> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vorel.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# SLA policies

> Set first-response and resolution targets per channel, track breaches, and report on hit rate.

An SLA policy is a commitment your workspace is measured against: how fast a customer gets a first
reply, and how fast their conversation gets resolved.

## Creating a policy

Under **Settings, then SLA**, create a policy with:

| Field                     | Meaning                                                                         |
| ------------------------- | ------------------------------------------------------------------------------- |
| **Name**                  | What this policy covers                                                         |
| **First response target** | Minutes from the conversation starting to the first reply reaching the customer |
| **Resolution target**     | Minutes from the conversation starting to it being resolved                     |
| **Channels**              | Which channels it governs. Leave empty for all of them                          |
| **Enabled**               | Whether the clocks run                                                          |

The two targets are independent. You can set a first-response target without a resolution target, or
the other way round.

Only owners and admins can edit policies. Agents and viewers can see them. An SLA target is a
commitment the workspace is measured against, so it is deliberately a tighter write gate than
personal conveniences like macros.

## Which channels are tracked

WhatsApp, email, web chat, Instagram and web forms.

Voice is not tracked for SLA. A phone call is answered in seconds by definition, and a first-response
clock on a live conversation measures nothing useful.

## What starts and stops a clock

**Start.** A clock starts when an inbound conversation is created on a tracked channel.

**First response satisfied.** When a reply actually reaches the customer. A drafted email that has
not been sent does not count, which matters if you run the [email channel](/channels/email) in draft
mode.

**Resolution satisfied.** When the conversation is resolved or closed.

An agent reply satisfies the clock the same way a human reply does. The customer does not care who
answered, only that someone did.

## Breaches

When a clock expires unsatisfied, the breach is recorded and two things fire:

* An **`sla.breach` notification**, which respects each person's notification preferences and can
  reach Telegram
* An **`sla.breached` webhook**, so your own systems can react

Breaches are recorded, not just alerted, so the report reflects what actually happened rather than
what someone remembered.

## Reporting

**Quality, then SLA** shows, per clock:

* Hit rate, with the change against the previous period
* Median and 90th percentile response time
* Breach count
* A daily hit-rate trend

Use the median for what normal looks like and the 90th percentile for what your worst-served
customers experience. A workspace with a good median and a bad 90th percentile has a queue problem,
not a speed problem.

## Setting targets that mean something

* **Start by measuring.** Run for a few weeks, look at your actual median and 90th percentile, then
  set a target you can hit most of the time.
* **A target you breach constantly is not a target.** It is noise that trains your team to ignore
  the alert.
* **Set different targets per channel.** Nobody expects an email answered as fast as a WhatsApp
  message, and a single policy across both means one of them is wrong.

## Escalation SLA is separate

Escalations have their own response clocks and their own reminder chain, covered in
[Routing and escalation](/inbox/routing). Conversation SLA measures your service to the customer.
Escalation SLA measures your team's response to each other.