Tickets solved metric discrepancy in Updates history Explore dataset

9 Commentaires

  • Commentaire officiel
    Eugene Orman
    Zendesk Product Manager

    Nathan Cassella, thank you for sharing this issue with the community. We will add some notes about this behaviour in the recipe you mentioned. 

    CJ Graeme Carmichael,  thank you for jumping in to help and investigate this issue.

    There are two reasons for the discrepancy between these metrics in the Tickets vs Updates history datasets.

    1st reason

    The Tickets dataset does not include any deleted tickets but the Updates history does. If you want to remove the deleted tickets from your report you need to use some ticket level attributes in your report, like Ticket status or Ticket group. The deleted tickets will be removed because of the join with the tickets table. If you want to understand the structure of the datasets better refer to this article - Metrics and attributes for Zendesk Support.

    2nd reason 

    In order to capture the unique creation and resolution events we use the field changes data.

    • Capturing this event is straightforward for the Tickets created metric because there is only one event when the ticket status is set from NULL to something. So, the Tickets created metric is normally quite accurate. The only reason why it might be not accurate is when tickets in an account are created incorrectly via the API and the original ticket status is not NULL.
    • Capturing the resolution events for the Tickets solved metric is more complex because there are multiple resolution events per ticket. To capture only the last resolution event the default metric uses this condition "[Update - Timestamp]=[Ticket solved - Timestamp]". The issue is that in rare cases these two timestamps are not equal. This happens because these timestamps are not recorded in the ticket API at the same time but one after another. The best workaround we currently found is to create a custom Tickets solved metric that uses this formula and is added to the report with the D_COUNT aggregator: 
    IF ([Changes - Field Name]="status" 
          AND [Changes - Previous Value]!="solved" 
          AND ([Changes - New Value]="solved" OR [Changes - New Value]="closed")
          AND ([Ticket Status - Unsorted] = "Solved" OR [Ticket Status - Unsorted] = "Closed") 
          AND [Update - Date]=[Ticket Solved - Date]
          AND [Update - Minute]=[Ticket Solved - Minute])
    THEN [Update ticket ID]
    ENDIF

    We will review the formulas used for the default Updates history metrics at the beginning of 2023 and will update them if there is a better configuration.

    Conclusion

    • If you want to get 100% accurate stats on the number of existing tickets that were created and resolved in your account use the Tickets dataset. Create two separate reports one filtered by the ticket creation date and another one by the ticket resolution date.
    • If you want to compare the ticket creation and resolution volume in the same report use the Updates history dataset but be aware that these stats are 99% accurate. 
  • Graeme Carmichael
    Community Moderator

    Nathan

    I can recreate this.

    The metric for Tickets_Solved in the Support Updates History dataset is defined as:

    IF ([Changes - Field name]="status" 
          AND [Changes - Previous value]!="solved"   
          AND ([Changes - New value]="solved" OR [Changes - New value]="closed")
          AND ([Ticket status - Unsorted] = "Solved" OR [Ticket status - Unsorted] = "Closed")   
          AND [Update - Timestamp]=[Ticket solved - Timestamp]) 
    THEN [Update ID] 
    ENDIF

    For some ticket known to be solved, does not return the ticket ID. It is something to do with the last line on the conditions:

    AND [Update - Timestamp]=[Ticket solved - Timestamp]

    For some reason this is causing the count to fail.

    It may be best for you to reach out to Zendesk support directly.

    You can create your own metric as a work around. Replace the last condition with a different way of checking the dates matching and return Ticket ID rather than the Update ID.

    IF ([Changes - Field name]="status" 
          AND [Changes - Previous value]!="solved"   
          AND ([Changes - New value]="solved" OR [Changes - New value]="closed")
          AND ([Ticket status - Unsorted] = "Solved" OR [Ticket status - Unsorted] = "Closed")  
          AND DATE_DIFF ([Update - Timestamp],[Solved Timestamp],"nb_of_seconds")=0
    )
    THEN [Ticket ID]
    ENDIF

    1
  • CJ

    I can recreate this, too. This is definitely a problem that effects everyone. 

    Edit: I did some digging and it seems to struggle and not count tickets that were solved more than once, even if the second solve was in the timeframe specified. 

    1
  • Nathan Cassella

    Eugene Orman

    I appreciate you clearing this up.

    I have a few suggestions going forward:

    1. Use the same logic across all datasets when defining Tickets Created and Tickets Solved: Those numbers are too important to have even a 1% variance.

    2. Create a metric in all datasets for deleted tickets: This is a useful metric on its own, but building a custom metric to extract this from the counts is a lot of unnecessary work.

    1
  • CJ

    This answer is not accurate. Please, look into this issue further. I investigated as well, and it was very clear the issue was not deleted tickets for me, but that the Updates dataset incorrectly does not count tickets solved during the time period selected, if they were previously solved. I was also able to write a custom metric that functions correctly and returns the right number on this dataset.  The suggested metric would not fix this issue. 

    I don't understand why Zendesk would leave a bad metric that return misleading, incorrect data in place knowingly. I think it's completely unacceptable for any dataset to be less than 100% accurate. These are metrics that literally impact the hiring and firing of agents. There should be a strong a sense duty here to ensure accuracy.

    Edit: If you need further proof, you can recreate this issue with the Ticket Solved filter on the Updates History dataset, "Tickets Created" is not a necessary component to invoke this problem. 

    1
  • CJ

    As a follow-up, the suggested workaround is double counting solves if the ticket was solved multiple times.  I was suspicious when I saw it was returning *more* solves than the default dataset was. 


    That's not what the metric "Tickets Solved" should return, as this report is asking for the number of tickets solved, not the number of solves that occurred. You shouldn't use "update_id" as the value returned for this reason. 

     

    1
  • Nathan Cassella

    CJ

    I agree with you; my main concern is that any dataset that reflects Tickets Created or Solved needs to use the same calculation, which is not the case. So if I am being told to use the Default Ticket Dataset because that's accurate, then I will use that to get what I need for the presentation to my Executive Leadership Team next week.

    Based on the reply, there are two ways to proceed:

    1. Use the Ticket default set and the brackets to see updates. This is a workaround to at least see First Time Reply and First Time Resolution, but only IF the logic used to calculate those brackets can be trusted (which I'm not confident in). While we are using that, we hope that Zendesk with actually make the fixes to this, and it doesn't end up sitting the development hell (like the 14 years it took to get Round Robin ticketing into a beta)

    2. (And the better option) Get a real reporting system like PowerBI or one of it's competitors and pull that data you want from Zendesk's backend. This is likely where I'll move my organization in the next few years.

    0
  • Graeme Carmichael
    Community Moderator

    Thanks Eugene

    Hopefully, this default metric can be updated.

    0
  • Eugene Orman
    Zendesk Product Manager

    Thank you for following up on this issue.

    My apologies I forgot to update the THEN clause of the formula. It needs to count the Ticket IDs, not the Update IDs to avoid double-counting. I have updated the calculated metric formula that I suggested above. The formula should end with "THEN [Update ticket ID] ENDIF" not with "THEN [Update ID] ENDIF".

    Also, I think I found a way to make the formula more accurate. Please try this formula and let me know if it will return more accurate numbers in your accounts: 

    IF ([Changes - Field name]="status" 
          AND [Changes - Previous value]!="solved"   
          AND ([Changes - New value]="solved" OR [Changes - New value]="closed")
          AND ([Ticket status - Unsorted] = "Solved" OR [Ticket status - Unsorted] = "Closed")   
          AND  DATE_TO_TIMESTAMP([Update - Timestamp])>=DATE_TO_TIMESTAMP([Ticket solved - Timestamp])) 
    THEN [Update ticket ID]
    ENDIF
    -1

Vous devez vous connecter pour laisser un commentaire.

Réalisé par Zendesk