Omnichannel SLA: measuring response time by channel

Short answer

To track an omnichannel SLA, first agree on the clock itself: when it starts, when it stops and which reply counts as a reply. A practical model has three measures: first response time, next response time and resolution time. Each is calculated per channel, in business hours, and reported as a percentage rather than an average: "90% of conversations got a reply within 15 minutes". Bot replies and human replies must be measured separately, or the number looks great while the customer is still waiting.

Why one SLA does not fit every channel

Customer expectations differ by channel. Someone in the website chat is sitting in front of the screen and will close the page after a few minutes. Someone on WhatsApp will read a reply an hour later — but after 24 hours the company can no longer answer them in free-form text. A reply to an Instagram comment happens in public, and a delay is visible to everyone.

So a single number — "30 minutes for every message" — is either far too slow for website chat or needlessly strict for email. Each channel needs its own target.

Three clocks: first, next and resolution

1. 1First response timeFrom the customer's first message to the company's first meaningful reply. This is where the customer feels heard.
2. 2Next response timeFrom each customer message mid-conversation to the reply that follows. The most forgotten measure: the first reply comes fast, then the conversation stalls for hours.
3. 3Resolution timeFrom opening the conversation to resolving and closing the request. For a sales question, "resolved" means the next step is agreed.

Contact centres record these moments separately: in AWS's contact record, for example, the moment a contact was initiated, the moment it entered a queue and the moment it connected to an agent are distinct fields. Written channels need the same logic — each moment with its own timestamp.

When the clock starts and stops

• Start: the moment the customer's message reaches the system. If the customer writes three short messages in a row, the clock starts from the first.
• Pause: when the conversation is in "waiting on the customer", the clock stops — the customer's delay should not count as the team's.
• Business hours: if the target is in business hours, the clock stops at 19:00 and resumes at 09:00. Write a separate target for overnight messages: "by 10:00 the next morning".
• Stop: a meaningful reply from the company. An automatic "we received your message" does not count.

Does a bot reply count?

A bot replies in seconds and drives first response time close to zero. That is true, but only half true. If the bot resolved the question, the SLA for that conversation is met. If the bot said "I'm passing you to an agent", the clock must not stop: the customer is still waiting for a person.

The practical answer is two separate measures: "first response" (bot included) and "first human response" (only for conversations handed to an agent). How to improve the bot's own first-reply speed is covered in chatbot first response speed.

Example targets

The figures below are illustrative and can serve as a starting point for your team. Set your real targets from your own baseline: measure for two weeks first, then write the target down.

• Website chat: first human reply within 3 minutes in business hours — 80% of conversations.
• WhatsApp and Instagram DM: first human reply within 15 minutes in business hours — 90%.
• Instagram and Facebook comments: within 2 hours — 90%.
• Overnight messages: by 10:00 the next morning — 100%.
• Next response: 30 minutes while the conversation is open — 80%.

Report percentages, not averages

Average response time misleads. If nine conversations were answered in 2 minutes and one in 6 hours, the average is about 38 minutes — neither good nor bad, yet one customer waited half a day. A percentage plus a list of the longest waits is more honest: "90% answered within 15 minutes; 6 conversations waited over 2 hours — here they are." Reading those six conversations teaches more than anything else.

The weekly report

1. 1SLA by channelTarget and actual percentage side by side for each channel.
2. 2By hourWhich hours break the target — often lunch and shift change.
3. 3The 10 longest waitsOne sentence of cause for each.
4. 4One decisionOne change from the report: a person added to a shift, a new bot answer, a new view.

An illustrative example

This is an illustrative example. A travel agency had set "10 minutes for every message" and saw 95% in its report. When it wrote down the measurement rules, it turned out that bot replies were counted and overnight messages were excluded. Under the new rules, first human response on WhatsApp came out at 61%, and half of the overnight messages were answered after 11:00.

The team did two things: the morning shift started 30 minutes earlier, and overnight messages got their own view and were handled first. The number went down, but for the first time it showed what was really happening.

Common mistakes

• Counting an automatic acknowledgement as a reply.
• Counting the customer's waiting time as the team's delay.
• Leaving overnight and weekend messages out of the report.
• Measuring only the first reply and missing long gaps mid-conversation.
• Using the SLA to punish agents — conversations then close fast, with empty answers.

Limitations

A fast reply is not a good reply. An SLA does not measure quality: a wrong answer in 2 minutes is worse than a right one in 20. So read the SLA together with reopened conversations and customer complaints. Platforms have their own delays too: a message may reach the company a few seconds or minutes late.

Response-time data in Vexvon

In Vexvon, conversations from every written channel are stored in one place: each message with its channel, time and sender — customer, bot or employee. Short messages that arrive in a row are combined into one reply, a conversation can be assigned to an employee and closed, and the working schedule is set by the company. SLA measurement is built from this data; we do not present it as a ready-made SLA dashboard, and we set the targets together with you. The list of channels is on the integrations page.

Next step

This week, write three rules: when the clock starts, when it stops and whether a bot reply counts. Then measure 30 of last week's conversations by hand. The daily inbox work is in the omnichannel inbox, and distributing requests between teams is in customer request routing. We can set your targets together during a demo.