Customer service metrics show you how fast, how well and how efficiently your support team works. The most important ones are first response time, resolution time, automation rate, ticket mix and CSAT. In our data from 131 shops, the first human reply without AI arrives after a median of 4.1 hours. A ticket is resolved after 3.0 days. Below you will find the formula, the measured median and the typical traps for each metric.
What do customer service metrics measure?
Each metric answers a concrete question from your day-to-day support. First response time shows how long a customer waits. Resolution time shows when her problem is settled. The automation rate shows how many tickets your team still handles itself.
A metric becomes truly useful once you have a comparison value from shops like yours. That is why every metric here comes with a measured median from the data of armincx by Chatarmin, our AI customer service for e-commerce. How AI works in support overall and which tools exist is covered in our guide to AI in customer service.
Which customer service metrics matter most?
To steer a shop's support, five customer service KPIs are enough. The table shows the formula, our measured value and the time period for each metric.
| Metric | Formula | Benchmark | Period |
|---|---|---|---|
| First response time (FRT) | Time from ticket arrival to the first human reply | 4.1 h without AI, 0.9 h after AI pre-sorting (median) | Jan to May 2026 |
| Resolution time | Time from ticket arrival to the resolved ticket | 3.0 days manual, 1.9 days AI-resolved, 1.0 day via workflow (median) | Jan to May 2026 |
| Automation rate | Tickets with AI or workflow ÷ resolved tickets with tracking data | 32.9% | Jun 2025 to Jun 2026 |
| Ticket mix | Tickets per request type ÷ all tickets | 25.2% "Where is my order?" | Jan to May 2026 |
| CSAT | Positive ratings ÷ all ratings | unpublished, base too small | Jun 2025 to Jun 2026 |
All time values are medians, meaning the middle value of all tickets. A few extremely slow tickets barely move the median, while they shift the average a lot.
How fast should the first reply in customer service be?
Without AI, customers in our data receive the first human reply after a median of 4.1 hours. When an AI pre-sorts the ticket and hands it to the team, it takes 0.9 hours. Both values come from the period January to May 2026.
What your customers expect depends on the channel. This is shown by a representative German survey by "Gewählt zum Kundenservice des Jahres" with 1,000 respondents aged 18 to 69. For written enquiries, people expect a reply within 24 hours at the latest, and on social media within one hour.
The two medians come from different groups of tickets and show a correlation, while saying little about the cause. They compare 31,098 tickets that an AI handed to the team with 1.15 million purely manual tickets. Our assessment: a pre-sorted ticket lands with the right person straight away, while in a shared inbox it waits until someone reaches it in order.
For tickets the AI resolves on its own, there is by definition no human first reply. That is why you compare the speed of AI and team via resolution time.
How long does it take to resolve a ticket?
Manually handled tickets are resolved after a median of 3.0 days, AI-resolved ones after 1.9 days. This is reported in the AI customer service benchmark across 131 shops. Tickets handled by a workflow, such as a shipment status request or a returns label, close after 1.0 day. That is around 37 and 65 per cent less time than manual handling.
Depending on the resolution path, first reply and resolution time vary widely. The table shows all four paths for the period January to May 2026.
| Resolution path | Tickets | First reply (median) | Resolution time (median) |
|---|---|---|---|
| Workflow | 242,185 | not applicable | 1.0 days |
| AI resolves alone | 91,180 | not applicable | 1.9 days |
| AI hands over to the team | 31,098 | 0.9 h | 2.7 days |
| Manual without AI | 1,146,411 | 4.1 h | 3.0 days |
Handed-over tickets get a quick reply and are still only resolved after 2.7 days. That is close to the 3.0 days for manual tickets. The quick first reply reassures the customer, and the actual work on the case still takes its time afterwards.
The fastest are workflows, because they carry out the same action for every request. The AI takes a little longer. Our assessment: the AI asks follow-up questions when a request is incomplete, for example for the order number, and every follow-up costs one reply cycle.
What is the difference between resolution time and average handling time?
Average handling time (AHT) measures the active working time on a ticket, resolution time the total duration until it is resolved. A ticket can be resolved after two days and still have taken only six minutes of work. Your customer feels the resolution time, your team feels the handling time.
Handling time requires time tracking per ticket, which our benchmark does not yet include. Formula, reference values and what AI shortens are covered in our article on handling time in customer service.
How many customer service tickets are automated?
AI or a workflow is involved in 32.9 per cent of resolved tickets, so one in three. We measured this across 2.75 million tickets with complete tracking data. A ticket counts as automated as soon as automation contributed to its resolution.
AI alone fully resolves 4.9 per cent of all tickets. When it touches a ticket, it resolves it without a human in one case out of five. The rest of the automation comes from workflows and from tickets where the AI contributes and your team closes the case.
For every automation rate, check the denominator first. A rate across all tickets and channels usually comes out lower than a rate across only the chats a bot touches. Many sales promises quote a high number and leave open what it refers to. So always ask about the ticket base, the channels and the time period.
The rate rises in peak season. From October to December 2025 it stood at 35.2 per cent, compared with 32.1 per cent for the rest of the year.
How wide is the range between shops?
The typical shop automates 16.3 per cent of its tickets, the top quarter 61.2 per cent on average. Calculated per shop, the average is 25.2 per cent. These values come from our automation benchmark with 140 shops and data as of July 2026.
On average, large shops automate more than small ones. Across all 2.62 million tickets in this study, the rate is 33.2 per cent, well above the 25.2 per cent per shop. That is only possible if shops with many tickets have higher rates.
You join the top quarter from 37.5 per cent. 35 of 140 shops achieve this. According to the study, the rate depends heavily on setup, industry and ticket mix. Our assessment from day-to-day work: the biggest levers are a clean connection between shop, inventory system and shipping, plus a well-maintained knowledge base.
For your shop, plan with the median as the starting point and the top quarter as the goal. That way you plan with values that real shops achieve.
How does the automation rate develop over the year?
The rate fluctuates from month to month between 24.8 and 38.5 per cent. The series comes from the customer service automation benchmark and starts in December 2025, the first month with over 90 per cent tracking coverage.
| Month | Automation rate |
|---|---|
| Dec 2025 | 36.8% |
| Jan 2026 | 38.2% |
| Feb 2026 | 38.5% |
| Mar 2026 | 30.0% |
| Apr 2026 | 24.8% |
| May 2026 | 35.6% |
| Jun 2026 | 34.0% |
In April the rate hit its lowest point of 24.8 per cent, with 1.9 times as many tickets as in February. In spring, the number of shops with tickets rose from 98 to 113. Our reading: new shops start with few workflows until connections and knowledge base are in place. In May the rate was back at 35.6 per cent.
Automation grows per shop over weeks and months. For your planning, the top quarter therefore says more about your potential than any single monthly value.
What do your customers write about?
Roughly one in four tickets is the question "Where is my order?". From January to May 2026 it made up 25.2 per cent of all tickets. Returns add 13.7 per cent and questions about invoices and payment 7.3 per cent.
A good half of the tickets belong to other groups. 27.1 per cent could not be assigned to a topic, and 26.8 per cent concern other topics. If you count only the assigned tickets, the parcel question reaches 54.6 per cent.
Email brings in 63.0 per cent of incoming tickets, the other channels 35.6 per cent combined. Instagram, at 15.4 per cent, is ahead of WhatsApp at 11.2 per cent. Facebook follows at 5.1, the website widget at 2.2 and TikTok at 1.7 per cent.
To get started, the parcel question is the biggest single lever, because the answer is already in your system. A workflow finds the order via email address or order number and sends the status and tracking link. What this looks like in practice is shown on the solution page for where-is-my-order requests.
Why is there no CSAT benchmark here?
In our data, only 15 of 131 shops capture CSAT cleanly, which is below our threshold of 20 shops. Below this limit we publish no values, because individual shops would shape the result too strongly.
CSAT itself is a good metric. It measures the share of positive ratings after a ticket, usually on a scale from 1 to 5. Alongside it there is NPS for willingness to recommend and CES for perceived effort.
In our cohort, only a minority measure satisfaction systematically. If you introduce CSAT, start with a single question directly after the resolution.
Which metrics can mislead you?
A metric without its counterpart leads to wrong decisions. You should know these four traps:
| Metric | Trap | How to measure it properly |
|---|---|---|
| First response time (FRT) | A quick first reply without a resolution creates a second message | Always analyse first reply together with resolution time |
| Automation rate | Without a denominator, the number says little | Fix the denominator: all tickets, all channels, fixed period |
| Averages | A few very slow tickets distort the average | Use the median |
| CSAT | A small number of ratings fluctuates heavily | Compare only above a fixed minimum number of ratings |
Translate the benchmarks to your own shop once. A D2C shop with 200 tickets a day handles around 33 tickets automatically at the median of 16.3 per cent. At the level of the top quarter, with 61.2 per cent, it is around 122 tickets. The difference of 89 tickets a day is work your team gets back.
Before any change, measure your own values for four weeks so that you can see the effect later. Record first reply, resolution time, automation rate and ticket mix per channel.
If you want to compare your values with these benchmarks, bring them to a conversation with us. Book an armincx demo, and we will go through your request types and look together at where workflows and AI can help you.
Where do the numbers come from?
All values come from aggregated and anonymised production data from armincx. The main study covers 131 shops with 2,907,990 resolved tickets from June 2025 to June 2026. Every shop was live for at least 60 days and resolved at least 50 tickets.
We calculate the rates across 2,746,345 tickets with complete tracking data. First reply, resolution times and ticket mix come from the period January to May 2026, when tracking coverage is almost complete. All time values are medians.
Range and monthly trend come from the automation benchmark with 140 shops and 2.62 million tickets from December 2025. We leave out earlier months because tracking was still being set up at the time. We publish shop segments only from 20 shops upwards.
FAQ: Common questions about customer service metrics
What metrics are there in customer service?
The most important are first response time, resolution time, handling time, automation rate, ticket mix, channel mix and satisfaction scores such as CSAT. For shops, ticket mix and automation rate are especially helpful because they show where work arises.
What is a good first response time in e-commerce?
In our data, the median is 4.1 hours without AI and 0.9 hours with AI pre-sorting. One hour on social media and 24 hours for written enquiries meet the expectations from the survey above.
What is a good automation rate?
The typical shop sits at 16.3 per cent, the top quarter at 61.2 per cent on average. According to the study, the rate depends heavily on setup, industry and ticket mix.
What is the difference between first reply time and resolution time?
First reply time measures the time until the first human reply, resolution time the time until the ticket is resolved. Your customer feels the first immediately, the second shows how efficiently your team works.
Is average handling time the same as resolution time?
No. Handling time measures the active working time on the ticket, resolution time the total duration until resolution including waiting time.
What is the most common request type in e-commerce support?
The question "Where is my order?", with 25.2 per cent of all tickets, so roughly one in four. Returns follow with 13.7 per cent and questions about invoices and payment with 7.3 per cent.
Conclusion: Which metrics should you measure from tomorrow?
Start with four metrics: first reply, resolution time, automation rate and ticket mix. Our medians give you the comparison values. Without AI, the first reply takes 4.1 hours and resolution 3.0 days. One in three tickets is automated and roughly one in four is a parcel question.
Read every metric together with its counterpart and with a clear denominator. Then you can see where your shop stands and which request types to tackle next. Your next step is a four-week measurement of your own values as a baseline.
