How metrics are calculated: AI & automationUpdated 2 hours ago
AI & automation analytics track how effectively your AI Agent and automation features are handling customer interactions and driving revenue. When reviewing your automation performance, you can use this guide to understand each metric and how it’s calculated.
Requirements
- Available on all Helpdesk plans with the AI Agent add-on
- All roles can review AI Agent & automation analytics reports
AI & Automation analytics
Your AI Agent and automation performance metrics are divided into two reports in the AI & automation section of your Gorgias analytics. To open a report, follow the steps below:
- From your helpdesk, click the dropdown menu in the top-left corner.
- Select Analytics.
- In the menu, locate AI & automation, then select either Overview or AI Agent.
The Overview page shows performance metrics across all automation features, including AI Agent, Flows, and Order Management. The AI Agent page focuses specifically on AI Agent's performance, including revenue generated from Shopping Assistant conversations.
Available metrics
To learn more about the metrics included in each report, select one from the list below to jump to its definition in this article, or scroll through this article to browse every definition.
Overview metrics
AI Agent metrics
Overview report
The Overview report surfaces high-level metrics across all automation features, letting you see the combined impact of AI Agent and any other automations you have running.
| Automated interactions* | |
| Metric type | Volume |
| Definition | Interactions resolved by automation features from start to finish, with no human involved. |
| How it's calculated | Calculated by counting the number of distinct billable automated interactions resolved by an automation feature with no human intervention. |
| Notes Each billable event is counted once. Spam tickets are not billed as automated interactions, and are not included in this calculation. Tickets where AI Agent sent a reply but then handed over are counted as handover interactions rather than automated interactions. | |
| Cost saved* | |
| Metric type | Volume |
| Definition | Estimated cost savings from conversations AI Agent handled instead of a human. |
| How it's calculated | Calculated by multiplying the number of automated interactions in the period by the cost difference between tickets resolved by humans (billable ticket cost† + $3.10) and tickets resolved by an automated interaction (cost of your current AI Agent plan ÷ AI interaction quota or $0.85 if your current plan doesn't include a quota) †Billable ticket cost = cost of your current Helpdesk plan ÷ the number of tickets included in the plan |
| Notes The calculation always uses pricing from your current plan, even when viewing historical data. If you change plans, past values will reflect your current plan pricing. Non-USD accounts see the $3.10 and $0.85 defaults in USD even when other components are in local currency. | |
| Decrease in First Response Time* | |
| Metric type | Median |
| Definition | How much faster shoppers receive a first reply when automation handles tickets vs. humans. |
| How it's calculated | Calculated by subtracting the median first response time for an automation feature (time in seconds from the first customer message to the first AI Agent message) from the median first response time for humans (time in seconds from the first customer message to the first human message). |
| Notes Tickets where AI Agent sent any message are excluded from the human first response time median to ensure tickets aren't counted twice. When the first response is sent by Flows, Order Management, Article Recommendations, or Autoresponders, a fixed 60-second response time is used. | |
| Decrease in Resolution Time* | |
| Metric type | Median |
| Definition | How much faster automation resolves tickets on average vs. humans. |
| How it's calculated | Calculated by subtracting the median resolution time for automation features (time in seconds from ticket creation event to the ticket close event by AI Agent) from the median resolution time for humans (time in seconds from ticket creation event to the ticket close event by a human). |
| Notes Tickets where AI Agent sent any message are excluded from the human resolution time median to ensure tickets aren't counted twice. When the ticket is resolved by Flows, Order Management, Article Recommendations, or Autoresponders, a fixed 60-second response time is used. | |
| Handover interactions | |
| Metric type | Volume |
| Definition | Interactions that ended with a shopper being connected to a human. |
| How it's calculated | Calculated by counting the number of conversations transferred from an automation to a human. Handovers are defined as conversations with AI Agent that are transferred to a human, requests to speak to a human during a Flow, or when an Article Recommendation or Order Management response is marked as "not helpful." |
| Notes Handovers on AI Agent tickets are deduplicated per ticket; Flows, Article Recommendations, and Order Management ticket handovers count per event. Spam tickets aren't explicitly excluded from the calculation, however handovers on spam tickets are rare in practice. | |
| Overall automation rate* | |
| Metric type | Percentage |
| Definition | Percentage of all shopper conversations fully handled by automation, with no human involved. |
| How it's calculated | Calculated by dividing billable automated interactions fully resolved by AI Agent or an automation by your total billable workload (all automated interactions + all human-handled tickets). |
| Notes Spam tickets are excluded from this calculation. The total billable workload denominator is used for each feature-specific automation rate, so the sum of all feature-specific rates equals your overall rate. | |
| Time saved by agent* | |
| Metric type | Median |
| Definition | Estimated time your team saves from conversations automation handled instead of a human. |
| How it's calculated | Calculated by multiplying median handle time (drafting time + working time + phone call time in seconds) of closed, non-automated tickets by the total number of automated interactions. |
| Notes Occasionally, the handle timer may silently fail to register a stop time. To prevent inflated handle times in these cases, their handle time calculation is capped to the report's selected period. | |
*Data delay: An interaction is only billed as automated once 72 hours have passed without a human handover, so the most recent 3 days aren't immediately reflected in this metric.
AI Agent report
The AI Agent report provides metrics specific to how AI Agent handles customer and shopper interactions. You can isolate metrics related to support-based interactions in the Support Agent tab. In the Shopping Assistant tab, you can see metrics related to sales, revenue, discounts, and product recommendations.
| AI Agent automation rate* | |
| Metric type | Percentage |
| Definition | Percentage of shopper interactions fully handled by AI Agent, with no human involved. |
| How it's calculated | Calculated by dividing the number of billable automated interactions resolved by AI Agent by your total billable workload (all automated interactions + all human-handled tickets) over the selected period. |
| Notes Spam tickets are excluded from the calculation. The total billable workload is also the denominator for the overall automation rate, so your AI Agent automation rate is a share of that total. | |
| Automated interactions* | |
| Metric type | Volume |
| Definition | Interactions AI Agent resolved from start to finish, with no human involved. |
| How it's calculated | Calculated by counting the number of distinct billable automated interactions resolved by AI Agent with no human intervention. |
| Notes Each billable event is counted once. Spam tickets are not billed as automated interactions, and are not included in this calculation. Tickets where AI Agent sent a reply but then handed over are counted as handover interactions rather than automated interactions. | |
| Average CSAT | |
| Metric type | Average |
| Definition | Average satisfaction (CSAT) score for interactions handled during the selected period. |
| How it's calculated | Calculated by dividing the sum of all survey scores (1-5) from tickets where AI Agent sent at least one response during the period by the total number of scored surveys. Tickets that received no survey response are excluded. When there are multiple surveys per ticket, only the most recent score is used. |
| Notes A ticket handled in one period whose survey arrives in the next will only count toward the period the survey was sent in. Unlike the CSAT metrics in the Support Performance and Quality reports (which only count tickets where AI Agent is the last assignee), this metric includes tickets that AI Agent handed over. | |
| Average discount amount | |
| Metric type | Average dollar value |
| Definition | Average value of discount codes generated by Shopping Assistant and used by shoppers. |
| How it's calculated | Calculated by dividing the sum of discounts on orders that used a Shopping Assistant discount code by the number of orders that used a Shopping Assistant discount code. Conversations where Shopping Assistant sent a code but the customer never ordered are excluded. |
| Notes An order is attributed to Shopping Assistant when the customer uses at least one code that Shopping Assistant sent in the conversation within 3 days of the last message from Shopping Assistant. If Shopping Assistant hands over a conversation to a human, attribution is voided and the conversation is excluded from the calculation. | |
| Average order value | |
| Metric type | Average dollar value |
| Definition | Average order value for purchases made within 3 days of a Shopping Assistant interaction. |
| How it's calculated | Calculated by dividing the total net sales (order total - refunds) from orders attributed to Shopping Assistant by the number of attributed orders. Amounts from non-USD stores are converted to USD at the order's exchange rate. Voided, pending, authorized, and refunded orders are excluded from the calculation. |
| Notes As Shopping Assistant has built-in spam detection, spam tickets are excluded from the calculation. An order is attributed to Shopping Assistant when a customer interacts with Shopping Assistant, then places an order within 3 days of the latest message from Shopping Assistant. If Shopping Assistant hands over a conversation to a human, attribution is voided and the conversation is excluded from the calculation. | |
| Buy-through rate | |
| Metric type | Percentage |
| Definition | Percentage of Shopping Assistant interactions with product recommendations that led to a purchase. |
| How it's calculated | Calculated by dividing the number of orders where the customer purchased a product recommended by Shopping Assistant (excluding voided, pending, authorized, and refunded orders) by the number of conversations where Shopping Assistant made a product recommendation (including conversations that were handed over), then multiplying by 100. |
| Notes An order is attributed to a Shopping Assistant product recommendation if the customer purchases the product within 3 days of the last message sent by Shopping Assistant in the conversation where it was recommended. | |
| Click-through rate | |
| Metric type | Percentage |
| Definition | Percentage of product recommendations that shoppers clicked on. |
| How it's calculated | Calculated by dividing the number of conversations where a shopper clicked a product recommendation or added it to their cart by the number of conversations that contain at least one product recommendation, then multiplying by 100. A click is counted as a shopper clicking a product recommendation or adding a product to their cart. |
| Notes As Shopping Assistant has built-in spam detection, spam tickets are excluded from the calculation. Conversations with multiple clicks or multiple product recommendations are only counted once. | |
| Closed tickets | |
| Metric type | Volume |
| Definition | Tickets AI Agent closed in the selected period, including those closed without sending a reply. |
| How it's calculated | Calculated by counting the number of tickets currently assigned to AI Agent where the closed date falls within the selected time period and the ticket is still closed. Tickets that were closed, but re-opened within the selected period are excluded. Tickets that re-opened and were re-closed within the selected period count once. |
| Notes Spam tickets are excluded from this calculation. Closed tickets without any messages are included in the count. If a ticket is closed by AI Agent but reassigned to a human, it won't be included in the count. | |
| Conversion rate | |
| Metric type | Percentage |
| Definition | Percentage of Shopping Assistant interactions that led to an order within 3 days. |
| How it's calculated | Calculated by dividing the number of orders attributed to Shopping Assistant by the total number of Shopping Assistant conversations (including handovers), then multiplying by 100. A single Shopping Assistant conversation that leads to multiple orders is only counted once. |
| Notes An order is attributed to Shopping Assistant when the customer makes a purchase within 3 days of the last message from Shopping Assistant. As Shopping Assistant has built-in spam detection, spam tickets are excluded from the calculation. | |
| Cost saved* | |
| Metric type | Volume |
| Definition | Estimated cost savings from interactions AI Agent handled instead of a human. |
| How it's calculated | Calculated by multiplying the number of AI Agent interactions in the period by the cost difference between tickets resolved by humans (billable ticket cost† + $3.10) and tickets resolved by AI Agent (cost of your current AI Agent plan ÷ AI interaction quota or $0.85 if your current plan doesn't include a quota) †Billable ticket cost = cost of your current Helpdesk plan ÷ the number of tickets included in the plan |
| Notes The calculation always uses pricing from your current plan, even when viewing historical data. If you change plans, past values will reflect your current plan pricing. Non-USD accounts see the $3.10 and $0.85 defaults in USD even when other components are in local currency. | |
| Coverage rate | |
| Metric type | Percentage |
| Definition | Percentage of tickets from AI Agent-supported channels that AI Agent attempted to handle. |
| How it's calculated | Calculated by dividing the number of tickets that AI Agent resolved, handled, or handed over by the total number of tickets in the helpdesk (including spam). |
| Notes For tickets resolved by AI Agent without human intervention, review the Success Rate metric instead. | |
| Decrease in First Response Time* | |
| Metric type | Median |
| Definition | How much faster shoppers receive a first reply when AI Agent handles the ticket vs. humans. |
| How it's calculated | Calculated by subtracting the median first response time for AI Agent (time in seconds from the first customer message to the first AI Agent message) from the median first response time for humans (time in seconds from the first customer message to the first human message). |
| Notes Tickets where AI Agent sent any message are excluded from the human first response time median to ensure tickets aren't counted twice. | |
| Decrease in Resolution Time* | |
| Metric type | Median |
| Definition | How much faster AI Agent resolves tickets on average vs. humans. |
| How it's calculated | Calculated by subtracting the median resolution time for AI Agent (time in seconds from ticket creation event to the ticket close event by AI Agent) from the median resolution time for humans (time in seconds from ticket creation event to the ticket close event by a human). |
| Notes Tickets where AI Agent sent any message are excluded from the human resolution time median to ensure tickets aren't counted twice. | |
| Discount codes applied | |
| Metric type | Volume |
| Definition | Purchases completed using a discount code Shopping Assistant generated and sent. |
| How it's calculated | Calculated by counting orders (including voided and refunded orders) placed within 3 days of the last Shopping Assistant message that used a discount code generated by Shopping Assistant. Individual orders that used multiple Shopping Assistant discount codes are counted once; multiple orders that use the same discount code are counted separately. |
| Notes As Shopping Assistant has built-in spam detection, spam tickets are excluded from the calculation. If Shopping Assistant hands over a conversation to a human, the conversation is excluded from the calculation. | |
| Discount usage | |
| Metric type | Percentage |
| Definition | Percentage of Shopping Assistant discount codes that shoppers used. |
| How it's calculated | Calculated by dividing the number of orders placed within 3 days of the last Shopping Assistant message where a Shopping Assistant discount code was used by the number of conversations where Shopping Assistant generated at least one discount code, then multiplying by 100. |
| Notes As Shopping Assistant has built-in spam detection, spam tickets are excluded from the calculation. If Shopping Assistant hands over a conversation to a human, the conversation is excluded from the calculation. | |
| Discounts offered | |
| Metric type | Volume |
| Definition | Interactions where Shopping Assistant generated a discount code for a shopper. |
| How it's calculated | Calculated by counting the number of conversations where Shopping Assistant sent at least one discount code. |
| Notes As Shopping Assistant has built-in spam detection, spam tickets are excluded from the calculation. If Shopping Assistant hands over a conversation to a human, the conversation is excluded from the calculation. | |
| First response time | |
| Metric type | Median |
| Definition | The median time between a customer's first message on a ticket and the first reply from AI Agent. |
| How it's calculated | Calculated by counting the time in seconds from the first customer message to the first AI Agent message. A ticket will be included in the selected period if AI Agent's reply fell within the period, even if the customer's first message did not. |
| Notes Spam and deleted tickets are excluded from the calculation. We use the median rather than the average so a few unusually slow tickets don't skew the number. | |
| Handover interactions | |
| Metric type | Volume |
| Definition | Interactions AI Agent transferred to a human for further support. |
| How it's calculated | Calculated by counting the number of conversations transferred from AI Agent to a human. |
| Notes Handovers on AI Agent tickets are deduplicated per ticket. Spam tickets aren't explicitly excluded from the calculation, however handovers on spam tickets are rare in practice. | |
| Human response time after AI handoff | |
| Metric type | Median |
| Definition | Median time for a human to first respond after AI Agent transfers a ticket to them. |
| How it's calculated | The median time (in seconds) from the last AI Agent message in a ticket to the first human response (excluding internal notes). A ticket will be included in the selected period if the human response fell within the period, even if AI Agent's last message did not. |
| Notes Spam and deleted tickets are excluded from the calculation. If a human replies to a ticket being handled by AI Agent without a formal handover event, that message still counts as the first human response. | |
| Median purchase time | |
| Metric type | Median |
| Definition | Median time between a Shopping Assistant conversation and a shopper placing an order. |
| How it's calculated | The median time (in seconds) from the last Shopping Assistant message to the customer placing an order. |
| Notes We use the median rather than the average so a few unusually slow tickets don't skew the number. When an order is made within 3 days of multiple Shopping Assistant conversations, the most recent is used for the calculation. If Shopping Assistant hands over a conversation to a human, the conversation is excluded from the calculation. | |
| Orders influenced | |
| Metric type | Volume |
| Definition | Orders placed within 3 days of a Shopping Assistant interaction where no human handover occurred. |
| How it's calculated | Calculated by counting the number of orders placed within 3 days of the last message in a Shopping Assistant conversation. Orders placed within 3 days of more than one Shopping Assistant conversation are only counted once. |
| Notes An order is attributed to Shopping Assistant when a customer interacts with Shopping Assistant, then places an order within 3 days of the latest message from Shopping Assistant. If Shopping Assistant hands over a conversation to a human, the conversation is excluded from the calculation. | |
| Product recommendations | |
| Metric type | Volume |
| Definition | Interactions where Shopping Assistant recommended at least one product to a shopper. |
| How it's calculated | Calculated by counting the number of conversations (including handovers) where Shopping Assistant recommended at least one product. |
| Notes As Shopping Assistant has built-in spam detection, spam tickets are excluded from the calculation. | |
| Resolution time | |
| Metric type | Median |
| Definition | The median time between a customer's first message on a ticket and the ticket's close — for tickets handled by AI Agent. |
| How it's calculated | Calculated by counting the time in seconds between a customer's first message and the ticket's close event on tickets that include at least one customer message and one reply, and are currently assigned to AI Agent. A ticket will be included in the selected period if it was closed (or re-closed) during the period, regardless of when the ticket was opened. |
| Notes Spam tickets are excluded from the calculation. If a ticket is closed by AI Agent but reassigned to a human, it won't be included in the count. We use the median rather than the average so a few unusually slow tickets don't skew the number. | |
| Revenue influenced | |
| Metric type | Volume |
| Definition | Total revenue from orders placed within 3 days of a Shopping Assistant interaction. |
| How it's calculated | Calculated by totalling the net order values (order total minus refunds) of all orders attributed to Shopping Assistant. |
| Notes An order is attributed to Shopping Assistant when a customer interacts with Shopping Assistant, then places an order within 3 days of the latest message from Shopping Assistant. Voided, pending, and refunded orders are excluded. Non-USD amounts are converted to USD. | |
| Revenue influenced per interaction | |
| Metric type | Average |
| Definition | Average revenue generated per Shopping Assistant interaction. |
| How it's calculated | Calculated by dividing the net revenue (order total minus refunds) attributed to Shopping Assistant by the total number of Shopping Assistant conversations in the period (includes handovers). Voided, pending, and refunded orders are excluded. |
| Notes Handovers are included in the total number of Shopping Assistant conversations to intentionally lower the average and demonstrate that not every Shopping Assistant conversation leads to a sale. | |
| Success rate* | |
| Metric type | Percentage |
| Definition | Percentage of AI Agent interactions fully resolved without escalating to a human. |
| How it's calculated | Calculated by dividing the number of billable AI Agent interactions by the total number of tickets handled by AI Agent (including handovers), then multiplying by 100. |
| Notes Spam tickets are excluded from the calculation. | |
| Time saved by agent* | |
| Metric type | Median |
| Definition | Estimated time your team saves from conversations AI Agent handled instead of a human. |
| How it's calculated | Calculated by multiplying median handle time (drafting time + working time + phone call time in seconds) of closed, non-automated tickets by the total number of AI Agent conversations. |
| Notes Occasionally, the handle timer may silently fail to register a stop time. To prevent inflated handle times in these cases, their handle time calculation is capped to the report's selected period. | |
*Data delay: An interaction is only billed as automated once 72 hours have passed without a human handover, so the most recent 3 days aren't immediately reflected in this metric.
