Suggest Response uses Generative AI to provide support agents with instant, contextually relevant responses, transforming how they handle queries. Agents are freed from the repetitive task of searching through knowledge bases or internal documents as the AI crafts responses using past interactions and the specific context of the current query. This capability reduces response times and improves agent productivity, regardless of the agent's training or workload, offering a faster and more seamless support experience.

Key Benefits

• Instant Context-Driven Suggestions: AI uses the specific context of each customer query to suggest the most relevant and accurate answers instantly.
• Consistent Quality: Agents, regardless of experience or workload, can deliver uniform, high-quality responses.
• Faster Resolutions: AI eliminates the need for agents to manually search multiple sources, enabling them to respond faster and resolve issues more efficiently.
• Increased Productivity: By generating instant responses, AI enables agents to handle more inquiries efficiently, freeing them up to concentrate on complex or urgent tasks.
• Data Privacy Assurance: A zero-day retention policy with OpenAI ensures all customer data remains private and secure.

Note: Please contact your customer success manager or the Helpshift support team to enable this feature.

Generating Suggest Response with AI

Once enabled, the Suggest Response feature allows agents to leverage suggested replies directly into their conversation pane for faster responses.

To generate a suggested response with AI, follow the steps given below:

1. Log in to your Helpshift account.
2. Go to the Issues section and select the conversation you want to respond to.
3. Click the Agent Copilot icon ( ) below the text input area in the Issue conversation pane. 
4. Click Suggest Response. Based on the context of the conversation, the AI will generate the most relevant response in the Suggest Response pop-up.
   
5. Click INSERT RESPONSE to insert the response to the text input area. You can send it as-is or make any changes before sending it.
      

The Suggest Response pop-up also includes additional options. These are as follows:

• Similar Issues found: AI will highlight three similar past issues and suggest responses based on them. Agents can click on the past issues and open it in a new tab for additional context if needed.
• Timestamp: Each suggested response includes a timestamp indicating the last generated time.
• Feedback option: Rate the quality of the suggested response.
  If the thumbs-down option is selected, a text box will appear for specific feedback to help improve future suggestions.

You can also view the Suggest Response on the Issue details page by clicking Open Page View at the top right corner of the conversation pane.

Response Generation Process

The Suggest Response feature fetches information from similar issues in the past and other relevant sources (future scope) in the context of the ongoing conversation. When each message is received, the AI evaluates the full context of all received messages and offers a suggested response, which agents can review, edit, and use as needed.

Data Privacy and Security

The system masks any Personally Identifiable Information (PII) to safeguard customer privacy before sending data to the AI model. Upon receiving the Suggest Response, the PII is unmasked before it’s displayed to the agent. Our OpenAI Enterprise license ensures zero data retention to protect sensitive information.

Common Failure Cases

The Suggest Response process can fail in the following cases:

• The AI may not generate a suggestion when there are no past issues or minimal conversation data. In these cases, agents may need to craft a manual response.
  
• If the AI cannot generate a suggestion due to a technical issue, retry the action. If the problem persists, please contact support for assistance.
  

Managing Suggest Response

Once enabled, admins (who have access to this feature) can manage the feature by accessing the SensAI tab.

Admins without access will see the MANAGE FEATURE button but cannot click it.

To configure and manage the sources for Suggest Response:

1. Click MANAGE FEATURE, and the Suggest Response Settings page will open.
    
2. Under Other sources, you can submit a suggestion to add relevant sources for future use in addition to Past Issues.
3. Click SUBMIT.
   

Best Practices for Using the Suggest Response Feature

The Suggest Response feature helps operators in suggesting efficient responses to customer queries. To use it effectively while ensuring high standards of quality and compliance, operators should follow these best practices:

1. Avoid personal data disclosure

Helpshift implements internal masking mechanisms for personal data before sending it to AI systems. In case, any irrelevant personal data is identified in the suggested responses, operators are advised to report using the feedback option within the response.

2. Verify AI responses for accuracy

AI-generated responses rely on the quality of the operator’s past responses. To ensure accuracy:

• Always review the AI generated response before sending it to the customer.
• Flag any irrelevant or inaccurate responses for AI system improvement.

3. Review and mitigate biased responses

AI generated responses may occasionally require adjustments to fully align with brand guidelines. To ensure this:

• Review all responses before sharing them with customers to report any inappropriate, biased, discriminatory, or offensive response.
• Immediately report any biased or unethical responses using the feedback option in the feature window - Agent dashboard UI.
• Stay updated on the limitations of the AI system, which will be outlined during training.

4. Manage high impact queries responsibly

For complex or high-impact topics like legal or financial issues:

• AI generated responses can serve as a helpful starting point, but operators should always draft the final response manually based on approved internal guidelines and policies.
• High impact queries require thorough verification, so double-check any suggested response.
• Ensure that you have undergone the relevant training and have access to the right resources as operators to identify such scenarios and respond to them appropriately.

Examples of high impact query management

• Legal proceedings threat: If an end user emails a threat to initiate legal proceedings, operators must understand the implications and ensure the response is carefully crafted without relying on AI generated responses.
• Requests from law enforcement: For support tickets received from law enforcement authorities requesting specific data, operators should follow approved processes and avoid using AI generated responses.
• Vulnerability reports or application issues: When a ticket highlights vulnerabilities or reports application issues, operators must escalate appropriately and avoid relying on AI generated responses.

5. Follow customer specific guidelines

Some situations, like legal claims or data rights requests, may require strict adherence to customer templates. Use the pre-existing templates provided for handling such scenarios and avoid AI suggestions where usage of response templates is advised.

6. Report errors related to AI features

Operators should report inaccurate AI generated responses or errors to your team lead/manager and seek further guidance. You could also use the in-app feedback option to report any AI feature related inaccuracies or errors you encounter.

7. Handle Data Subject Access Requests (DSARs) appropriately

Use predefined templates for DSARs to ensure compliance with customer-specific protocols without relying on AI suggestions.

8. Exercise caution with children’s data

If the operators encounter an end user (children), they should not use the AI generated answers to respond to their queries.

9. Limitations of AI

AI generated responses may not be accurate as the responses are generated from a limited pool of knowledge, and the AI features may provide incorrect or inaccurate information. Therefore, it's advisable to review every response generated by the AI features.