FAQs for Copilot (BETA)

How can a user access Copilot in Gainsight?

Copilot is available in the left navigation bar of the Gainsight application. Users can click on the Copilot icon to open the chat window and start interacting with the AI.

What type of users can access Copilot? 

Copilot is available to all user types with Full Licenses, Viewer Analytics, and Viewer Licenses. It is included in the default permission bundle for all license types. 

Is Copilot available for all users, or can it be assigned to specific users for an account? 

Copilot access can be assigned to specific users based on permission bundles. Refer to Configure Permission Bundles to Enable Copilot for Users in the Copilot Overview article for details on setting up permissions.

What customer data is used to power Copilot?

Copilot uses multiple data sources to provide contextually relevant answers based on historical customer interactions and product knowledge. It retrieves information from:

• Company Data: Structured details from the Company object.
• Timeline Data: Historical insights from up to one year of Timeline activities.
• Product Documentation: Feature-specific information from Gainsight product documentation.
• CS Best Practices: Expert guidance from internal knowledge and ChatGPT’s trained expertise.

Note: Copilot does not scan documents attached in Timeline activities or any customer-shared documentation. It only references Gainsight's official product documentation.

How long does Copilot take to generate an answer?

On average, Copilot generates an answer in 30 seconds to 1 minute, depending on the complexity of the question and the amount of data it analyzes. If Copilot does not respond after 3 minutes, it shows a timeout error, which allows users to retry their query. When you encounter a timeout error, retry your question for a new attempt at generating an answer.

What type of timeline activity will be processed by Copilot?

Copilot processes all activities present in the Company Timeline across all companies. It includes entries from Scorecard, Cockpit (CTAs), Success Plans, Opportunities (Renewal Center), C/R360, and external plugins, as long as they are configured to appear in the Company Timeline. 
Note: Journey Orchestrator data is not included.

What is the duration of Timeline data that Copilot can access to answer questions?

By default, Copilot can retrieve and analyze up to one year of Timeline data to answer questions. If you need information from a shorter timeframe, you can specify the desired period in your query.

How does Copilot determine if it should query Global or Company Timeline?

A large language model (LLM)-based classifier determines if a query is a Company query or a Global query. Depending on this classification, the appropriate parameters are selected, and the relevant Timeline activities are retrieved.

Does the Copilot provide sources from where the answers are generated?

When Copilot answers using Timeline data, it includes links to the original Timeline posts, allowing users to trace back the context. However, responses generated using product documentation do not include links.

Is there any limit on the number of Timeline Activities shown per response?

Up to 5 Timeline entries are shown upfront. Users can view all activities from which the response is generated by using the view more option.

What actions can users take on the link to the Timeline activities given for an answer?

Users can interact with the Timeline activities within the Copilot interface just as they do in Timeline, meaning they can comment, copy, share via email, edit, or delete, depending on their user permissions.

Can the Copilot generate reports or access information beyond the Timeline?

Copilot can access Company Data and Timeline Data but cannot generate reports or retrieve data from other sources. It is limited to:

• Company Data: Structured information from the Company object.
• Timeline Data: Up to one year of historical activities for context-based insights.

Can a user provide feedback on answers given by Copilot?

Yes, users can provide feedback by giving a “Thumbs Up” or “Thumbs Down” to the answers. This allows a user to give Thumbs Down to an answer if it’s inaccurate, irrelevant, or incomplete.

How is feedback data utilized? Does it affect the query response? 

We collect user feedback to review and improve the model over time. This feedback does not affect the current results directly and does not immediately impact future responses.

What are the best practices while asking a question to get better, more accurate answers?

• For optimal performance, queries should ideally involve only one service. While multiple services can be utilized, breaking down the query into sequential, single-service queries is recommended.
• Provide full names when inquiring about a person or a company to ensure precise responses.
• Ensure your queries are clear and specific to avoid vagueness and improve the accuracy of responses.
• Understand the subtle differences between each service. For example, a query like "What are CTAs?" might be directed to Brainstorming Assist instead of Product Documentation Assist.
• Queries such as "summarize my last three meetings" will apply a user filter. 
• User specific queries are not fully supported in Timeline Assist. Example: How many meetings has xxx attended?

Are there restrictions on the type of questions users can ask within Copilot? What would happen if such questions were asked?

Yes, there are some restrictions on the types of questions users can ask within Copilot. Copilot can answer queries based on Company Data, Customer and Global Timeline, Gainsight Product Documentation, and CS Best Practices. 
Note: If a user asks a question outside these areas, Copilot might provide a general response or ask for more details to better understand the query.

Are there any restrictions on the length of conversations?

No, Copilot does not have any restrictions on the length of conversations.

Will there be chat history support? 

Yes, Copilot retains chat history session-wise, allowing users to continue from previous sessions or start new chats. The chat history is retained for the 30 days and users can delete or rename chat sessions.

Can a user share a chat history of their conversation with Copilot? 

No, currently users cannot share chat history of their conversation with Copilot with other users.

Can AI Copilot maintain concurrent sessions? How is session state managed?

Yes, Copilot can maintain concurrent sessions, with queries running in parallel. Users will see running states for all queries in different chats.
Note: In one chat, a user can run only one query at a time.

Do we support Copilot in Sandboxes?

No, Copilot is not supported in Sandbox environments. 

What are the limitations of Copilot?

• Copilot's support documentation is up-to-date until July 2024.
• By default, Copilot applies a 180-day filter if no timeframe is specified in a query. We are working on making this dynamic in the next release. Specify specific timeframes in your query, e.g., "last 60 days," "last 90 days," or "last quarter."
• If Copilot is not providing relevant responses, starting a new session may yield better results as we are currently improving the Copilot orchestrator.
• Currently, Copilot does not support queries related to relationships.
• By default, Copilot uses active company data. If you need information on inactive or churned accounts, please specify this in your query. Note that Copilot can only work with one status at a time (active, inactive, or churned)

User specific queries are not fully supported in Timeline Assist. Example: How many meetings has xxx attended?

What technology are we using to provide answers to users' questions? 

We primarily use OpenAI's models to generate responses. If latency issues arise, we redirect traffic to an enterprise host service offered by Microsoft, which incorporates OpenAI models within the Azure environment.

How do we ensure data privacy?

Data inputs for Copilot include publicly available product documentation and Timeline data. Relevant PII data is masked before being sent to the vendor (name and position may remain unmasked for relevance). Data from one tenant is not shared or used for training another tenant. We follow a Zero Retention Policy, ensuring no data resides on the OpenAI or Azure platforms, and all data in transit and at rest is encrypted. Rigorous security testing is performed, and all communications with our AI feature are API-based, with security checks completed as part of API integration. For more information, refer to the Data Privacy FAQs for Generative AI Supported Features article.