Task Mining
latest
false
Banner background image
Task Mining
Last updated Mar 26, 2024

FAQs

General

How is Task Mining different from Task Capture?

  • Task Mining is AI-driven and applies machine learning to mine user desktop activity across users to determine automation candidates. The word mining implies the nontrivial extraction of previously unknown and potentially useful information from activity data.

    Task Capture is human-driven and captures the subject matter expertise and assembles the ideal step-by-step workflow of a known process. Experts can transform siloed knowledge into organizational wisdom by “taking selfies” of their expertise.

    Both products can expedite automation by generating PDD without messy manual screenshots or templates and export a skeleton workflow to Studio in seconds.

What is the difference between Process Mining and Task Mining?

  • Task Mining and Process Mining complement each other to generate a holistic view of how work gets done across the organization. Task Mining focuses on user desktop activities. It mines useful information from activity data across users to determine automation candidates. Process Mining helps discover automation potential from information trapped in business systems (like ERP and CRM) and provides a detailed understanding of your complex business processes.

What is the difference between Assisted and Unassisted Task Mining?

  • Assisted Task Mining is one of the Task Mining approaches that empowers your employees to capture variations of a known task and easily merge these into a comprehensive picture of the existing process.

    Unassisted Task Mining is another Task Mining approach that lets you mine the unknown area by recording your employees' desktop activities and discovering repetitive tasks suited for automation and process improvement.

What is the difference between Assisted Task Mining and Task Capture?

  • Assisted Task Mining can be considered an evolution of Task Capture. Both of the products cover the area of capturing a known task. However, there are several differences between them:
    • Task Capture is a full desktop application, while Assisted Task Mining is a combination of both - the admin portal on Automation Cloud (for project management, data storage, collaboration, etc;) and the desktop recorder client for capturing the traces.
    • Assisted Task Mining speeds up the process of variations comparison and automatic merging of multiple traces into a full picture. With Task Capture, it’s possible to work with one task graph at a time and there's in-built AI to cover this need.

How can I try Assisted Task Mining?

  • Assisted Task Mining is available as part of the Task Mining service experience.

    Task Mining service can be enabled for all Automation Cloud organizations included with all enterprise licenses starting 22.10 general availability.

What are the benefits of Assisted Task Mining?

  • Check out the Introduction page for Assisted Task Mining to find out more about the benefits.

Does Assisted Task Mining work on RDP/Citrix environments?

  • Yes, it's possible to capture Citrix applications with the Assisted Task Mining recorder. However, the generated selector quality, in this case, will be on a basic level and the screenshot will show the entire Citrix window, not the specific active application. It is recommended to install the client on the Citrix environment rather than on the desktop for capture.

Does Assisted Task Mining use a machine learning model for the merging? Does this consume AI units?

  • Yes, the ML model is used for the auto-merge functionality. No AI Units are required for now to use this. The AI Units are required for Unassisted Task Mining projects only.

Can Assisted Task Mining be used on Mac?

  • Assisted Task Mining consists of two parts - the web-based component for project management, data storage, trace editing, etc. and the desktop recorder client. The web-based part is cross-platform as it’s on cloud only for now, so it can be used on Mac OS. However, the desktop recorder can be used on Windows OS only.

    Please read the Hardware and software requirements page for more information.

Can Assisted Task Mining support languages other than English?

  • Assisted Task Mining is localized in all the same languages as the rest of the UiPath platform.

    Assisted Task Mining isn't language specific, though it has been trained and tested on Latin character languages.

Does Assisted Task Mining mask PII data?

  • Yes, the PII data masking is enabled by default. PII masking is provided by Microsoft Cognizant Services, you can view Azure Cognitive Services page to find out more.

Will current Task Capture features be incorporated into Assisted Task Mining in the future?

  • As of 22.10, the Assisted Task Mining contains core features of Task Capture + brings new value with the merging functionality, collaboration and sharing capabilities, etc.

    Our plan is to incrementally improve the product by bringing more valuable Task Capture features in 2023+, such as action editing, export to StudioX, etc.

    Share your feedback to let us know what we’re missing now!

Can we export the task graph from Assisted Task Mining and open it in Task Capture?

  • At this time this functionality doesn’t exist, but it’s on the backlog.

Can we upload Task Capture projects to Assisted Task Mining?

  • No, there's no option to reuse Task Capture projects in the Assisted Task Mining.

Where can I watch a demo of the Assisted Task Mining features?

Artificial Intelligence system

How is transparency ensured in the Artificial Intelligence (AI) system for fairness across different demographic groups and mitigation against potential biases?

  • Our AI and Machine Learning (ML) algorithms are designed around clustering screenshots of desktop activities. We do not take any human attributes such as demographics, age, gender, and race into account as these data points are irrelevant for the problem we’re solving. Our algorithms classify which screens came from the same application across machine themes, window sizes, browsers, and operating systems, etc. Training for these models is done using data generated by UiPath, although in some older projects previous customer data could be used.

Are AI-generated outputs securely logged, timestamped, and attributed to specific AI agents to prevent repudiation in legal or compliance contexts?

  • We have unique ML models for specific tasks. With these models, we can track which decision was made where in our pipelines.

How are model versioning and deployment tracked to maintain an audit trail for accountability?

  • Model updates are tracked by UiPath release and change management system. We share updates publicly with customers via our release notes.

When someone changes role or leaves an organization, is their access removed?

  • Yes, access is removed when changing roles or leaving an organization.

How are privacy and data security concerns addressed when AI models process sensitive or personally identifiable information?

  • PII data is removed from the collected images by using Microsoft Azure Cognitive Services before they are processed by the AI models. The content is not what's important for the screens, it's the context of the screens that matters.

Are there mechanisms to enforce data anonymization or differential privacy to mitigate information disclosure risks?

  • Yes, in addition to removing PII data, customers can choose to anonymize who the data was collected from.

How do you ensure any personal data is not shared among other tenants in a multi-tenant environment?

How are user-generated prompts filtered to prevent the AI system from disclosing sensitive data that is used to create inferences but not allowed to be exported raw?

  • There are no user generated prompts in UiPath Task Mining.

How is the AI model protected against unauthorized copying, distribution, or reverse engineering to safeguard intellectual property?

  • The AI Models are run on UiPath servers and are not exposed directly to customers.

Are there techniques employed to watermark AI models to trace their origins and ownership?

  • No, we don’t employ techniques to watermark AI models as they are not exposed directly to customers.

What protections are in place to defend against adversarial attacks targeting the AI system, such as input perturbations, model overloading, or resource exhaustion?

  • We have throttling mechanisms in place on the server side to prevent adversarial attacks. In addition, our AI system is not directly accessible by customers.

How do you prevent unauthorized access to AI model parameters, as well as their modification or extraction?

  • AI model parameters are not exposed to customers. They are configured on the server side and controlled through change management/release/DevOps systems operated by UiPath.

Assisted Task Mining

Recorder client

Can I use the 2022.4 and older Task Mining recorder client for Assisted Task Mining projects?

  • Starting with 2023.4 release, there is just one unified recorder client. It can be used to capture data for Assisted Task Mining projects and just by switching the project, you can record for Unassisted Task Mining projects.

What are the software requirements for the Assisted Task Mining recorder client?

Can the Assisted Task Mining recorder work offline?

  • Not yet, an internet connection is required to capture the Assisted Task Mining or Unassisted Task Mining traces.

ATM project management

Unassisted Task Mining requires a minimum of 50K of actions to be captured. Does Assisted Task Mining also have a minimum or a maximum number of required actions?

  • There is a maximum of 500 actions for each trace in ATM . Our guidance is to create distinct traces for individual tasks and capture them in less than an hour.

Unassisted Task Mining is limited to working with up to 10 users for a project. How many users can be invited to an Assisted Task Mining project?

  • There is no limitation on the number of users invited to the project.

Unassisted Task Mining

Recorder application

What needs to be installed on the client desktops to run Task Mining?

  • Each user needs to install the Client and the Processing Queue to capture their tasks. For details, please check the Installing the Recorder from the User Guide.

Does the recording application support a multi-monitor setup?

  • Yes, it does. Depending on the setup, it might consume more CPU / RAM

Does Task Mining support the Remote Runtime for working with Citrix XenApps?

  • No, not yet. We use OCR screenshots, so task/process discovery usually works well in Citrix, but if you export the result as a XAML you won't get the needed selectors. It is recommended to install the client on the Citrix environment rather than on the desktop for capture.

Are there any limitations when recording in virtual / Citrix environments?

  • We did some testing, and there were no limitations identified.
  • The Desktop Client can be installed on a XenDesktop persistent VDI and works properly, recording a XenApp while having the Desktop Client installed locally works properly.

    Still, keep in mind that in Citrix apps, you can capture only the window selector.

How can I uninstall the recording application?

  • To uninstall the Task Mining recording application, please follow the generic process to uninstall an application: access your PC > Settings > Apps > search for the Task Mining app > select it > click the Uninstall button.

What can I do if the recorder is capturing too slow?

  • Ensure that there are no 3rd party antivirus programs or security applications which can slow down the recording speed. Try to allowlist Task Mining application in 3rd party applications.

  • Install the specific recording extensions for browsers you are working with.

  • Exclude specific applications that are too slow, using Edit list from Troubleshooting, to disable selectors for specific applications.

Data collection

How to choose users for a project?

  • Select 2-5 users working on similar tasks that are available during the timeframe allowing them to capture their task execution multiple times. Usually, a minimum for 2-3 users would be two weeks or one week for 3-5 users.

Is there a minimum amount of data or users required to get meaningful results?

  • We recommend 50K actions as a good baseline. This can typically be captured by 2-5 users in a week.

How to select applications for a project?

  • Select applications that users can access without administrator permissions, installed on their desktops and can be accessed via .exe.

Can URLs be allowed for recording for browser apps?

  • Yes, you can specify URLs for allowing recording in the settings.

What application languages can TM recognize?

  • English is the only language supported in Task Mining now.

Will TM record processes in mainframe apps?

  • Yes, but we do not recommend capturing processes heavily reliant on mainframe applications

Data upload and security

How is recorded data uploaded?

  • It depends on the selected option in the Output Configurations setting.

What is the data residency regions supported for the cloud product?

What are the data types, and where are they stored during the Task Mining data flow?

  • Initially, captured data resides on the user’s desktops until uploaded as an archived file to an Azure Blob storage, then extracts metadata.

Analysis and visualization

What is a task?

  • A Task is defined as a set of activities employees do to complete an entire or part of a process. Things like completing an expense report, checking if the PO amounts are correct, or spending thirty minutes in excel. It is a collection of mouse clicks, keystrokes, and screenshots.

How are tasks named and sorted?

  • The task's default rank is based on the machine learning model’s confidence of the identified task.

What is a variant?

  • A variant is a grouping of similar traces within a task. A grouping provides a view of similar traces indicating areas where they vary.

What is a trace?

  • Trace is one iteration of the task performed by one user.

What is a step? Why it has a number?

  • The number is the unique identifier for this step/cluster in the dataset.

How are the names of the steps determined?

  • Unsupervised algorithms are used to generate model representations and labels. If needed, manual editing could be used for labeling after the Analyzer algorithm is applied.

What are the numbers between steps?

  • The number between two consecutive steps (rectangles in the task graph) represents the number of traces (individual instances of the task performed) that executed those steps. For example, in the diagram below, we have 11 traces/instances where a user executed Step 4 - Open Expense Window and Step 5 - Input Expense Date. Then, only 7 traces/instances had the user proceed to Step 5a - Input Expense Category, followed by Step 6 - Enter Amount. The remaining 4 traces/instances went from Step 5 - Input Expense Date directly to Step 6 - Enter Amount.
    docs image

How can I export a task?

My analysis result shows that my task only involves one user. Why is that?

  • Our AI model looks for automation opportunities. Therefore, it tries to identify the most repetitive workflows (i.e., tasks) regardless of the number of users performing them. If a task was performed repetitively and sufficiently by a single user, it would be identified as a task in the result. The AI model does not require a task to be performed by more than one recording user. In this specific case of a task having one user, it is likely that other users completed other repetitive tasks but they are less frequent than the task performed by that single user.

In one trace, I see that a screenshot is marked as not associated to a step. What does that imply?

Not associated with any step has two meanings:

  1. The screenshot is only clustered with a few other similar screenshots, implying similar actions were performed only a few times, so it is treated as noise in the result; or

  2. This action is treated as a gap in the variant. When the AI model tries to identify the most frequent sequence of actions for the result, it allows gaps between steps to handle random actions. When the screenshot is marked as not associated to a step, it means that it is treated as a gap and was skipped.

What do the minimum, maximum and average in the Min/max utilization chart in Project Insights represent?

  • The maximum utilization represents the maximum total amount of time spent on an application/website domain across the recording users. Similar computation across the recording users is performed for minimum and average.

    For example, over the entire recording period, supposed User 1 spent 4 hours on one application, User 2 spent 5 hours on the same application, User 3 spent 7 hours, and User 4 spent 1 hour. The chart will indication a maximum of 7 hours, a minimum of 1 hour, and an average of 4.25 hours.

In the chart showing Utilization by day, if a day’s utilization is 4 hours, what does it represent?

  • Utilization by day represents the total amount of time to perform the actions across all users for a given day.

The total recorded time for my project is greater than the total time in the utilization chart. Why is that?

  • The Unassisted Task Mining algorithm treats any action that took more than 10 minutes from its previous action as idle and disregards it for the purpose of computing utilization. A previous action can be an action by an application or website domain not in the allowlist, and because it is not in the allowlist, it is not included in the CSV export from the project.
    docs image

Can I filter Project Insights at a user level?

  • No, Project Insights does not filter by users. To do that, you can analyze actions carried out by users by downloading the .CSV log file.

How do I see all actions recorded by users in my project?

  • You can download a .CSV file that contains all actions recorded by your users. See Export to CSV.

In the .CSV file, what is the reason that the Associated Task field is empty?

  • The Associated Task field is empty if an action is not included in any of the tasks identified by the AI model.

Why does an URL appear in the top 5 context switching but not in the top 5 URLs?

  • The URL showing up in the top context switching means that it is frequently switched from or to another application/website. However, it is not in the top 5 URLs because the recording users spent more time in total on those other URLs.

Troubleshooting

Why is the user's recording status switched to Error when I start the recording?

  • Please see the description for all errors here: Troubleshooting.

    You can download the user logs to see the error's details and context or share it in the support request.

Why recorder apps shows the ‘You are not assigned to a project’ message when a user signs in?

  • Because the user was invited to Automation Cloud organization but was not invited to a Task Mining project.

Was this page helpful?

Get The Help You Need
Learning RPA - Automation Courses
UiPath Community Forum
Uipath Logo White
Trust and Security
© 2005-2024 UiPath. All rights reserved.