Understanding the performance dashboard

You will learn

What each metric on the Customer Agent performance dashboard means, how to filter the data, and how to use it to find opportunities to improve Customer Agent’s behavior.

Before you begin

You’ll need:

• An Owner, Admin, Manager, or Analyst role to view the dashboard. An Analyst can view the data but cannot edit Customer Agent settings.
• Customer Agent live on at least 1 channel — there’s nothing to measure until you’ve launched

Where to find it

In Customer Agent, select Analytics. You may need to expand Performance first. You can also open it directly at klaviyo.com/kagent/performance.

Top-level metrics

The dashboard shows 6 headline metrics:

• Total resolved by AI — Count of conversations classified as resolved by Customer Agent. See Conversation status below for how outcomes are classified.
• Percent resolved by AI — Resolved conversations divided by the total of resolved, routed, and closed conversations. Open conversations are excluded.
• AI generated sales — Revenue attributed to Customer Agent. For web chat, this includes purchases of Shopify products within 24 hours of a product click or add-to-cart from Customer Agent. SMS attribution follows your Klaviyo SMS attribution settings.
• Add to carts — Add-to-cart clicks from Customer Agent in web chat. This metric is unavailable when your channel filter excludes web chat.
• Link clicks — Clicks on links Customer Agent shared in responses.
• Avg. order value — Average order value of AI-generated sales.

These metrics fall into two lenses:

• Resolution metrics (Total resolved by AI, Percent resolved by AI) — Tell you if Customer Agent is doing the job.
• Revenue and engagement metrics (AI generated sales, Add to carts, Link clicks, Avg. order value) — Tell you if Customer Agent is driving commerce alongside service.

Most brands optimize for both.

Filters

Use filters to scope the data:

• Date range — Select Last 7 days, Last 30 days, Last 90 days, or Custom above the chart. Custom ranges can span up to 365 days and include both the start and end dates. The range applies to the chart, key metrics, skill performance, and conversations table.
• Channel — Choose from the channels available in your account, such as web chat, email, SMS, or WhatsApp. Set the Channel filter separately for Key Metrics, Skill performance, and Conversations; selecting a channel in one section does not filter the others.

The chart groups daily totals using your account’s timezone, set in account settings. Conversation date filters use UTC, so their day boundaries can differ.

Skill performance

The Skill performance section shows usage volume by skill, including live custom skills you’ve created, with the busiest skills first. A single conversation can contribute multiple skill uses. It does not show a resolution rate per skill. Use it to choose which skills to investigate:

• Skills handling high volume: Review their conversations for unanswered questions, missing content, or instructions that need improvement.
• Skills rarely getting picked: For a custom skill, check whether When to use this skill clearly describes the requests it should handle.
• Skills handling the wrong types of requests: Review the conversations to confirm which requests are being handled before adjusting the skill’s instructions.

If a skill name is a link, select it and open its Performance tab to review that skill’s conversations.

Conversations table

Below the metrics, the Conversations table lists conversations that match the selected date range and table filters. Columns:

• Question — The shopper’s opening message
• Channel — Where the conversation took place, such as web chat, email, SMS, or WhatsApp
• Status — Conversation outcome (see below)
• Tags — Conversation topic tags, such as Order tracking or Returns
• Message sent — When the conversation last had activity

Hover over the avatar beside a question to view profile details, if available. Click a conversation row to open the conversation preview and read the messages.

Conversation status

Status reflects how the conversation ended (or where it is):

• Resolved by agent — The conversation ended and was classified as resolved by Customer Agent. This can include conversations where the shopper stopped replying; the status does not require the shopper to confirm that their issue was solved.
• Routed to team — Escalated to a human (via escalation rule or shopper request)
• Closed — The conversation ended without being classified as resolved by AI, for example after a shopper opted out or an agent response failed to send.
• Open — Still in progress

Acting on the data

A few common patterns and how to investigate:

• Low Percent resolved by AI — Filter the Conversations table by Status > Routed to team and open conversation rows to review the messages. Look for common questions and check whether each handoff was expected before updating content or skill instructions.
• Low AI generated sales — Shopping skills (Product Recommendations, General Q&A) may not be active or surfacing well. Confirm they’re enabled and that your product catalog is connected.
• Low link clicks — Review conversations to see whether Customer Agent includes relevant product-page or help links. Check that your content includes URLs where relevant, and review communication style rules if the responses need adjustment.
• High volume on a single skill — Review that skill’s conversations for recurring problems, then update the relevant content, How to respond instructions for custom skills, or tools.

Next steps

• Test in the Preview tab of the sidebar to check the revised responses. See How to test your Customer Agent.
• Iterate on Guidance and content based on patterns you see
• Review the event data reference to build segments from conversation data — see Customer Agent event data reference

Additional resources