Introduction
Escalations by Topic shows you which conversation subjects your AI hands off to a human most often. Instead of a single handoff number, you get a ranked list of the subjects behind it — so you can see whether shoppers reach your team about order delays, returns, sizing, or something specific to your store.
It's the card to read when you want to know where your AI needs reinforcement. A topic near the top is one your AI is least able to finish on its own, and usually the fastest place to add a policy, an FAQ answer, or a support skill.
To find it:
- Log in to your Rep Console.
- Click AI Support in the left navigation menu.
- Select Support Analytics.
- Escalations by Topic is the first card on the page, directly under the filters.
What the card shows
The card lists up to ten topics, with four columns.
Topic
What it shows: The subject of the conversations, such as Order status, Refund policy, or Damaged order.
Why it matters: The topic name tells you what shoppers were actually asking about when your team got involved. It's the difference between knowing that handoffs happen and knowing what they're about.
How to use it: Click any topic name to open your Conversations list, already filtered to the escalated conversations on that topic for the same time range and channel. That takes you from a number to the real exchanges behind it in one click.
Escalations
What it shows: How many conversations on that topic were handed to a human.
Why it matters: This is the volume side of the picture — the raw workload a topic creates for your team.
Conversations
What it shows: How many conversations touched that topic in total, whether or not they were escalated.
Why it matters: It's the denominator that makes the rate meaningful. Twelve escalations out of 15 conversations is a very different situation from 12 out of 900.
On narrow screens this column is hidden and the rate column is shortened to Rate.
Escalation rate
What it shows: The share of that topic's conversations that needed a human, shown to one decimal place. A topic with 8 escalations across 10 conversations reads 80.0%.
Why it matters: This is the number that tells you how reliably your AI handles a subject. A high rate means that whenever the subject comes up, your AI usually can't finish the job.
How the list is ordered
Rows are ordered by escalation rate, highest first — not by how many escalations there were. The topic at the top is the one your AI most often can't complete, even if another topic further down produced more handoffs in total.
This is deliberate, and it's worth getting used to, because the two views answer different questions. Rate tells you where your AI is weakest. Volume tells you where your team spends its time. To switch, click the Escalations column header to sort by volume instead — clicking any of the three number columns re-sorts the list instantly, and clicking the same one again reverses the direction. The Topic column isn't sortable.
Your sort choice resets to escalation rate whenever you change the date range or the channel.
Choosing your time range and channel
The card follows the filters at the top of the Support Analytics page.
Date range. The page opens on the last 30 days, ending yesterday. You can choose Today, Yesterday, This Week, Last Week, This Month, Last Month, or pick a custom range on the calendar and click Apply. The current day isn't selectable, because topics are assigned the day after a conversation happens.
Channel. Switching between channels re-runs the card, so you can look at web chat separately from your helpdesk.
What you see with Deep Research
Without the Deep Research addon, topics come from Rep AI's standard subject list — a fixed set of common ecommerce subjects that works across every store.
With Deep Research, the card uses your own topics instead: the custom topics you define, plus topics your AI discovers from your conversations. That surfaces subjects specific to your business, like Express delivery delays or Gift card balance issues, rather than a general category. Deep Research also adds an Ask AI option when you hover over a row, which opens an insights view for that topic without leaving the page.
If you don't have the addon, a banner below the table offers to activate it.
Things to know about the numbers
The percentage column isn't a share of your total. Each row's percentage is that topic's own escalation rate — its escalations divided by its conversations. The column won't add up to 100%, and it isn't meant to.
Very small topics are held back. A topic needs at least three conversations in the period before it's ranked, so a single escalated conversation can't take the top row on a technicality. If almost nothing clears that bar, the card fills the remaining space with the topics that produced the most escalations, so you're not left with a nearly empty list.
Topics with no escalations don't appear. The card is a list of where handoffs happen, so a subject your AI handled without help every time is left out, however popular it was.
The list stops at ten topics. Anything past the tenth isn't shown.
Only classified escalations are counted. A conversation is included once it's been assigned a topic, which happens the day after it takes place. Escalations that haven't been classified aren't in the card, so on stores where classification coverage is still building, the totals here can be lower than the handoff count elsewhere on the page. Widening the date range is the quickest way to see a fuller picture.
Generic subjects are filtered out. Topics that only restate that the shopper asked for support don't name a reason, so they're excluded to keep the list actionable.
Making the Most of Escalations by Topic
- Start at the top row and click it. Read five or six of the real conversations. The fix is usually obvious once you see what the shopper actually asked.
- Check the Conversations column before you act. A 100% rate on four conversations is a hint. A 60% rate on 400 is a priority.
- Sort by Escalations once per review. The highest-rate topic shows you the biggest gap; the highest-volume topic shows you where your team's hours are going. Fix both kinds.
- Close the gap where it belongs. Missing policy details usually belong in your AI's knowledge; recurring scenarios that need an action are usually better handled by a support skill.
- Re-check the same date range a few weeks later. A topic dropping down the list is the clearest proof that a change worked.
Troubleshooting
The card says "No escalations in this period".
Possible causes: Every conversation in the range was handled by your AI without a handoff, or the range is very short.
Solution: Widen the date range. This message means there were no handoffs at all to break down.
The card says "No topics for these escalations".
Possible causes: Conversations were handed off, but none have been classified into a topic yet. Topics are assigned the day after a conversation happens, so a range that includes only very recent days may have nothing to show.
Solution: Choose a range that ends a day or two earlier, or widen it.
The order looks wrong — a topic with fewer escalations sits above one with more.
Possible causes: The list is sorted by escalation rate, not by escalation count.
Solution: This is expected. Click the Escalations header to sort by volume instead.
The percentages don't add up to 100%.
Possible causes: Each percentage is that topic's own escalation rate, not its share of all escalations.
Solution: No action needed — the column isn't meant to total 100%.
A topic you expected to see is missing.
Possible causes: It had no escalations in the period, it had fewer than three conversations, it fell outside the top ten, or its conversations haven't been classified yet.
Solution: Widen the date range, then check the Conversations page filtered to that topic to confirm the volume.
You see an error about the time range.
Possible causes: The selected range is longer than 400 days.
Solution: Choose a shorter range.
Need Help?
If the numbers don't look right, contact our support team with your store name, the date range and channel you selected, and the topic in question, and we'll take a look.
Related Articles
- Measuring Your AI Support Performance: Understanding the Support Analytics Dashboard
- Understanding Shopper Intelligence: Turn AI Conversations into Actionable Business Insights — uses the same topics to show conversation volume and satisfaction, rather than escalation rate
- Connect Customers to Your Support Team with AI-Powered Contact Support
- Build Your Own Support Skills: Automate Any Support Scenario with Custom Support Skills