ADD-ON FEATURE Insights Agent is available at no cost during this preview period, with paid pricing expected later this fall. Reporting access in Gladly is required to use it. To get started, contact Gladly Support or your SAM if you’d like to activate this feature. |
Insights Agent performs best when your question is specific. The more context you include, the more precise the result. Think of it like filling out a report filter: each piece of information you add narrows the query and improves the answer.
Four ingredients for a good question
The following example includes all four ingredients that produce the best results:
.png?sv=2026-02-06&spr=https&st=2026-06-18T19%3A00%3A55Z&se=2026-06-18T19%3A14%3A55Z&sr=c&sp=r&sig=nDzPG4MN%2B8Zr1vi%2F81MBFat2hsy%2Flt%2FRSD7bPR4p9%2BA%3D)
Always specify a date range. Leaving out dates is the most common cause of poor or empty results. Use natural language like "last 30 days" or "April 2026."
Specify the time granularity. “By day" for a 30-day window, "by week" for a 12-week window, "by month" for a full year.
If you're not sure of the exact metric, just describe what you want to know
The four ingredients above give Insights Agent the most precise information to work with, but you don't always need all of them. Insights Agent is designed to interpret intent, so if you're not sure of the exact metric name or how to frame the question technically, describe the outcome you're looking for and let it figure out the rest.
For example, "Which of my Agents are the best at handling Conversations quickly?" is enough. Insights Agent will interpret that as an average handle time ranking, identify the relevant data sources, and return a ranked result, without you needing to know the field name or which explore to use. If the result isn't quite right, you can refine from there.
What works and what to avoid
Do this | Avoid this |
|---|---|
Always include a time range ("last 30 days," "Q1 2026") | Leaving out dates entirely ("show me missed calls" – which period?) |
Be specific about the metric you want | Combining multiple unrelated questions in one message |
Tell it how to slice the data (by Channel, by Agent, by Topic) | Using internal system names the agent won't recognize (e.g. try "Voice Channel" rather than a custom IVR name) |
For Agent metrics, specify aggregation granularity (by day / by week / by month) | Asking for metrics that don't exist in Gladly's reporting |
Ask one clear question at a time | Asking about individual Customer records or PII |
Break complex questions into smaller follow-ups | |
Start with "Tell me about this data" to discover what's available | |
Stick to metrics that exist in your standard Gladly reports |
Complex question? Break it up
If you're hitting "Unable to complete request" errors, ask for one metric first, then follow up. Insights Agent maintains context within a session.
Warm up
If you're not sure what to ask, start with a warm-up prompt. Insights Agent will describe what data it has access to and suggest starting questions based on your org's reports.
Try one of these to get oriented:
"Tell me about this data"
"What are some questions I could ask?"
"What metrics do you have for Agent performance?"
The start screen also includes a "View supported question examples" link that opens a reference of general question types Insights Agent can handle, such as metric trends over time, breakdowns by dimension, and top-value rankings.
Once you know what's available, follow up with a specific question. Insights Agent remembers context within a session, so you can build on previous answers without repeating the full question. For example, after asking about handle time by Channel, follow up with "Now compare that to last month."
Example questions
Use these as a starting point. Swap in your own time windows and specifics.
Conversation volume and Channels
"Show me Conversation volume over the last 30 days, broken down by Channel."
"Compare average handle time this month vs. last month."
"Show me SLA performance by channel for the last 4 weeks."
Topics and reasons for Contact
"What were the top 5 topics that caused Customers to contact us in Q1 2026?"
"Show me topic breakdown for the last 14 days, sorted by volume descending."
Agent performance
"What are Agent occupancy rates for the last 30 days, aggregated by week?"
"Which Agents had the most missed calls last month? Show me names and counts."
"Show Guide resolution rates and number of Conversations per Guide for the last 7 days."
Gladly AI sessions
"Show me AI session resolution rate vs. recontact rate for the last 30 days, by day."
"Can you look at my transfer rationales for the month of April 2026 and break them down into categories? Filter out sessions where transfer rationale is null."
"Compare AI session volume this month to last month, broken down by day."
Read a response
A successful response has several layers, from a plain-language summary down to the underlying query details.
Click on the icons below to learn more about these components:
When something goes wrong
When Insights Agent cannot answer a question, it tells you why. Expanding the error detail gives you more context to diagnose and fix the issue.
Error | What caused it | How to fix it |
|---|---|---|
No date range | No time window was given, so the Agent defaulted to a single day or returned nothing. | Add a specific time window: "last 30 days," "April 2026," "last 12 weeks." |
Too many steps | The question required too many operations at once ("unable to complete within the maximum allowed reasoning steps"). | Break into smaller questions. Ask for one metric first, then follow up with the next. |
Metric not found | The Agent searched for a field that doesn't exist in Gladly's data. | Verify the metric exists in your standard Gladly reports. Use common names rather than system-specific labels. |
Zero results | The query ran but returned no data – often a one-day default window or a granularity mismatch in Agent metrics. | Add a longer date range and specify granularity (e.g. "by week for the last 12 weeks"). |
Chat timed out | The session expired, typically on complex queries. | Restart from the left menu. Simplify the question or split it into steps. |