stage, meaning the feature is more mature now having been validated by customers, iterated on, and extended substantially since the initial EAP began. There's no more waiting list, if you sign up now you'll be accepted into the program generally within 2 working days.

What is Automatic Answers?

Very simply: Automatic Answers adds artificial intelligence to self-service. Questions from customers are interpreted and relevant knowledge base articles are suggested. Customers are happier with the faster resolution and the support team can focus on things only people can do.

For more information, take a look at zendesk.com/automatic

To find out more about how the feature is setup and used, read the Using Automatic Answers article.

Joining the Early Access Program

To join the Automatic Answers Early Access Program, head to Zendesk Automatic Answers. and click Sign Up for Early Access.

Eligible customers must meet the following requirements:

  • The feature only works in English currently, so you will need to configure it accordingly

  • Must have 10 or more detailed, published articles in English that speak to the problems your customers commonly face.

What's changed between the first release and now?

1 - The model

When the EAP started, we built individual models for each customer, this meant we could only offer the feature to select accounts with the required data volume needed to effectively build the model. Since then, we have been working a new, deep learning model that can be used by any customer without requiring as much or any data. Simply create enough articles that represent the problems you commonly see and turn the feature on! That easy.

2 - The end-user experience

We've adapted and refined the end-user experience substantially. We now have a fully integrated Help Center experience: when a user clicks on an article, they are taken to the article in HC and a widget appears allowing them to solve their request right there, if the article was correct. 

3 - Configurability

Admins now have more control over the experience than ever before. We have 3 placeholders that are used in the Automatic Answers trigger that allow customisation over how the Automatic Answers email is presented. Admins also now have more control targeting specific articles based on labels.

4 - Reporting

We now provide detailed deflection reporting in Insights to ensure that you have full visibility into the feature performance. Understanding what self-service value you are driving with the feature has never been easier. 

 

Have more questions? Submit a request

Comments

  • 0

    Hello Mike,

     

    is it an option for you to develop such automatic answers for Zendesk Chat as well (e.g. ChatBot).

    If yes, might it be an option let i integrate with our knowledgebases than the one from Zendesk Support as well?

     

    Hope to hear from you soon! Thanks!

  • 0

    Hey Kundenbetreuung, we are definitely considering how to apply such a solution to other products and channels at Zendesk while ensuring that the delivery mechanism (chatbot or other) is the most well suited to each product / channel. In terms of integrations with other knowledge bases, we'll need to see... Currently we feel we can deliver the best experience with a deep and seamless integration to Help Center

  • 0

    When we sign up for the Early Access Program, is the detailed deflection reporting in Insights available immediately?

  • 0

    Hey Pete, we're just finalising the initial default dashboard and we should have it activated on the EAP customer insights projects within 48 hours.

  • 1

    Mike, that's great news, thanks! Another question: can you provide more explanation on how "artificial intelligence" works? Our HC search field appears to use Instant Search, which displays article suggestions, but is apparently limited to article titles. Our ticket submission web form uses Smart Search, where article suggestions are displayed based on article titles AND article content. In the sub-section above titled "Configurability", it mentions the usage of labels. I'm trying to understand what our expectation should be with Automatic Answers. Thanks, Pete    

  • 0

    I would also like an answer to Pete LeRoy's question posted on Monday at 15:15 about how it chooses which article to show. 

  • 0

    Hi Pete (and Brian), Automatic Answers is powered by a deep neural network that understands the language of customer support. What this means is that instead of matching keywords, it tries to understand the meaning behind sentences in a customer query in the subject and description of the ticket. We then locate help center articles that address similar topics. It returns the top 3 closest articles to the question being asked.

    Your expectation should be that this uses a completely new and different way to interpret and select the suggested articles - we built it this way because our aim is to offer self-serve capabilities for email and most emails are long form questions (not keyword oriented as you would find in search or even the ticket submission web form).

    Let me know if there are further questions?

    Thanks!

  • 0

    Naturally, the next question is if/when will the "deep neural network that understands the language of customer support" be available in other deflection hotspots like the widget, HC search, support ticket requests, etc.?

  • 1

    Mike, thanks for the response to my question, but I could benefit from having specifics on the logic. I've created several test tickets so far and the resulting Automatic Answers are not relevant to the request. One of our KB articles is titled "HTML Widget". I tried to get this article to display as one of the suggested articles in the Automatic Answers notification. So, I submitted a ticket using "HTML Widget" as the email subject line and "Does G5 Marketing Cloud have an HTML Widget?" as the email content. The suggested articles didn't include our "HTML Widget" article. Your previous reply states that Automatic Answers does not match keywords, but still, why would the deep neural network not produce it as a result so obviously a match? So far, I favor the logic used in Smart Search much better. Thanks, Pete          

  • 0

    Mike / Pete - I'm having the same exact issue. 

    This is the thing that's really keeping us from launching automatic answers. We currently have our top 10 FAQ answers in our existing auto responder. I am putting in test support tickets trying to get automatic answers to at least pull up the most relevant articles for these top 10 and we're struggling.

    I believe it said that it gets better over time. How does that work? 

    Does it also look at thumbs up / thumbs down on the help center article itself?

  • 0

    Will automatic answers account for articles that are not available to the user. 

    For example, if I have a section set up just for members of the organization "Microsoft", if a user from organization "Apple" searches the helpcenter, they will not see articles for "Microsoft" that would otherwise match search criteria.

    Will this be the case for Automatic Answers, so "Microsoft"-only articles are not presented to "Apple" users?

  • 0

    Hey everyone!

    Joel, great question - we are definitely considering the best way to integrated this functionality into other channels, in the most effective way possible.

    Pete, can you send in the ticket IDs to automatic-answers@product.zendesk.com so we can take a look at your specific examples to understand what's going on? It should match so there may be an issue with your model or we don't have all the articles embedded.

    Brian, same deal - please send us some example ticket IDs and the articles you think should have been suggested so we can look into it.

    Steve, we built Automatic Answers to work with the existing restrictions you've created in your help center. Only the articles that each user has access to will ever be returned back to them. If you've built those sections and made them restricted, you can be assured that we enforce those same restrictions using this feature! :)

    Thanks for the questions and feedback - keep 'em coming!

     

  • 0

    Mike, I just sent an email with the ticket ID I used to test. Thanks for your attention to this! Pete  

  • 0

    Hi Mike, 

    We're trying to get Automatic Answers to use an organization field as in input, i.e our knowledge base responses are highly contextual so this field is an important signal for any search / neural net. Is this possible? 

    Cheers,
    Richard

  • 0

    No Deflection tab in Insights, we already activated this since last week. We want to check the report to verify if Automatic Answers Program is suitable or helping us.

Please sign in to leave a comment.

Powered by Zendesk