Measuring Your AI Support Performance: Understanding the Support Analytics Dashboard

The Support Analytics dashboard provides valuable insights into how your AI Concierge assists with customer service inquiries. This powerful tool helps you measure automation effectiveness, understand human escalation patterns, and quantify the efficiency gains from using Rep AI to handle support tickets.

1. Log in to your Rep AI Console

2. Click "AI Support" in the left navigation menu

3. Select "Support Analytics"

What it shows: The subjects your AI hands off to a human most often, each with its escalations, its total conversations, and its escalation rate. This is the first card on the page.
Why it matters: It turns a single handoff number into a list of specific subjects you can work on. Rows are ordered by escalation rate, highest first, so the top row is the subject your AI is least able to finish on its own — not necessarily the one with the most handoffs. Click any topic name to open the conversations behind it.

For the full breakdown — sorting, Deep Research topics, and why a topic might be missing from the list — see the help article "Understanding Escalations by Topic: See Which Subjects Send Shoppers to Your Team."

What it shows: The share of conversations your AI handled on its own, together with the numbers behind it — for example 94.33% · 632 of 670 conversations. The first number is the conversations your AI handled without passing anyone to a human agent; the second is all the conversations it handled in the period you selected.
Why it matters: The rate tells you how well your AI is performing. The counts tell you how much it did — the absolute volume you can report internally without asking anyone to pull it for you.

What it shows: The share of conversations that went to a human agent, with the counts beside it — for example 5.67% · 38 of 670 conversations. The first number is the conversations handed off to a human; the second is the same total your AI handled.
Why it matters: Read together with Resolved by AI, this shows exactly how many customers your team still needs to touch. On web traffic the two success counts add up to the shared total, so the two cards always tell one story.

What it shows: The share of shopper questions your AI answered, with the counts beside it — for example 92.59% · 1,299 of 1,403 questions.
Why it matters: This is your AI's knowledge coverage in absolute terms. Note the unit: this card counts questions, not conversations — a single conversation can contain several questions.

What it shows: The number of customer questions your AI couldn't answer effectively.
Why it matters: This highlights knowledge gaps that you can address to improve your AI's performance.

What it shows: The number of customer-initiated support conversations that were fully handled by your AI.
Why it matters: Each deflected ticket represents time saved for your support team.

What it shows: An estimate of cost savings from AI-handled support tickets (calculated as a set amount per deflected ticket).
Why it matters: This helps quantify the ROI of your AI investment in concrete financial terms.

What it shows: How your helpful and not-helpful ratings compare across the period you've selected. Shoppers can rate a conversation by tapping the rating icon in the chat widget, and your AI also asks for a rating automatically when a conversation reaches its end — both feed this chart as a combined total.
Why it matters: This provides direct insight into how customers perceive their AI support experience.

What counts toward this chart: the total includes shoppers who gave a helpful or not-helpful rating and shoppers who followed an AI product recommendation, so it is broader than the count of explicit ratings on their own. To read the ratings conversation by conversation, go to Conversations and filter by Reported as helpful or Reported as unhelpful.

Each of the three rate cards — Resolved by AI, Handled by humans, and Question-answering rate — shows three things stacked together:

1. The rate, as a percentage.

2. The counts behind it, in the form "632 of 670 conversations" — how many succeeded, and how many were counted in total.

3. A one-line definition of exactly what is being counted and what it is divided by.

The definitions read like this:

• Resolved by AI — "Conversations the AI handled without handing off to a human, divided by all conversations it handled."

• Handled by humans — "Conversations handed off to a human agent, divided by all conversations the AI handled."

• Question-answering rate — "Questions the AI answered, divided by all questions shoppers asked — counted in questions, not conversations."

The definition uses whatever name you gave your AI, so it reads the same way as the rest of your Console.

The counts always match the rate. Divide the first count by the second and you get the percentage shown beside it, to the same two decimal places. Both numbers come straight from your data, so a card never disagrees with itself.

The counts follow your filters. Change the date range, switch between new and returning customers, or pick a traffic source, and the counts move with the rate. What you see is always the slice you selected.

Where the counts appear. The counts are shown for your web traffic — the Web, Desktop, and Mobile platform options. On the messaging and email channels (Instagram DM, Facebook DM, WhatsApp DM, Helpdesk) and on the combined All view, the cards show the rate together with its definition. On messaging and email channels the definition changes to match how the rate is measured there: Resolved by AI reads "Conversations the AI resolved, divided by all engaged conversations on this channel."

Quiet periods say so. If the period you selected has no conversations at all, the card reads "No data for this period" in place of the counts, and the trend line and the period-over-period comparison are hidden — so a placeholder percentage is never mistaken for a measured result.

Question-answering rate and Unanswered questions count different things. The Question-answering rate is measured in questions asked and answered. Unanswered questions is its own metric with its own source, so it will not equal the gap in the Question-answering rate. Use each on its own terms: the rate for coverage, Unanswered questions for the specific gaps to fill in your AI's knowledge.

The bottom section of your dashboard displays specific support skills and their performance metrics:

• Skill: The type of support task your AI can handle (e.g., Cancel Order, Change Order Address)

• Created At: When the skill was added to your AI

• Conversations: Number of times this skill was used

• Successfully Handled: How many conversations were fully resolved by the AI

• Success Rate: Percentage of successful resolutions

If you use AI Email Answering, set the Platform filter to Helpdesk to see the Email Resolution Chart. It is a pie chart showing what happened to every email ticket your AI processed, so you can measure how much work it took off your team and fine-tune the rules that control which emails it answers.

The center of the chart shows Tickets Analyzed — every email ticket the AI processed in the selected date range — divided into four outcomes, each with a count and percentage:

What it shows: The AI answered the customer and the conversation closed without a human stepping in.
Why it matters: This is your AI handling tickets end-to-end.

What it shows: The AI replied and is waiting for the customer to respond.
Why it matters: The ticket is not stuck — the ball is in the customer's court.

What it shows: The AI could not fully answer and left the ticket for a human.
Why it matters: A rising slice here points to knowledge gaps worth filling.

What it shows: The AI intentionally skipped the email because it matched a Do-Not-Answer rule or automatic spam filtering.
Why it matters: This is a good outcome — the right emails reach your team instead of getting an automated reply.

One ticket, one slice: Each ticket is counted only in its final outcome, so the four slices always add up to your Tickets Analyzed total. A ticket that was Pending and later Resolved appears under Resolved only.

Alongside the chart you will also see the resolution rate (resolved divided by analyzed), the block rate (blocked divided by analyzed), and period-over-period comparison on each slice — improvements shown in green, the opposite in red.

Every part of the chart is clickable, so you can go from a number to the actual tickets behind it:

• Tickets Analyzed total: the full list of email tickets processed, with subject, customer, date, and outcome.

• Blocked by Do-Not-Answer Rules: a second pie chart breaking blocked tickets down by individual rule, each labeled Default (a Rep AI built-in rule) or Custom (one you created). Click a rule to see the exact tickets it caught.

• Could Not Be Resolved: the escalated tickets, with the reason the AI could not answer.

• Pending Customer Response: tickets awaiting a reply, including how long each has been open.

• Resolved: the tickets the AI closed on its own.

To adjust the rules behind the Blocked slice, see the help article "Automate Customer Support with the AI-Powered Answer Emails Support Skill."

When a drill-down opens an individual blocked or unanswered conversation, an indicator sits under the customer's message so you do not have to guess why the AI stayed quiet:

• Blocked by a rule: the indicator shows the rule name and a Default or Custom label. Click it to review the rule — custom rules are editable, default rules open as read-only.

• Could not be resolved: the indicator shows the reason with a neutral label, so you can tell the AI deliberately handed the ticket off rather than missing it.

• No email answering yet, or not on your plan: the chart shows a prompt to set up AI Email Answering or to upgrade.

• No activity in the period: the chart shows an empty state — try widening the date range.

• Email answering toggled on or off mid-period: only tickets from the days when it was active are counted, keeping your numbers accurate.

• The chart is not showing: make sure the Platform filter is set to Helpdesk, AI Email Answering is enabled, and your plan includes it.

• The slices do not add up as expected: remember each ticket is counted once, in its final outcome only.

• The Blocked slice is larger than expected: this usually means spam filtering and your rules are working. Click the slice to see which rules are catching tickets and adjust any that are too broad.

To gain more targeted insights, use the filtering options available:

1. Date Range Selection:

   • Click the date picker to select a custom time period

   • Use preset options like Today, Yesterday, This Week, or Last Month

   • Apply custom date ranges using the calendar interface

2. Customer Segments:

   • Filter by New or Returning customers

   • See how different customer types interact with your AI

3. Traffic Source:

   • Filter by Direct, Referral, Search, or Social traffic

   • Understand how acquisition channel affects support needs

4. Platform:

   • Select Web (Desktop + Mobile), Desktop only, Mobile only, or Simulator

   • Compare AI performance across different platforms

   • View messaging platform data when available (Instagram DM, Facebook DM, Whatsapp DM)

   • Select Helpdesk to switch the dashboard to your email ticket metrics and view the Email Resolution Chart

• Compare to Previous Periods: Toggle the comparison feature to see how your metrics have changed over time

• Report the Volume, Not Just the Rate: The counts on Resolved by AI, Handled by humans, and Question-answering rate give you the absolute numbers for the period you selected — ready to paste into a weekly or monthly report

• View Conversations: Click "See conversations" links to review actual chat transcripts related to each metric

• Address Unanswered Questions: Review and update your AI's knowledge base to fill identified gaps

• Calculate ROI: Use the Potential Savings metric to demonstrate the value of your AI investment

If your metrics show room for improvement:

1. Review unanswered questions to identify knowledge gaps

2. Check your Support Skills settings to ensure they're properly configured

3. Consider expanding your AI's training with additional product information or customer service policies

4. Adjust escalation thresholds if too many or too few conversations are being handed to human agents