Response training

Overview

The Conversation View is where you’ll find a library of the conversations your end users have had with your AI Agent. You can use your AI Agent’s conversation history to see a full record of how it responded to end users, and use that information to fine tune your AI Agent.

Conversation definition

When it comes to AI Agents, a conversation is the interaction between your AI Agent and the end user.

Conversation organization

The left-hand panel of the Conversation View contains the conversations library. This is a scrollable list grouped by date, with the most recent conversations appearing at the top of the list. Each card leads with the conversation’s primary Intent and a short summary of what the end user asked.

Filters and search appear across the top of the page. You can filter the list by resolution, date, or by a number of conditions, such as language, browser, variable, or whether or not the conversation contains a handoff to an agent. The filters, search, and date range you apply stay in place as you move between conversations and reload the page, and the conversation count reflects the conversations matching your filters.

Data updates and retention

The Conversation View stays up to date on its own: new conversations appear in the list automatically, and an open transcript updates as new messages arrive—typically within 10 seconds—so you don’t need to refresh the page. Conversations stay visible in the dashboard for about the last six months today, extending over time up to two years. Older conversations then get moved to the archive.

Limitations

The Conversation View has the following constraints:

• Archival: Conversations remain available in the conversations library for about the last six months today, extending over time up to two years. Older conversations are then archived and can no longer be accessed from the Conversation View. If you need assistance or more information, contact your Ada team.
• Resolution classification: Before July 31, 2024, AI Agents only assessed a statistical sample of conversations for automated resolutions. As a result, conversations that happened before this date may not have classifications.

Use cases

The Conversation View helps you understand and improve your AI Agent’s performance.

• Review AI Agent responses: View full transcripts to see how your AI Agent responded to end users and identify areas for improvement.
• Debug and troubleshoot: Use the reasoning log to see knowledge base searches, Actions, and Processes that ran during a conversation.
• Filter by criteria: Find specific conversations by date, resolution status, language, channel, CSAT score, or other filters.
• Validate configuration changes: Check how Custom Instructions, Actions, and Playbooks affect responses in real conversations.

Capabilities & configuration

The Conversation View provides tools to review, search, and understand your AI Agent’s interactions with end users.

Conversation library cards

The conversations library appears in the left-hand panel of the Conversation View. Conversation cards are grouped by date, with the most recent conversations appearing at the top, and lead with the conversation’s primary Intent and a short summary so you can scan and find conversations quickly.

The following list breaks down the various elements that make up a conversation card:

• Primary Intent: Each card leads with the primary Intent your AI Agent assigned to the conversation. If the conversation hasn’t been classified, the card shows Unclassified.

• Conversation summary: Beneath the Intent, a short statement summarizes what the end user was asking. When your AI Agent doesn’t have enough dialog to determine intent, the card shows the end user’s first message instead.

• Resolution status: An icon in the top-right corner indicates whether the conversation was classified as Resolved — a green checkmark for Resolved, or a red icon if it wasn’t.

• Satisfaction: If the end user was served a Satisfaction Survey, an icon in the top-right corner shows their score. The icon varies depending on your survey rating method.

• Channel and end user: At the bottom of the card, an icon shows the channel the conversation took place over (chat, email, or voice), followed by the end user label and the time:

  • Name or email address – If your AI Agent collected a name or email variable for the end user, that appears as the end user label, with the name taking precedence.

  • Anonymous – If your AI Agent collected neither variable, Anonymous appears instead.

  • Test User – Conversations you or your team had with the test AI Agent are tagged [Test User].

Conversation summaries

Conversation summaries help you understand the theme of a conversation at a glance. With enough end user questions in a conversation, your AI Agent can determine the end user’s intent. Key phrases or sentences are then presented on the card in the conversation library. These phrases should give you a good idea about what the end user is trying to achieve, helping you identify conversations of interest quickly and easily.

You’ll also find Conversation Summaries at the end of the individual conversation transcripts. Of course, an end user can have several conversations with your AI Agent. Each of those may have their own summaries.

• Summaries appear on conversation cards in the Conversations library.
• Summaries also appear at the end of relevant transcripts.
• Summaries include up to three end user messages.
• It’s possible that an end user’s intent is not detectable. In that case, there won’t be any summary attached to that particular conversation.
• Summaries only consider the end user’s messages, not AI Agent messages.
• Summaries are available for conversations in all languages.

Reasoning log

After connecting your AI Agent to your knowledge base and configuring Custom Instructions and Actions, use the information from the Conversation View to see how your AI Agent uses those inputs to automatically resolve end user inquiries. This information is collectively called the “reasoning log,” because you can use it to understand how your AI Agent reasoned through resolving an end user’s inquiry.

In conversations where your AI Agent searched your knowledge base for content or ran an Action, you can see a record of its decisions to get some insight behind its reasoning.

• If your AI Agent searched your Knowledge base, you can see the search term that your AI Agent used to search for knowledge, and the article snippets that it retrieved. For more information on the content retrieval process, see Understand how your AI Agent generates content from your knowledge base.

• If your AI Agent ran an Action, you can see whether the Action ran successfully, the data that your AI Agent sent to the API, and the data the API sent back to your AI Agent.

• If your AI Agent ran a Process, you can see which Process it decided to run. If your AI Agent needed to gather additional information for the Process or for an Action within the Process, the action Gather Arguments For Function indicates that your AI Agent asked the end user for that information.

Additional conversation details

The Details pane on the right-hand side groups everything Ada knows about the conversation. You can see:

• Whether the conversation was classified as Resolved. For more information, see Understand and improve your AI Agent’s automated resolution rate.

• The conversation’s Topic and Intent classification, which you can review from this pane.

• Any Coaching that was applied to the conversation.

• All of the Custom Instructions that were active at the time of the conversation, regardless of whether your AI Agent took the Custom Instructions into account while generating responses. For more information, see Customize your AI Agent’s generative content with Custom Instructions.

• Meta variables about the end user and the technology they were using, in case you need those details for troubleshooting.

Quick start

Review your AI Agent’s conversations in a few steps.

To view a conversation:

For detailed instructions on filtering and searching, see Implementation & usage.

Implementation & usage

Use the Conversation View to review transcripts, filter conversations, and understand your AI Agent’s reasoning.

View conversations

You can open any conversation to read the full transcript and view end user details.

To view a conversation:

1. On the Ada dashboard, go to Conversation View.

2. In the conversations library in the left-hand panel, click a conversation card to open it.

   This displays the transcript of the end user’s conversations with your AI Agent in the center panel. The right-hand panel groups the conversation’s Coaching, Topic and Intent, and end user details.

3. To hide the Details pane, click the collapse arrow in its header. To open it again, click the expand arrow in the upper-right corner.

Find and filter conversations

By default, your Conversation View shows all conversations from the last 180 days, excluding test conversations. You can search or filter for specific criteria to narrow down your conversations. A banner above the list shows how many conversations match your current filters; click Reset to clear all filters, the date range, and search back to their defaults.

To find and filter conversations:

1. On the Ada dashboard, go to Conversation View.

2. If necessary, search for specific text in the conversation transcript, or use the filters to narrow down the conversations that appear in the list.

   • To filter conversations by when they occurred, click the calendar and either select a default time frame or choose your own dates within the last 180 days. Then, click Apply to see conversations that occurred in your chosen date range.

   • To apply additional filters, click the Filter button .

     1. In the Select filter type list, select a filter type:
     • Include test user: Include conversations originating from the Ada dashboard test AI Agent. Test conversations are excluded by default.
     • AR classification: The automatic resolution classification for the conversation.
     • Coaching: View conversations in which one or more Coaching instructions were applied.
     • CSAT: The customer satisfaction rating the end user gave the conversation.
     • NPS: The Net Promoter Score the end user gave the conversation.
     • Article Source: View conversations that relied on Articles from one or more specific Article sources.
     • Article: View conversations relevant only to a specific Article being utilized.
     • Action: View conversations relevant only to a specific Action being called.
     • Playbook: View conversations relevant only to a specific Playbook being used.
     • Conversation topic: The Topic and Intent your AI Agent automatically assigned to the conversation.
     • Engaged: Conversations where an end user sent at least one message to your AI Agent.
     • Handoff: Conversations where the end user was handed off to a human agent.
     • Live agent: View conversations handled by a specific human agent.
     • Language: The language the conversation was in.
     • Channel: The channel the conversation took place over (e.g., Chat, Voice).
     • Browser: The browser the end user used.
     • Device: The device the end user used.
     • Variable: Select a variable, so you can search for conversations where that value had specific values associated with it.
     • Moderation: View conversations where the moderation filter triggered. Select one or more categories: Self-harm, Violence, Hate speech, Harassment, or Sexual content.
     2. By default, searches default to using = Is as the operator to return exact matches to your query. Depending on the filter type, you can click the = list to select a different operator.
     3. In the Value box, enter a value to search for.
     4. If required, click And to add additional filter conditions. You can also hover over a condition and click the Delete button that appears to remove it.
     5. Click Apply.

   • To include or exclude test conversations, toggle the Include test user setting.

3. Select a conversation in the conversation library to read through the entire transcript.

   If you want to read more conversations from the end user in an open conversation, under their name, you can click the number of conversations to change your filter to conversations your AI Agent had with them.

Search syntax

Search matches against the text of the messages in a conversation. It is case-insensitive, so Refund, refund, and REFUND return the same results.

Related features

Explore additional conversation analysis and optimization options.

• Conversations: Learn about conversation management and analysis.
• Resolution rate optimization: Understand and improve automated resolution classifications.
• Coaching: Create instructions to refine your AI Agent’s responses.
• Knowledge: Manage the content your AI Agent uses to respond.
• Actions: Create automated actions that appear in conversation logs.
• CSAT: Collect and analyze customer satisfaction data.