Adaptive Learning in Auto Answers and Bots

MicroStrategy ONE (June 2024) adds support to learn from user interactions in Auto Answers and Bots. This new component enhances response accuracy and precision by listening to user feedback and adapting responses accordingly. Auto Answers and Bots will then be able to continuously improve and provide more contextually relevant and precise responses that are tailored to individual user preferences.

Adaptive learning is used per-user which allows the system to apply what it has learned across the application during subsequent conversations with a user and ensure personalized and accurate responses across relevant dashboards and Bots.

Starting in MicroStrategy ONE (June 2024), MicroStrategy AI only learns from user feedback when the original question is classified as ambiguous. MicroStrategy AI classifies questions as precise or ambiguous. Ambiguous questions trigger smart alternative suggestions in the Did you mean? window. For more information, see Smart Suggestions. For other instances, the feedback is stored for administrator review and potential refinement to improve answer quality.

For information on enabling adaptive learning, see Configure Adaptive Learning.

Limitations

• User learning string length: Feedback cannot be greater than 300 characters, including white spaces.

• Number of adaptive learning entries per user: Each user can have 500 learning entries generated through both smart suggestions and thumbs-down feedback.

Learning Process

Choose Smart Suggestions

When MicroStrategy AI detects an ambiguous query, it provides a default answer and suggests alternative options to help users clarify their queries and get more accurate results.

For more information on smart suggestions, see Smart Suggestions.

When a user clicks on a suggestion, MicroStrategy AI generates the response based on the understanding. MicroStrategy AI then learns from these clarifications to improve future response accuracy.

Not every interaction results in MicroStrategy AI learning something new. If there is nothing new to learn from an interaction with MicroStrategy AI, the user receives a response without the learning prompt. MicroStrategy AI accurately identifies when valuable insights can be gathered for future use.

Feedback

After a response in MicroStrategy AI, a user can click a thumbs-down icon and add a comment or choose a predefined option such as Incomplete answer or Incorrect format. MicroStrategy AI then evaluates this feedback to learn and improve. For more information on providing user feedback in MicroStrategy AI, see Using Auto Answers or Using a Bot.

MicroStrategy AI only learns from user feedback when the question is classified as ambiguous.

The user provided feedback is available to Bot and dashboard owners via the Usage tab in edit mode. Administrators are also available to access the feedback in the Auto Adoption dashboard in the Platform Analytics project.

You can provide feedback in Auto Dashboard but adaptive learning is not yet supported. However, all user feedback will be displayed in the Last 30 Days page of the Auto Adoption dashboard in the Platform Analytics project.

Adaptive Learning Workflow

See the following workflow to ask a question, detect ambiguity, generate learning entries, and generated learning entries:

Apply User Learning Entries in Conversations

User learning entries are applied to questions that don't yield the expected results based on a user's understanding. See the following example of an employee dataset with various objects:

When a user asks an ambiguous question, MicroStrategy AI may interpret it based on its closest understanding of the question or using the knowledge assets defined in the Bot, dashboard, or application.

Learning Creation

Users can create new learning entries using smart suggestions or feedback. The following example demonstrates a learning entry using feedback:

The user asks a question where MicroStrategy AI provides an answer using information from the linked knowledge asset file. Therefore, MicroStrategy AI chooses the Average Capacity metric.

If the user finds that the data is incorrect, they can click the thumbs-down button and select the Incorrect Data tag.

Choosing a tag is optional. These tags help MicroStrategy AI evaluate feedback and making improvements.

After a user submits their feedback, the Bot processes the user input and generates a new learning entry. The user has the option to forget the newly generated learning if it is deemed unhelpful.

Learning Consumption

When a user asks the same question, MicroStrategy AI will use the learning entry to provide an answer. You can check which learning entry was used in the footnotes of the interpretation window.

Priority Handling

When there is a conflict between knowledge assets and user learning entries, MicroStrategy AI prioritizes the user learning entry. This prioritization ensures a personalized experience.

Example: If a knowledge asset states "High revenue is classified as revenue over USD 1.2 million per quarter", but a user learning entry states, "High revenue is classified as revenue over USD 1.5 million per quarter", the user learning entry is prioritized. This personalization is only applied to the user who generated it. Other users will see results based on the knowledge asset.

If there is a conflict between two user learning entries, the most recent learning entry is prioritized.

Manage User Learning Entries

Users can manage their own learning entries using forget or by overwriting their entries by creating new ones.

Forget: The forget button next to generated learning entries removes the recent entry for the user and does not apply the entry to future questions. However, learning entries can be forgotten only during the active session. Previous learning entries and Forget will not be displayed if the user creates a new session.

If the user clears the chat history during the existing session, the previous learning entries and Forget will clear, as well.