> This page is for version Generative (default).
> For other versions, use one of these documentation indexes:
> - Generative (default): https://developers.ada.cx/generative/llms.txt
> - Scripted: https://developers.ada.cx/scripted/llms.txt

> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://developers.ada.cx/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://developers.ada.cx/_mcp/server.

# Resolution rate optimization

Measuring your AI Agent's success is a challenging task. It's impossible to
go into every conversation that customers have with your AI Agent, and with
all of the different reports and metrics you can look at, it can be hard
to know which ones to focus on.

Ada has created the **automation resolution** metric, which is not only
a measure of whether your AI Agent was able to automatically resolve a
customer's inquiry without handing them off to human support, but also
whether it was able to successfully address the reason why your customer
came to your AI Agent in the first place. Your AI Agent uses AI language
understanding to assess both customers' inquiries and its own responses
to figure out whether a successful automatic resolution did or didn't
happen.

## Why automated resolution?\[#why-ar]

Historically, metrics like the [containment rate](/generative/docs/monitoring/performance/reports/performance-reports#containment-rate)
have come close to measuring success, but they couldn't provide
a complete picture. The containment rate tells you which proportion of
conversations that customers had with your AI Agent ended without being handed
off to a human agent. But without context, this metric can only tell you
so much. Can you differentiate between a customer who ended a
conversation with your AI Agent because they were satisfied, as opposed to
one who got frustrated and gave up?

While containment can be a useful metric to get a quick glance of the
proportion of bot conversations that didn't escalate to a human agent,
automated resolution takes it a step farther. By measuring the success
of those conversations and the content they contain, you can get a much
better idea of how helpful your AI Agent content really is.

## Understand how your AI Agent classifies automatic resolutions\[#ar-classification]

Your AI Agent analyzes each conversation to understand both the
customer's intent and the AI Agent's response. Based on that analysis, it then
assigns a classification of either Resolved or Not Resolved to each
conversation.

For a conversation to be considered automatically resolved, the
conversation must be:

* **Relevant** - Ada effectively understood the customer's inquiry, and
  provided directly related information or assistance.

* **Accurate** - Ada provided correct, up-to-date information.

* **Safe** - Ada interacts with the customer in a respectful manner and
  avoided engaging in topics that caused danger or harm.

* **Contained** - Ada addressed the customer's inquiry without having to
  hand them off to a human agent.

To get an overview of what percentage of conversations
were automatically resolved, you can use the [Automated Resolution Rate](/generative/docs/monitoring/performance/reports/performance-reports#automated-resolution-rate)
report.

> **Note**
>
> Before February 24, 2025, only knowledge content was used to assess conversation accuracy and predict whether a user's issues were resolved. Processes, Actions, Coaching, and Company Descriptions were not yet considered.

> **Note**
>
> 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.

## Conversation Lifecycle

For contained conversations, the AI Agent will assess a conversation for automated resolution after the conversation ends.

Conversations may end immediately if:

* A customer clicks End Chat on web chat
* A customer hangs up during a voice call, or
* A human agent ends an escalated conversation.

Idle conversations will automatically be ended after 24 hours in web chat and social channels, and 72 hours in email.

For escalated conversations, the AI Agent marks the conversation as **Not Resolved** immediately following the handoff.

## Provide feedback on how your AI Agent classified a conversation\[#provide-feedback-ar]

You can read through individual conversations to see how
customers interacted with your AI Agent. If you disagree with how your AI Agent
automatically classified a conversation, you can provide feedback to
Ada.

> **Note**
>
> This feedback is only so Ada can improve our classification model. It
> does not override the automatic classification assigned to the
> conversation.

1. Open a conversation. You can do this two ways: from the Automated
   Resolution Rate report, or from the Conversations view.

   * **From the Automated Resolution Rate report**

     If you're looking at the [Automated Resolution Rate](/generative/docs/monitoring/performance/reports/performance-reports)
     report, scroll down to the Conversations section. Click a
     conversation to read the entire transcript.

   * **From the Conversations view**

     On the Ada dashboard, go to **Conversations**. Optionally, you can use the **AR Classification** filter to narrow down
     conversations by selecting **Resolved** or **Not Resolved**
     to look through those conversations.

     Select a conversation in the conversation library on the left side
     of the screen to read through the entire transcript.

2. If the sidebar on the right side of your page is collapsed, click
   **Details** to expand it. There, you can read through your AI Agent's
   automatic assessment of the conversation:

   * **Inquiry Summary** - An automatically generated summary of what
     the customer wanted to accomplish.

   * **Classification** - The classification of either Resolved or Not
     Resolved that your AI Agent assigned to the conversation.

   * **Reason for Classification** - Your AI Agent's understanding of what
     happened in the conversation, which led it to the classification
     that it assigned.

   Take a second to understand the reasoning that your AI Agent used to make
   its classification, so you can give more targeted feedback.

3. Beside **Agree?**, click **Yes** or **No**.

   If you clicked **No**, enter some additional details about what your
   AI Agent got wrong in its assessment.

4. Click **Submit**.

## Find trends so you can improve your AI Agent's content\[#find-trends-improve-content]

Automated resolution data can be very powerful in helping you find gaps
in your AI Agent's content that you can improve. While going through your
automatic resolution data, ask yourself:

* **What kinds of patterns am I seeing?**

  For example, let's say there are a lot of unresolved conversations
  caused by your AI Agent being unable to understand what your customer was
  asking for. Are there [improvements](/generative/docs/knowledge/agent-logic/knowledge-setup) you can make to your knowledge base content to improve recognition?

* **Is there additional information I can add?**

  If you see that multiple unresolved conversations relate to the same
  topic, consider adding that information to your knowledge base to address it.

Keep in mind, it's not a realistic goal to
get to 100% automated resolutions in your AI Agent. Some customers ask for
handoffs immediately, and others might ask inquiries that are off-topic
or even abusive, and you can't expect your AI Agent to resolve those queries.

---

Have any questions? Contact your Ada team, or email us at [](<mailto:help@ada.support%20%E2%80%9CHelp Docs inquiry%E2%80%9D>).