First response time (FRT) is how long a customer waits between contacting you and getting the first reply that addresses their request. It is one of the core customer service metrics, and it only means something once you've decided three things: what counts as a reply (an automatic acknowledgement, an AI answer or only a human agent), which clock you use (business or calendar hours), and whether you report the median or the average.
Get those wrong and the number looks excellent while customers wait: an auto-reply saying "we've got your message" brings FRT close to zero and changes nothing. This guide is for teams answering order, delivery, return and refund questions at volume on WhatsApp, Instagram, email, chat and phone.
In short
- A first response answers the request, asks for the one thing needed to act, or says what happens next and by when. Acknowledgements don't count.
- Measure per channel from the customer's first message, in business hours and in calendar hours.
- Report the median, the 90th percentile and the share within target, plus conversations still waiting for any reply.
- Count an AI reply only if it passes the same test as a human one, and track the human wait after handover separately.
What first response time measures, and what it doesn't
Several different waits get called "response time":
| Measure | Clock starts | Clock stops | What it tells you |
|---|---|---|---|
| Time to acknowledgement | Customer's first message | Any automatic reply (a ticket number, "we'll get back to you") | That your auto-reply works. Not a service measure |
| First response time | Customer's first message | The first reply that addresses the request, from an AI agent or a human agent | How long the customer waited to be helped |
| First human response time | Customer's first message, or the AI agent's handover | A human agent's first reply | Whether your queue and staffing keep up |
| Average (or next) reply time | Each customer message | The reply to it | Responsiveness across the whole conversation |
| Resolution time | Customer's first message | The request is completed | How long the whole problem took |
First response time is not resolution time: a fast first reply can still lead to a week of back-and-forth. Nor is it average handle time, which measures how long an agent spends working a conversation, not how long the customer waits for one. On the phone, the equivalent is speed of answer (see the channel table below).
Three decisions behind the number
1. What counts as a first response
"First meaningful reply" is the usual phrase. To make it testable, a first response must do one of three things with the customer's request:
- It answers it. "Your order shipped on Tuesday. The courier's last scan was this morning, and delivery is expected on Thursday."
- It asks for the one thing needed to act. "Please send a photo of the damaged part so we can arrange a replacement."
- It says what happens next and by when. "Our returns team is checking the pickup with the courier and will reply by 11am tomorrow."
Three things don't count: an automatic acknowledgement, "Hi, how can I help?" after the customer has already explained, and "We're looking into it" with no next step. Report those as acknowledgements.
2. Business hours or calendar hours
Report both, because they answer different questions. Take a team that works 10am to 7pm, Monday to Friday. A customer messages on WhatsApp at 9:40pm on Friday, and an agent replies at 10:15am on Monday.
- Business-hours FRT: 15 minutes. The team answered quickly once open. Use this for staffing and agent performance.
- Calendar FRT: 60 hours 35 minutes. This is what the customer experienced. Use it to decide whether you need evening and weekend cover, AI after hours, or 24/7 customer support.
The reply also missed WhatsApp's own clock, which matters wherever WhatsApp is the main support channel, as it is for many consumer brands in India. On Meta's WhatsApp Business Platform, a customer's message opens a 24-hour customer service window; once it closes, the business can send only pre-approved templates. The Monday reply could go out only as a WhatsApp message template, not as a normal answer.
3. Median, percentile and share within target
A few long waits distort an average. Ten WhatsApp conversations answered in 2, 3, 3, 4, 5, 6, 8, 9, 12 and 240 minutes have an average FRT of 29 minutes and a median of 5.5 minutes. The average overstates the typical wait; the median hides the customer who waited four hours.
So report three numbers per channel: the median for the typical wait, the 90th percentile for the slow tail, and the share of conversations answered within your target. Set targets on the last one.
How to calculate first response time
For one conversation:
First response time = time of the first reply that addresses the request − time of the customer's first message
For a period, take the median across conversations, and include every conversation that arrived, not only the ones that have been answered.
That last point is where helpdesk reports mislead. Freshdesk divides by the tickets whose first responses were sent in the selected period, and Gorgias counts a ticket once its first agent reply falls in the period. So on a bad Saturday, the conversations nobody has answered are missing from Saturday's figure. Track the conversations still waiting for a first response, and the oldest one's age, next to FRT.
Check, too, where your tool starts and stops the clock (help documentation checked on 28 September 2026):
- Zendesk measures from ticket creation to the first public agent comment, in both calendar and business hours, and ignores most automated and bot actions. If a ticket is created only when a bot hands over, the time before that isn't in the number.
- Gorgias ignores messages sent by rules, replies sent from outside the helpdesk (such as directly from Gmail) and tickets started by an agent, includes time outside business hours, and reports the median.
Where the clock starts on each channel
Two channels have deadlines set by the platform: for WhatsApp customer support and Instagram DMs, Meta's messaging windows decide whether a late reply can be a normal message at all.
| Channel | Start the clock when | Counts as the first response | Watch for |
|---|---|---|---|
| The customer's first message about a new issue | The first reply that addresses the request, from the AI agent or a human agent | The 24-hour customer service window: after it closes, only approved templates can be sent. Messages on an issue that's already open are reply time, not a new FRT | |
| Instagram DMs and Messenger | The customer's first DM | The same test | A 24-hour window for replies; Meta's human agent tag allows a manual reply for up to seven days. Comments on posts are a separate queue, so measure them separately |
| The email arrives in the support inbox, not when someone opens it | A reply from the support inbox that addresses the request | Auto-replies with a ticket number; tickets agents create themselves; replies from personal mailboxes the helpdesk can't see | |
| Live chat | The customer sends their first message, not when the widget opens | A reply that addresses the question | Welcome messages and "an agent will be with you shortly"; customers who leave before any reply (count them) |
| Phone | The caller joins the queue after the IVR menu | A human agent, or a voice AI agent, answers | Abandoned calls and callbacks: report abandonment and time to callback next to speed of answer |
Count the first response once per issue, not once per channel. If a customer asks about a damaged order on Instagram and then on WhatsApp, the second message belongs on the open ticket, as it would in omnichannel customer service, not on a new clock. Merge duplicates before you report, or the faster duplicate flatters the number.
In Flowcall's AI Helpdesk, WhatsApp, email, Instagram, Messenger, Live Chat and phone conversations arrive in one inbox. When the same customer writes on another channel while their issue is open, Flowcall keeps it on the same ticket, with the whole history visible; a new issue or a different order gets a ticket of its own. That way, the ticket already matches the issue, and there are fewer duplicates to merge before you report.
A first response time report
| Measure | How to count it | Why it matters |
|---|---|---|
| Median FRT by channel | First qualifying reply minus the customer's first message, in business hours | The typical wait while you're open |
| 90th-percentile FRT by channel | The same conversations, 90th percentile | The slow tail the median hides |
| Share within target | Conversations answered within target ÷ all conversations that arrived, answered or not | The number to set targets on |
| Calendar-hours FRT | The same, in clock time | What customers experience at night and at weekends |
| Waiting for a first response | Open conversations with no qualifying reply, and the oldest one's age | The backlog that answered-only reports leave out |
| Human FRT after handover | From the AI agent's handover to a human agent's first reply | Queue health when AI answers first |
| Repeat contact | Customers who wrote again about the same issue within 24 hours of the first response | Whether first responses actually helped |
| Replies after the WhatsApp window | First responses on WhatsApp sent more than 24 hours after the customer's last message | Conversations that could only get a pre-approved template, not a normal reply |
Split each measure by request type as well as channel: a "where is my order?" question and a damaged-product claim shouldn't share one target.
What's a good first response time?
It depends on the channel and on what you've told customers to expect. These are the published figures we could date (checked on 28 September 2026):
| Source | What it measured | Finding | Date |
|---|---|---|---|
| Toister Performance Solutions | Email reply expectations of more than 3,200 consumers in the US, UK and Canada | A reply within one hour meets the expectations of 88% of them | Survey, March 2020 |
| Sprout Social Index 2025 | 4,044 consumers in the US, UK, Canada and Australia | About three-quarters expect a response on social media within 24 hours or sooner; 73% say they'll buy from a competitor if a brand doesn't respond on social | Fieldwork September 2024, published January 2025 |
| Gorgias research | First response time of ecommerce brands on Gorgias in the $10M gross merchandise value (GMV) band: the median per brand, then the median across brands | All-industry median of 6.3 hours; 1.6 hours in the fastest vertical and 9.1 hours in the slowest | Published April 2026, data as of March 2026 |
| SuperOffice | Customer service emails sent to 1,000 companies | Average response of 12 hours 10 minutes; 62% didn't respond | Reported January 2018; still widely quoted as current |
Expectations matter as much as the clock. Across three studies, Caruelle, Lervik-Olsen and Gustafsson (Journal of Retailing, 2023) found that waiting less than expected raises satisfaction substantially, while waiting longer than expected lowers it only slightly. The effect had two limits: when the expectation came from the business itself, such as a wait-time estimate, and when the wait was much longer than expected. So the time in your acknowledgement becomes the customer's clock.
We found no dated, public benchmark for WhatsApp first response time, and the phone figure repeated on ranking pages ("three minutes or less") comes without a study date. Set targets from your own data:
- Measure where you are: the median and 90th percentile for each channel and request type, in a normal month and your busiest week.
- Write each target as a share within a time, such as "90% of chats answered within X minutes".
- Keep WhatsApp and Instagram targets well inside the 24-hour windows, overnight and at weekends too.
- Tighten targets by priority: failed deliveries, payment problems and angry customers first.
- Publish only a time you can meet at the 90th percentile.
If FRT is one of your customer service KPIs, put the share within target on the scorecard.
In Flowcall, SLA policies set response and resolution targets for each team or workflow, with timers that follow your business hours and holidays or run around the clock. You see which tickets are at risk of missing their target, the people you choose are alerted before the breach, and breached tickets move up the queue, so a slow first response gets attention while the customer is still waiting, not when the weekly report comes out.
What pushes first response time up
In a consumer brand, slow first responses usually come from these causes:
- Spikes you can see coming. Sales, festive seasons and courier delays bring waves of order questions. Plan cover and automated answers in advance, as for Black Friday customer service.
- Staffing that doesn't match arrival times. If customers message in the evening and at weekends while the team works office hours, calendar FRT grows even when business-hours FRT looks fine. Workforce management starts from hourly volume by channel.
- Routing by channel instead of by issue. A refund question waits in the email queue while the WhatsApp team is quiet. Skill-based routing sends it to the next agent who can answer it.
- Looking things up before replying. If agents must open the store admin, the courier portal and the payment tool first, every first response waits for that.
- Language. A message in a language only one agent reads waits for that agent; multilingual customer support closes that gap.
- Repeated questions and duplicates. WISMO ("where is my order?") questions take agent time the harder cases need, and customers who haven't heard back write again on another channel.
How to improve first response time without gaming it
Several ways of lowering the number don't help the customer at all:
| Tactic | Why the number falls | What the customer gets | Do this instead |
|---|---|---|---|
| Counting auto-acknowledgements | Every conversation gets a "reply" in seconds | The same wait | Exclude them; report time to acknowledgement separately |
| Holding replies ("We're looking into it") | The clock stops | No answer and no time | Count only replies that answer, ask for what's needed, or give a next step with a time |
| Asking for details the customer already gave | The clock stops | They repeat their order number | Find the order from the phone number or email before replying |
| Answering easy conversations first | The median falls | Hard cases wait longer | Work by priority and age; watch the 90th percentile |
| Reporting answered tickets only | The unanswered backlog disappears from the report | Silence | Report conversations still waiting and the oldest one's age |
The same applies to agent performance metrics: ranked on raw FRT, agents learn to pick easy conversations. Compare them on the share answered within target, by request type.
What lowers first response time honestly:
- Remove contacts that don't need an agent. Send delivery updates before customers ask, and answer order-status questions from live courier data. This is where customer service automation earns its place.
- Route by issue and priority, with frustrated customers and failed deliveries moved up the queue.
- Put the order in front of the agent, so the first reply can be an answer rather than a question.
- Use canned responses that insert the customer's details, such as the order's status or the return window.
- Set timers and alerts for each channel's target, and on WhatsApp and Instagram, alert well before the 24-hour window closes.
- Check quality alongside speed. Pair FRT with first contact resolution and customer satisfaction metrics such as CSAT. If FRT falls while repeat contacts rise, the first replies got faster, not better.
Flowcall builds the second and third of these into the AI Helpdesk. Each ticket goes to the right team or agent by skill, language, workload, sentiment or escalation risk, and frustrated customers, or those at risk of escalating, skip the queue. On the ticket, Customer 360 shows the customer's orders, delivery, payments, refunds and past conversations, so the agent can answer in the first reply without opening Shopify or a courier site.
How AI replies change first response time
An AI agent can reply to every message within seconds, day and night. If every AI reply counts, FRT falls close to zero and tells you nothing about your queue. If none count, you ignore conversations the AI actually settled. For AI customer service, the answer is the test you already apply to a human reply.
| What the AI sent in reply to "Where's my order?" | Counts as the first response? | What to measure next |
|---|---|---|
| "Thanks for your message. Our team will get back to you soon." | No. It's an acknowledgement | Human FRT, from the customer's first message |
| "Orders usually arrive in 3–7 days. Track your order here." | Usually not: it doesn't use the customer's order | Repeat contact about the same order |
| "Your order shipped on Tuesday. The courier's last scan was this morning, and delivery is expected on Thursday." | Yes | Repeat contact about the same order |
| For a cancellation request: "Your order is cancelled, and your refund has been started to the original payment method." | Yes, and the request is resolved | Resolution time, and actions a human agent had to reverse |
| "I've passed this to our delivery team with your order details. They'll reply by 11am tomorrow." | Yes, if the team meets that time | Human FRT after handover, and whether the promised time was kept |
We built Flowcall's AI Agent so that its first reply can be one that counts: it checks current order, delivery and payment data in your connected systems and answers from the customer's own order. When your rules allow it, it completes the request in the same conversation, such as cancelling an order or changing the address. When a human agent is needed, it creates a ticket with the conversation, a summary, the order context and what's left to do, and routes it to the right team. Outside business hours, it tells the customer when the team is back.
Once AI answers first, report FRT across all conversations and, for conversations that reach a human agent, the human wait: from the customer's first message and from the handover. A tool that starts its clock at ticket creation may start it at the handover, hiding the time the customer spent with the bot.
Three more checks:
- Fast and wrong is worse than slow. An instant answer with the wrong delivery date creates a second contact. Reviewing what the AI actually said is what AI quality assurance does at scale.
- A draft is not a reply. If an agent approves AI email replies before they go, FRT includes that wait.
- Speed stops being the question. Once AI answers most first messages, resolution time, repeat contact and the human wait after handover carry the information.
How Flowcall handles first response time
Together, these pieces make the first response part of the work rather than a timestamp. The AI Agent answers or completes what your rules allow on WhatsApp, Instagram, Messenger, Live Chat and email, and the AI Helpdesk brings the rest to the right human agent with the order on the ticket and an SLA timer running. Flowcall's reports show each agent's first response time next to resolution, the AI automation rate and CSAT, with CSAT split by AI and human. Flowcall's AI Quality Assurance checks whether the fast replies were right: it flags wrong or incomplete AI answers and late handoffs, and human agents who ask again for details the customer already gave.
To see how first responses, handovers and SLA timers would work on your own channels and order data, book a demo.
Frequently Asked Questions
What is first response time in customer service?
First response time (FRT), also called first reply time, is the time between a customer's first message and the first reply that addresses their request. Measure it per channel, leave out automatic acknowledgements, and report the median rather than only the average.
What is FRT in a BPO or contact center?
In a BPO or contact center, FRT usually refers to email and chat: the time from the customer's message to the first agent reply. For calls, the equivalent is speed of answer, the time a caller waits in the queue before someone answers.
Does an automatic reply count as a first response?
No. An automatic acknowledgement tells the customer their message arrived; it doesn't address the request, which is why Zendesk and Gorgias leave automated and rule-generated messages out of their first response metrics. An AI reply is different when it answers the request or asks for what's needed to act on it.
What is the difference between first response time and average response time?
First response time measures only the wait for the first reply; average (or next) response time measures the wait for every reply. A team can answer the first message quickly and then go quiet, so track both on chat and messaging.
