- Release notes
- Task Mining overview
- Setup and configuration
- Notifications
- Task Mining
- Additional resources
DEPRECATEDBest practices
Unassisted Task Mining is designed to mine the unknown by recording your employees’ desktop activities and discovering repetitive tasks suited for automation and process improvement.
When starting to use Unassisted Task Mining, a best practice is to always present the rationale and the expected result to your end users.Unassisted Task Mining only records the relevant parts of the work end user activity, specifically screenshots and metadata, into use AI and create a clear picture of the task at hand.
Key call-outs include:
-
The primary purpose for recording actions within the task is to find variations in task workflows for further optimization, not to track user level actions.
-
Once recording has started, it will capture everything that has been allowed in the settings, so it is important to prepare the environment for recording prior to selecting start.
- Only applications that have been allowed will be recorded, so applications that users are concerned about can be denied for recording.
- Personal Identifiable Information masking is on by default and can be configured.
- User anonymization is always an option.
Overall, the goal of recording this data is to understand the inefficiencies in the task better in order to optimize the team/task further. Explaining the why as to your reasoning behind doing this is important.
Navigate to your Task Mining service and create an Unassisted Task Mining Project. It's best to create a project per team/department.You should provide a unique and relevant title and description during project creation. Add up to 10 users to your project, as the recommended amount is 2–7 users. For best results, it's recommended to have experienced users that are familiar with the day-to-day job.
Each user receives an email asking them to download and install the recorder. Check out the recorder settings page to configure the recorder to your specifications.
Indicate which applications you’d like to apply or deny recording to as this limits noisy data collection. It's recommended to apply recording to applications you know are core to the task at hand.Specify URLs for web applications as precisely as you can without limiting the recording too much.
Example: A project might deny Slack since data collected on communication outside of the task completion isn't part of the task steps and might be considered noisy data. This frequently applies to other communication tools as well.
Communication tools like Slack, Zoom, and Microsoft Teams are inherently noisy as recording users will navigate through various different messages. It is not recommended to record these applications as they impact the output of the tasks from the AI analysis.
By adding a website to the allow list, you are giving permission for that specific site to be recorded regardless of what browser is used to access it. Even if the browser itself is not explicitly added in the allow list, the permitted website will still be recorded.
Provide each recording user context on the goal of the project and the amount of recording necessary. The recommended amount of actions collected is 20k-70k to limit noisy data. As a quick way of estimating, an active user typically provides 200-1,000 actions per hour. Indicate the importance of recording only relevant task-related content or pausing the recorder when they have to navigate away from day-to-day tasks.
Tips for recording:
-
Finish the task that you start → minimize interpretation/breaks during the recording.
-
For authentic and accurate results, the behavior during the recording should remain the same.
-
Limit task switching, if possible, while recording. Minimize the usage of non-supported apps.
-
Pause recording if other things come up during the completion of a task.
From accuracy and financial wise, it's best to run the evaluation one time after reaching the threshold and finishing the recording.
Be certain when running the AI evaluation pipeline that you are ready to receive task output, as running an analysis consumes 5,000 AI Units per analysis, regardless of the number of actions.
Once you’ve received your discovery results, review the tasks with a business process Subject Matter Expert following the Unassisted Task Mining Analysis Guide.
Tips for task analysis:
-
All detected tasks are ranked by default based on repeatability.
-
The table that shows all detected tasks contain lots of insights on the task potential automation value.
-
When reviewing tasks, we can expose all detected steps by toggling off the key steps only. Additionally, hovering will expose additional information on each parameter in the graphic.
-
When reviewing a selected trace, the admin can see all screenshots taken from the user who performed it. While viewing a selected step, the admin can see all the screenshots taken from all the users' actions in that step.
