AHT, SLA and First Response Time: How to Calculate and Set Targets
Formulas for average handling time (AHT), SLA and first response time, with illustrative examples, common measurement mistakes and a method to set targets by type of operation.

Marlos Carmo
Founder of Tolky
Updated on November 16, 2026

Share
Three acronyms show up in almost every customer service meeting, and it is common for each person in the room to calculate one of them differently. The result is a dashboard that shows 12 minutes to one manager and 40 to another, for the same operation in the same month.
The problem is rarely the tool. Nobody wrote down where the clock starts, what counts as time, or whether the average is even the right statistic. Without those decisions, any target becomes a matter of opinion.
This guide fixes the definitions, shows the formulas and explains how to set targets without making up a reference. All numbers in the examples are illustrative: they show the math, not a comparison with the market.
Updated in November 2026.
What is AHT and how do you calculate it?
AHT (average handling time) is the average time an operation takes to complete a conversation, from its start to its resolution. The formula is:
AHT = total time of all completed conversations ÷ number of completed conversations
Illustrative example: in one day, 200 conversations were completed and added up to 3,000 minutes. AHT is 3,000 ÷ 200 = 15 minutes.
Two decisions change the result and must be written down:
- Where the clock starts. At the customer's first message, when they enter the queue, or when a person takes over? The first option includes waiting time; the last measures only the agent's work.
- What counts as time. On asynchronous channels such as WhatsApp, customers disappear for hours and come back. If the clock runs during the silence, AHT shows the customer's routine, not the team's efficiency.
Many operations measure active time (between messages) instead of elapsed time, and record the rule in their metrics handbook. What matters is consistency: changing the rule mid-year breaks the historical series.
What is first response time?
First response time is the interval between the customer's first message and the operation's first reply. It is the metric customers feel most, because it decides whether they believe they were heard.
First response = time of first reply − time of the customer's first message
Three measurement precautions:
- An automatic acknowledgment is not a response. If a "we received your message" resets the clock, the number looks great and the experience does not change. Decide whether only a reply with content counts.
- Separate business hours from off hours. A message at 11:00 PM answered at 8:00 AM is not the same kind of failure as a delay at 2:00 PM.
- Use the median and percentiles, not just the average. See the example below.
Illustrative example: six first responses, in minutes: 1, 1, 1, 2, 2, 60. The average is 11.2; the median is 1.5. The average describes a customer who did not exist: five were served within 2 minutes and one waited an hour. The median shows the routine, and the 90th percentile and the SLA show the case that generates complaints.
What is a customer service SLA and how do you measure it?
An SLA (service level agreement) is the time or quality commitment an operation makes, such as replying within 5 minutes or resolving within 24 hours. What you track is not the deadline itself, but the percentage of conversations that met it:
SLA compliance (%) = conversations within the deadline ÷ total conversations in the period × 100
Illustrative example: the SLA is "first response within 5 minutes". Of the day's 200 conversations, 180 were answered within that deadline. Compliance is 180 ÷ 200 = 90%.
A well-written SLA has five parts: which metric, which deadline, during which hours, for which customers or priorities, and what compliance target. "Reply fast" is not an SLA. "90% of first responses within 5 minutes, during business hours, for priority customers" is.
Note the other use of the term: in technology contracts, SLA also describes service availability. This guide covers the customer service SLA.
How do the three numbers relate to the queue?
The service queue is where waiting time accumulates, which is why it explains much of the variation in first response. If the team is sized for average volume and a spike arrives, the queue grows, first response worsens and the SLA drops, even with a stable AHT.
A useful dashboard reads the numbers together, as in the table below.
| Symptom | Likely cause | Where to look |
|---|---|---|
| High first response, normal AHT | Long queue or too few people at that hour | Queue size and wait time by hour |
| High AHT, normal first response | Complex cases or missing context | Contact reason and transfers |
| Low AHT, high reopen rate | Premature closing | Reopen rate |
| SLA drops only at certain hours | Poorly distributed schedule | Compliance by time band |
| Good average, customers complaining | Long tail hidden by the average | 90th percentile and cases past the deadline |
For the cost of waiting on sales and retention, see the invisible cost of slow customer service and how to avoid losing leads on WhatsApp.
Common mistakes when calculating these metrics
The most frequent mistake is comparing numbers calculated under different rules. Before blaming the team, check whether the dashboard measures what you think it measures.
Mixing channels and hours in the same average
Webchat during business hours and WhatsApp at night behave differently. A single average hides the fact that one of them is outside the deadline. Separate by channel, time band and priority.
Counting reopened conversations as new ones
If the customer returns the next day and the platform opens another conversation, the first looks quickly resolved and the second looks like a new case. AHT improves on paper and the experience gets worse. Track reopens and, where possible, repeat contact about the same issue.
Measuring by agent without looking at case type
Whoever receives the most complex cases will have a higher AHT, and that is not poor performance. Compare within the same type of issue before drawing conclusions.
Using the target as punishment
When the target becomes individual pressure, the team learns to close early and send empty messages. Use the metrics to size schedules and fix process, alongside a satisfaction measure such as NPS.
How does AI change these metrics?
AI agents tend to cut first response time to seconds, but that does not guarantee a lower AHT or a better SLA. What AI changes is the mix: part of the conversations is resolved without a person, and the human queue receives only the rest, usually more complex.
So when measuring an operation with AI, separate the groups:
- Conversations resolved by AI alone: their own first response and AHT, usually short.
- Transferred conversations: the time until transfer and the time after it. If the handoff to a human loses context, human AHT goes up.
- Total view: useful to understand the customer experience, but mixing everything can hide that human AHT rose because the easy cases left the queue.
For the resolution rate itself, see how to measure AI resolution rate without fooling yourself. For the concept of deflection, see ticket deflection with AI.
How do you set targets by type of operation?
An operation's target should start from its own baseline and from what customers expect in that context, not from a market number. We do not cite benchmarks here because there is no reliable public reference that holds for every topic, channel and hour. The method below works for any operation:
- Measure 30 days without changing anything. Record the median and 90th percentile of first response, active AHT and reopen rate.
- Ask customers what matters. For an urgent order, minutes count; for a registration question, hours are enough.
- Set targets as percentiles, not averages. "90% within X" protects the long tail.
- Change one target at a time and re-read the other two metrics afterward.
- Review every quarter. A target that never fails is probably too loose.
The table shows which metric to prioritize by type of operation. It has no target values: they come out of step 1.
| Type of operation | Main metric | Common trap |
|---|---|---|
| Support and customer service | First response and reopens | Cutting AHT by closing early |
| Sales and lead qualification | First response, because leads cool down | Replying fast without qualifying |
| Collections and relationship | Customer reply and resolution time | Measuring only sends, not replies |
| Internal support, for example on Microsoft Teams | Resolution time by priority | Treating every ticket the same |
| After-sales and retention | First-contact resolution | Successive transfers |
In operations that mix AI and a human team, set separate targets for each layer. Also worth reading: automating customer service without losing humanity and the guide to AI governance in customer service.
Where Tolky fits in
Tolky is a conversational AI CRM where AI agents and the human team work in the same environment, with each conversation's history recorded. That history is the raw material for the metrics in this guide: the times of the first message, first reply, transfer and closing live in the conversation itself, across the active channels (WhatsApp, webchat, Microsoft Teams and voice).
The platform does not set the target for you. Measurement rules and SLA deadlines are your operation's decision, and it is worth confirming with the team which reports fit your case before planning the dashboard. The platform's public figure of 60% less response time is a platform metric and not a promise of results for your company.
Frequently asked questions
What is the difference between AHT and first response time?
AHT measures the duration of the entire conversation, until resolution. First response time measures only the wait until the first reply. An operation can have excellent first response and high AHT if it replies fast but takes long to resolve.
What is a good AHT?
There is no single number valid for everyone. It depends on the topic, the channel and the complexity of cases. The path is to measure your baseline, compare it with what customers expect and track the trend together with the reopen rate.
Is the SLA calculated using the average?
No. The SLA is calculated as the percentage of conversations that met the deadline. The average hides cases far beyond the limit, which are exactly the ones that generate complaints.
Should AHT on WhatsApp count the time the customer does not reply?
It depends on the rule you set, as long as it is written down and consistent. Many operations measure active time between messages so as not to confuse the customer's routine with the team's efficiency.
Does AI improve the SLA?
It can improve first response compliance, because it replies in seconds and relieves the human queue. But the effect on AHT and resolution depends on how much AI actually resolves and on how the transfer to the team happens. Measure each layer separately.
How often should I review targets?
Once a quarter is a good rhythm, or sooner if volume, channel or case type changes. A target that is never missed tends to be too loose, and one that always fails tends to be unrealistic.
Conclusion
AHT, SLA and first response only become management when the calculation rules are written down, the reading uses median and percentile, and targets start from your own baseline. Begin by measuring 30 days without changing anything and adjust one target at a time.
Explore the WhatsApp solution, see the plans or talk to the team →. Let's talk?
Share
Tags
- how to calculate AHT
- average handling time
- customer service SLA
- first response time
- support targets
- WhatsApp support metrics
- customer service KPIs
- #conversational CRM
Cited in
- How to Choose an AI Agent for WhatsApp: A 12-Question Checklist
- AI to human handoff: 7 rules to keep the context
- Tolky platform conversations report 2026: metrics and definitions
- AI resolution rate: how to measure it without fooling yourself
- Voice AI and Semantic IVR: The End of Press 1 in Customer Service
About the author

Marlos Carmo
Founder of Tolky
Marlos Carmo is an AI entrepreneur and founder of Tolky, the conversational-era infrastructure and AI CRM that unifies intelligent service, multi-channel support (such as WhatsApp and voice), live CRM, and operational intelligence in a single ecosystem. He is a finalist for the SXSW Innovation Awards and a member of Francesco's Economy, a global network of young entrepreneurs focused on innovation and social impact. He works connecting Artificial Intelligence and digital transformation in projects for large organizations.
Next step
Want to bring this to your operation?
Tati, Tolky's AI assistant, understands your scenario in a few questions, suggests the best path and connects you with someone from the team, with full context.


