Working with Satisfaction Prediction Scores

Return to top
Have more questions? Submit a request


  • Wouter van Gessel

    Wow, this seems like an amazing tool! Customer service is our USP, therefore we are very eager to use this. 

    In this article you mention "you need a minimum of 500 satisfaction ratings", however in another article ( you mention "this feature also requires a minimum of 500 satisfaction ratings per month".

    We do have more than 500 ratings in total, but only around 150 per month. We really hope we can use this feature. Can we and if not, will it be possible in the future?

  • Omari Whyne

    I second this comment!

    Look forward to hearing a reply.

  • Josh Greenwald

    Maybe there is a way to train the model on historical data? Maybe an ask for one of their internal teams ?

  • Tsvetan

    Same here, if we have that minimum of 500 in total, but each moth these are less than 500, is this going to work?

  • Aimee Spanier
    Zendesk Documentation Team

    Hi, all. Sorry for the error - as I understand it now, you need 500 *per month*, and should have at least 1,000 ratings total if possible.


  • Charles Magnuson

    It seems this feature isn't working as described. We receive around 2,000 CSAT ratings per month, ~500 per week, yet I am told we don't have enough ratings:


    My interpretation is that we all need to wait at least one month for Zendesk to build up enough data about our accounts for this tool to begin working correctly. @Zendesk, can you confirm this?

  • Leandro Gomes

    Only enterprise? or not!.... 


    What you doing zendesk?

  • Alfred Diez

    This is very cool. We get escalations sometimes and it's great to be able to have the system predict which ones may go side ways before it actually occurs.

  • Jon Thorne

    We've enabled this and didn't get the same message Charles did, but we haven't got the predictions showing yet. As per Charles' message, my questions is also:

    "My interpretation is that we all need to wait at least one month for Zendesk to build up enough data about our accounts for this tool to begin working correctly. @Zendesk, can you confirm this?"


  • Jessie Schutz
    Zendesk Team Member

    Hi Jon! Sorry about the delayed response here.

    Support is currently working with Product to find out what's going on here. I'll be sure to post here when we get it figured out!

  • Jessie Schutz
    Zendesk Team Member

    Hey Jon!

    Sorry for the delayed response. Here's what I was able to find out about this:

    "Satisfaction Prediction uses a Machine Learning process to train a model for your account. Although you meet the minimum initial criteria to have a model built, there is a validation process that is applied to the model to ensure it can accurately predict the outcome.

    If the model doesn't perform well enough, then the feature cannot be enabled. To overcome this problem we simply need more data. It is possibly due to the types of tickets you have - if they are too varied, with insufficient good and bad ratings then the model won't perform well. The automated model building process will continue to attempt to build a new model for your account. Once it is able to successfully build a model that meets the threshold, the feature will become active and your account administrators will receive an IPM and email to notify them."

    Please let me know if you have any other questions!

  • Andrea Brown

    I'm curious about how satisfaction prediction works with Hub and Spoke. Will the prediction come over to my Hub account with the ticket? I see that I need to turn it on at the Spoke level, but I'm curious about whether I'll need to also log into the Spoke to see this feature.

  • Megan Howell

    @Andrea, The satisfaction prediction stays inside the Spoke accounts and isn't carried to the Hub, so you will need to log into the Spokes to see those predictions! Sorry for any inconvenience this might cause, but satisfaction is generally separated so one Spoke doesn't have a great influence on all the others. Let us know if we can answer anything else!

  • Wouter van Gessel

    We were very enthusiastic about this feature at start. After using this for 1 month, we find that the predictions are way off, making this function useless. Therefore we've removed the function. Perhaps it works well for larger companies with thousands of tickets per day.

    CAUTION: We did find out that clicking "Enable satisfaction prediction scores", automatically enabled the System Automation "Request customer satisfaction rating".

    This caused a month of unwanted satisfaction rating requests emails to our customers. Bad bug Zendesk...

    Automations should never be activated without the user knowing...!

  • Mike Mortimer
    Zendesk Product Manager

    Hi Wouter, thanks for sending your feedback through. I'll try answer the 2 parts of your experience, first - the prediction scores: Because of how we normalise the prediction scores (to ensure stability for your automations / triggers) the score itself may not seem accurate - we are currently reviewing alternative options for showing the score but the best way to interpret it is to take your long term CSAT rating (i.e. your current CSAT score overall, over a period of time). 100 - this score should give you an accurate threshold for where a ticket moves between good and bad, e.g. 100 - 88 = 12. Scores below 12 are high risk for negative ratings. We've validated this across many accounts using the feature.

    In terms of the second issue - this is actually not related to satisfaction prediction but rather an unfortunate bug related to the core CSAT feature where saving that form again will cause the system to interpret this as activating CSAT even though it was already active. This is now a known issue and should be resolved shortly. Indeed, if you simply updated a standard CSAT related setting then the re-activation of the trigger also would have occurred. 

  • DJ Jimenez

    We have multiple teams using Zendesk but not all of them are using satisfaction surveys.

    • How do the tickets belonging to the teams not using CSat affect the satisfaction prediction scores, or do they?
    • Is there a way to isolate the machine learning to tickets for specific teams?
    • Is there a way to isolate the satisfaction predictions to tickets for specific teams?
  • Claudia Pelagatti
    Zendesk Team Member

    Hello DJ,

    I will address your questions below:

    1. Tickets that don’t get a CSAT score don’t get used in model training, and won’t influence satisfaction prediction scores.
    2. No, there isn’t. The ticket ingestion pipeline that feeds the model training process cannot discriminate on the basis of teams.
    3. Unfortunately no. Satisfaction prediction is enabled globally, there’s no way to configure which teams will receive predictions. You can customize whether or not an agent will see a score at the top of a ticket thread, but even when not showing the prediction a score is still calculated.


  • Tim von Gimborn

    Is there a way to route tickets with low satisfaction prediction to certain agents that have the best CSAT in their group / skill? Can the satisfaction prediction model automatically select the best agent for the ticket?

  • Sharon Couto

    Not all tickets are displaying a satisfaction prediction score. Currently, there are no automations or triggers set up that could be interfering. What could be causing this?

  • Dave Dyson
    Zendesk Community Team

    Hi Sharon,

    As noted at the top of this article, we will be discontinuing the Satisfaction Prediction feature on September 30, 2021. This feature was launched in 2016, and no longer meets our reliability standards, so those issues may be what's causing the behavior you're seeing.

    You may be able to find some third-party solutions that can provide similar functionality by searching for "sentiment analysis" in our Apps Markeplace: Sentiment analysis


Please sign in to leave a comment.

Powered by Zendesk