Conversational analytics: ask the agent about your data

The agent can analyze your workspace's messaging and behavioral data. You can ask for information across automations, channels, and screens, gaining greater insights into your data than you could otherwise get from our UI alone—all without having to export data to a spreadsheet or another reporting platform. Ask it reporting questions in plain language, from ranking your automations to tracing conversions when audiences engage with your messages.

How it works

The metrics pages in Customer.io show you standard reports: delivery metrics, journey metrics, link clicks, and so on. But AI can reach out to a database with access to all of the messaging and behavioral data in your workspace, so you can get answers to things that don’t have ready-made charts in the UI. You can ask the Customer.io agent in the app, and the same answers are available through the Customer.io MCP server and the agent in Slack.

For example, you might ask about your top automations by “click-to-open rate”:

The Customer.io agent can even build charts on the fly, helping you visualize data-driven answers to complex questions. Using our click-to-open rate from above as an example, the Customer.io agent might show you a chart like this:

Your data reaches the agent’s database within about five minutes of an activity happening, including changes to profile attributes, objects, and relationships. So a message you sent 10 minutes ago can show up in an answer.

Questions about the current state of your audience, like how many profiles are on each plan, are the exception. The agent answers those from a snapshot that Customer.io rebuilds once a day. If you imported new plan values 10 minutes ago, the count of profiles on each plan can take up to a day to reflect them.

You don’t need to do anything special to ask the agent questions about your data. You can ask questions the way you’d ask a teammate, and your agent will attempt to find the right data to answer it. If you ask an open-ended question, like “what are my most successful automations?”, the agent recommends a way to measure success and lets you choose before it runs the numbers.

When an answer comes from the analytics database, the agent in the app can add a View and edit this query link under it. The link opens the query behind the answer as a form, next to the SQL it runs. You can change the metrics, grouping, time window, filters, and sort and limit, and the SQL updates as you edit, with changed lines marked.

• Run again runs the query exactly as the agent wrote it.
• Preview runs your edited query and shows the rows.
• Send to conversation sends your edited query to the agent as your next message, so the agent answers it in the chat.
• Reset query returns the form to the agent’s version.

The link only shows up on answers that come from the analytics database, so you might need to ask a deeper question, like a ranking or a trend over time, to see it.

Get analytics on a schedule

You can set the agent to run recurring tasks with routines. Analytics and health reporting are great candidates for routines, so you can see how your data changes over time and spot problems (or opportunities) as they arise.

For example, you might tell the agent to “rank last week’s one-time sends by click rate and email me the top and bottom three every Monday morning.” The agent will analyze the data and email you a summary of the results on a schedule.

Routines can also watch for problems instead of only reporting numbers by asking the agent to judge what matters to you. For example, you might tell the agent, “email me last week’s send volume, open rates, and click rates by channel every Monday morning. Flag anything that looks unusual compared to the previous four weeks.”

Ask the agent about your data

We’ve provided some examples below, but you can ask the agent about your data in a wide range of ways. Try asking it questions about your data, from rankings and comparisons to trends and funnels. Check out the examples below for inspiration.

The agent may ask you clarifying questions about methodology when it’s answering a question that involves judgment, like attribution windows or minimum send volumes for a fair ranking. You can ask or tell the agent to change its methodology as you go. For example, if you ask “What are my highest-performing messages?”, the agent will ask you to define “highest-performing”—whether you want to track performance by click rate, open rate, or another metric.

You can skip that back-and-forth by providing your agent the metric, the time window, and a minimum volume in your question. Instead of asking “What are my highest-performing messages?” You might tell your agent something like, “Rank my automations by click rate over the last 30 days, and leave out any that sent fewer than 500 messages.”

Message performance

Try asking the agent about how your messages perform, from rankings and comparisons to trends and funnels. This can give you a better understanding of how you’re engaging with your audience more generally than individual message and automation performance alone.

For example:

• “Which of my automations had the best click rate this quarter?”
• “Compare open rates for my welcome and onboarding automations.”
• “How many people received my product update email but never opened it?”
• “What time of day and day of week get me the best email open rates?”
• “Compare the A/B test results for my personalization automation, with open and click rate by variant.”
• “Break down my open and click rates by inbox provider over the last 30 days.”
• “Chart my message volume by channel over the last 90 days.”
• “Which of my automations has the biggest drop-off between the first and second message?”
• “How many people received more than five messages from me last week, and which automations sent them?”

Look for conversions

The agent can connect the messages people receive to the events they perform afterward, so you can trace conversions and other indicators of success:

• “Which automations drove the most purchase events last month, and how soon after the message do people typically make their purchases?”
• “How many people upgraded within a week of receiving the promo?”

Customer.io credits a conversion to the latest message a person received (within the past 30 days). Questions about conversions include the last 90 days of event history.

Website and app activity

If you send page and screen events to Customer.io, the agent can help you understand your traffic without having to export data to a spreadsheet or rely on another reporting tool:

• “How many people visited our pricing page last month?”
• “What are the most visited pages on our site this month?”
• “What time of day do people browse our site the most?”

Audience growth and changes

You can ask the agent to figure out how your audience has changed over time and which touchpoints are most effective at growing it.

• “How many people were added in the last 30 days?”
• “What is the source of most of our new people from the last 7 days?”
• “How many people changed their plan attribute this month?”
• “How many devices registered for push in August?”
• “How many companies were created in the last 30 days?”
• “Which accounts gained the most people this month?”
• “Who was added to the Acme account in the last 7 days?”
• “Which company attributes changed most often this quarter?”

If you ask the agent to analyze your custom objects—like the questions above ask about your companies or accounts—know that data only goes back to March 5, 2026. But if you ask about an earlier period, the agent tells you when the data starts instead of reporting that nothing happened.

Current attributes and objects

The sections above look at what changed over a period, and those answers include activity from about five minutes ago. You can also ask about the current state of your audience, like which attribute values your profiles hold or who belongs to an account. The agent would have to page through profiles one at a time to answer these questions, so it uses a snapshot instead.

Customer.io rebuilds the snapshot once a day, early in the morning US Pacific time. These answers can be up to a day old, so a change you make today shows up in them the next day.

For attribute values, if your profiles have attributes like plan, country, and lifetime_value, you could ask:

• “How many people are on each plan?”
• “How many people have a country attribute set?”
• “How many people are on the pro plan and in DE?”
• “What’s the range of lifetime_value across my audience?”
• “Which attributes are filled in for the most people?”

For objects and their relationships, if your object type is called Accounts, you could ask:

• “How many accounts do I have?”
• “How many accounts are on each plan?”
• “Who is in the Acme account?”
• “How many accounts is this person in?”
• “Who are the admins of the Acme account?”

The snapshot also keeps deleted and merged profiles, so profile counts can be higher than the number of profiles in your workspace. When you ask about one profile’s attributes, the agent checks your live data instead. You can also ask the agent to check live data when you need today’s value for one object or an exact count of profiles that match a single condition.

Custom events and funnels

You can ask the agent to surface patterns in activities that lead to activation, purchases, and other milestones.

• “Which events have people performed most often over the last 30 days?”
• “Break down how many unique people performed the signup_completed event by its cohort property over the last 30 days.”
• “Show me the funnel from signup to first purchase over the last 90 days.”