Skip to main content
Question

Getting difference DA results when running from the control room vs from a task bot

  • January 26, 2026
  • 6 replies
  • 158 views
Uzumaki
Forum|alt.badge.img+12

Hello,

I am getting different results from the document automation tool when it is run from the control room learning instance vs when it is run from a task bot. Has anyone seen this issue before?

When run in the task bot I am using the Document Extraction: Extract Data command and selecting the appropriate learning instance. I am not providing a LI version or any other additional settings. Within the learning instance I am using the ABBYY service provided and the Anthropic Claude model for Gen AI extraction

When run via the learning instance, the extraction is perfect and all of the fields are mapped exactly as they need to. When I run via a task bot, some fields are missing, some fields are in the wrong column, and the file outputs to the Validation folder. Note that this is only the case with one pdf of 5 that I am working with, the other 4 appear to work just fine when run from the task bot.

Any feedback on why I might be getting different results is appreciated. Thank you

    6 replies

    Lu.Hunnicutt
    Pathfinder Community Team
    Forum|alt.badge.img+19
    • Lu.Hunnicutt
    • Pathfinder Community Team
    • January 27, 2026

    ​@Uzumaki -- checking on this.

    Let's Connect! www.linkedin.com/in/lu-hunnicutt-1290jan
      Lu.Hunnicutt
      Pathfinder Community Team
      Forum|alt.badge.img+19
      • Lu.Hunnicutt
      • Pathfinder Community Team
      • January 28, 2026

      ​@Uzumaki -- I spoke with Alex. He said:
      A few ideas that can be verified on your side:

      • different learning instances are selected
      • different Advanced settings in the Extract Data action (BYOK settings)
      • additional pre-processing or post-processing at the process level
      • some hidden exceptions that happen during bot runner extraction (can't get access to feedback due to certificate issues, etc)
      • random inconsistency at the LLM level

      Let me know if this helps at all and I’ll circle back to him.

      Let's Connect! www.linkedin.com/in/lu-hunnicutt-1290jan
        Uzumaki
        Forum|alt.badge.img+12
        • Uzumaki
        • Author
        • Navigator | Tier 3
        • January 28, 2026

        ​@Uzumaki -- I spoke with Alex. He said:
        A few ideas that can be verified on your side:

        • different learning instances are selected
        • different Advanced settings in the Extract Data action (BYOK settings)
        • additional pre-processing or post-processing at the process level
        • some hidden exceptions that happen during bot runner extraction (can't get access to feedback due to certificate issues, etc)
        • random inconsistency at the LLM level

        Let me know if this helps at all and I’ll circle back to him.

        Hi Lu,

        I appreciate you digging into this.

        Maybe its possible that there is something preventing the extract data package from seeing the feedback results of the learning instance.

        • different learning instances are selected
          • Same learning instance in both
        • different Advanced settings in the Extract Data action (BYOK settings)
          • Additional settings in the extract data action are set to ‘none’
        • additional pre-processing or post-processing at the process level
          • I removed all pre-processing before testing all of this.
        • some hidden exceptions that happen during bot runner extraction (can't get access to feedback due to certificate issues, etc)
          • This one could be possible.
        • random inconsistency at the LLM level
          • Possibly, but it works every time when run from the learning instance and fails every time when run from the bot. Seems too consistent to be random

        I will continue to dig into it, but I appreciate the follow-up.

          Lu.Hunnicutt
          Pathfinder Community Team
          Forum|alt.badge.img+19
          • Lu.Hunnicutt
          • Pathfinder Community Team
          • January 28, 2026

          ​@Uzumaki -- I spoke with Alex. He said:
          A few ideas that can be verified on your side:

          • different learning instances are selected
          • different Advanced settings in the Extract Data action (BYOK settings)
          • additional pre-processing or post-processing at the process level
          • some hidden exceptions that happen during bot runner extraction (can't get access to feedback due to certificate issues, etc)
          • random inconsistency at the LLM level

          Let me know if this helps at all and I’ll circle back to him.

          Hi Lu,

          I appreciate you digging into this.

          Maybe its possible that there is something preventing the extract data package from seeing the feedback results of the learning instance.

          • different learning instances are selected
            • Same learning instance in both
          • different Advanced settings in the Extract Data action (BYOK settings)
            • Additional settings in the extract data action are set to ‘none’
          • additional pre-processing or post-processing at the process level
            • I removed all pre-processing before testing all of this.
          • some hidden exceptions that happen during bot runner extraction (can't get access to feedback due to certificate issues, etc)
            • This one could be possible.
          • random inconsistency at the LLM level
            • Possibly, but it works every time when run from the learning instance and fails every time when run from the bot. Seems too consistent to be random

          I will continue to dig into it, but I appreciate the follow-up.

          I’ll pass this response along and see if he’s got any other recs. Please keep me posted. If we need to hop on a call to walk through it, happy to do so. 

          Let's Connect! www.linkedin.com/in/lu-hunnicutt-1290jan
            Lu.Hunnicutt
            Pathfinder Community Team
            Forum|alt.badge.img+19
            • Lu.Hunnicutt
            • Pathfinder Community Team
            • January 28, 2026

            ​@Uzumaki 

            one more idea - the process can use Cloud Extraction, which is slightly different from the package extraction, thus the outputs might be different. 

             

            regarding hidden exceptions:

            • enable trace mode for bot runners
            • run the document through a task bot, collect a bot runner log here:
               C:\ProgramData\AutomationAnywhere\BotRunner\Logs

              it will be Bot_Launcher_...log with a user name.
              check if there are any exceptions in the log file.
              Ideally, compare logs with the same run through the process
            Let's Connect! www.linkedin.com/in/lu-hunnicutt-1290jan
              Uzumaki
              Forum|alt.badge.img+12
              • Uzumaki
              • Author
              • Navigator | Tier 3
              • February 2, 2026

              ​@Uzumaki 

              one more idea - the process can use Cloud Extraction, which is slightly different from the package extraction, thus the outputs might be different. 

               

              regarding hidden exceptions:

              • enable trace mode for bot runners
              • run the document through a task bot, collect a bot runner log here:
                 C:\ProgramData\AutomationAnywhere\BotRunner\Logs

                it will be Bot_Launcher_...log with a user name.
                check if there are any exceptions in the log file.
                Ideally, compare logs with the same run through the process

              Hi Lu,

              I appreciate the advice. Unfortunately I can’t figure out how to enable trace logs. I may put in a ticket with support to try to get 1-on-1 advice with this.

              Thanks though, I appreciate the help getting me this far

                Terms & ConditionsCookie settingsAccessibility statement