Overview of Agent Analytics

Agent Analytics lets you understand how visitors interact with AI‑powered conversational interfaces in your product, such as chatbots and virtual assistants. It surfaces real prompts and conversations so you can see how often people use your AI agents, what they ask for, and where things break down. You can also use Leo, Pendo’s AI assistant, to quickly explore patterns and themes in this data.

In this context, an AI agent is the conversational UI in your product that receives natural‑language questions or instructions from visitors and returns generated responses. The messages visitors submit to the agent are referred to as prompts, and the full back‑and‑forth between a visitor and the agent is treated as a conversation.

When to use Agent Analytics

Agent Analytics is useful when you want to:

• Understand how customers are actually using AI agents in your product.
• Identify common intents, questions, or workflows handled by the agent.
• Detect confusion, frustration, or failures in conversations.
• Measure the impact of changes to agent prompts, behavior, or coverage over time.

How we capture agent data

Agent Analytics supports two data capture methods. Both capture data at the time of interaction, but they differ in the depth of insight they provide.

For web and mobile apps, Pendo automatically sets up full conversation capture. For browser extension apps, you choose the capture method during setup.

While prompts-only capture helps you understand what visitors ask, full conversation capture provides visibility into both visitor prompts and agent responses. Capturing both sides of the interaction enables analysis of issues, frustration signals, conversation length, and overall agent performance over time.

For step-by-step instructions on configuring an AI agent, see Add and configure AI agents in Pendo.

For implementation details on full conversation capture, see Capture full conversations with the Conversations API.

What you can analyze

Agent Analytics processes captured data to provide several types of insights:

• Overview metrics track usage volume, unique visitors and accounts, retention, user reactions, issue rates, rage prompt rates, unsupported rates, and error rates over time. For more information, see Analyze interactions with AI agents.
• Use cases are automatically surfaced based on semantic similarity in prompts and conversations. You can track emergent patterns or define specific topics to monitor over time. For more information, see Analyze and track use cases for your AI agents.
• Issues are automatically detected from conversation data and grouped by theme. You can also manually track specific problems to monitor them over time. For more information, see Identify and track issues with your AI agents.
• Conversations show threaded chats of both visitor prompts and agent responses so you can review full transcripts, validate changes, and spot examples to improve training data and documentation. For more information, see View full conversations.
• Experiments let you compare performance between different agent configurations, models, or tools to measure the impact of changes. For more information, see Compare AI agent versions with experiments.

You can also explore Agent Analytics data beyond the Agent Analytics module:

• Use AI agent events in paths, funnels, dashboards, and stickiness widgets to see how AI fits into the rest of your product.
• Start from the AI engagement dashboard template for a pre-configured view of key metrics and trends.
• Ask Leo, Pendo's AI assistant, questions about use cases, issues, and trends in natural language.
• Send scheduled Agent Analytics digests to Slack so key metrics and emerging issues land in your team's channels.

Who can access conversation data

Access to Agent Analytics features depends on your role in Pendo.

Subscription admins have full access to all Agent Analytics features across all apps. The AI Agent Admin role is an app-level role that grants similar access, but only for the specific apps it's assigned to. A user who is AI Agent Admin for one app but not another can only manage agents and view conversation data for the app they're assigned to.

Other users without admin or AI Agent Admin permissions can access Agent Analytics in read-only mode. They can view the agent overview, use cases, issues, and existing reports, but can't view actual visitor prompts or conversations.

For instructions on assigning the AI Agent Admin role, see Assign the AI Agent Admin role. For a full permissions reference, see Roles and permissions.

Where Agent Analytics works

Agent Analytics can capture data from any AI agent where you have access to either the browser context (through the Pendo Web SDK or browser extension) or the agent's backend code (through the server-side Conversations API).

This includes agents that run in:

• Your own web applications (using the Pendo Web SDK).
• Third-party web applications where the Pendo browser extension is deployed for employees (prompts-only).
• Mobile apps, backend services, or custom integrations where your development team can call the server-side Conversations API (for example, a Slack or Microsoft Teams agent that routes conversations through your own backend).

Agent Analytics can't capture data when you don't have access to either the browser context or the agent's code. For example:

• Third-party agents accessed through desktop applications where the browser extension can't be deployed (for example, Copilot).
• Third-party agents embedded in your application through an iframe or similar container where you can't add the Pendo Web SDK and don't have access to the agent's backend code.

Note: Agent Analytics respects your privacy and exclusion settings. Visitors and accounts marked as Do Not Process don't generate Agent Analytics data, and prompts from visitors and accounts on your exclude list aren’t included in your prompt counts and don’t count toward your monthly prompt limit.

Usage limits and prompt quotas

For customers who don’t have a paid Agent Analytics subscription, Pendo applies a monthly prompt limit.

• Each subscription can collect up to 500 prompts for each calendar month.
• When you reach this limit, Pendo stops collecting new prompt events for the rest of that month. Existing data and reports remain available.
• Limits reset on the first of every month at 00:00 UTC. When the quota resets, Pendo starts collecting new prompts again automatically.

In Agent Analytics, we surface in‑app banners when you’re approaching or have reached your prompt limit so you can decide whether to upgrade before data collection pauses.

To see how prompts are counted and monitor usage across your subscription, see Track AI prompts in organization settings.