A customer asks for your SLA before they sign, and you realise you have a number you try to hit and nothing to attach to a contract.
An SLA, short for service level agreement, is a contract between a service provider and a customer that defines the service, the performance to expect, how performance is measured, and what happens when it is not met, as IBM describes it. Companies also use the term internally, for an agreement between two teams in one organisation.
The word that carries the weight is contract: an obligation, with a named consequence for missing it. An internal target is a number you aim at and review yourselves. Most teams end up with a target, which is usually right.
The parts of an SLA
SLAs differ by company and product, but they settle the same six questions.
| Part | What it fixes |
|---|---|
| Scope | Which service, for which customers, over which period |
| Response target | How long until someone replies |
| Resolution target | How long until the issue is fixed |
| Hours | When the targets run, in which time zone |
| Exclusions | What does not count against the number |
| Remedies | What happens when a target is missed |
The agreements IBM’s component list describes usually add an overview, a description of services and named contacts. The six above change how the promise behaves.
The parts compound. A four-hour response target with office hours, a long exclusion list and no remedy is a weaker commitment than a one-day target with none of those escape routes. Whether a particular clause commits a particular company is a question for a lawyer. What the document contains is published, and that is what this page covers.
What an SLA attaches to a missed number
The remedies are what separate an agreement from a target. A target missed is a conversation. A target missed under an SLA is a claim.
IBM’s component list records penalties, escalation procedures and compensation, which can be financial or take the form of service credits. Credits sometimes come with an earn-back, a provision that lets the provider recover them by meeting the service level for a defined period afterwards. Wikipedia adds the harder end: the customer’s right to terminate the contract or ask for a refund for losses (Wikipedia, Service-level agreement).
Exclusions matter for the same reason. Where an agreement exempts downtime caused by the customer’s own equipment, by events outside reasonable control, or by scheduled maintenance windows, it names in advance which failures will not count.
An SLA and an internal target
The two are easy to confuse because they use identical sentences. “We reply within one working day” is a target in one company and a contract term in another.
| SLA | Internal target | |
|---|---|---|
| What it is | An agreement with a customer | A number the team holds itself to |
| Who can enforce it | The other party to the agreement | Nobody outside the team |
| What a miss costs | Credits, penalties, or the relationship | A review |
| Who writes it | Whoever signs for the company | The people doing the work |
| Where it lives | A contract, or a published service commitment | A doc, a saved view, a status page |
Support tools have not helped. Several let you label a rule or a report “SLA”, which means the feature watches a clock you set yourselves. That is a target with a contractual name on it.
An internal SLA between two departments is a real agreement with an escalation path behind it. Read the term that way only when there is somebody on the other side of it.
Why a small team should not promise one
The arithmetic is the first reason. A 95% first reply target sounds strong until you run it over a month. On 200 threads, that allows 10 misses before anything has gone wrong, and twelve months of it is a pattern a customer who reads contracts will read.
The second is who carries the obligation. On a team of five, the person replying is often the person fixing the bug.
The third is the clause you write to protect yourself, which is the one the customer argues with. Every exclusion is an admission that something was left out.
The last is exposure you cannot see. Once your number is public and contractual, it is a comparison someone else can put in a table next to theirs.
What to do instead
You can be reliable without becoming a counterparty. A target written down, published and read every week gets a small team most of the benefit.
- Commit in plain language, with no remedy attached. “We reply to every message within one working day” is a promise a customer can hold you to and a sentence you can keep. It just is not a claim.
- Read the oldest unanswered thread every morning. It names the person waiting now, which is what a percentage hides. What first response time is covers how that number is measured.
- Be consistent about your own hours. If your target runs in office hours, write that in the promise too.
- Revisit it when the team changes. A target set for three people quietly becomes a different target at eight.
If a customer genuinely needs a signed agreement, that is a business decision rather than a support setting. Take it to whoever signs contracts, and treat the number as the smallest part of it.
Where Chirp fits
A promise with no owner is not a promise. On a shared address the question an SLA cannot answer is who is answering this, and what that customer has already been told.
Chirp is a shared inbox for small software teams that track their work in Linear. Press i in a thread to link
it to an issue, create one, or search for the one that already exists. Snooze until the issue is done, and the thread
comes back with a reply drafted for you that names the issue.
The Issues view (g i) sorts linked issues by how many customers are waiting, which is closer to the reading that
matters than any percentage. Rules in Chirp can close, tag, assign and snooze, and write drafts; none has a send
action, so no acknowledgement leaves on its own.
Chirp connects to Gmail, Google Workspace or IMAP and sets access per inbox, with the roles Admin, Agent and Viewer. Which reporting model your team needs is the subject of shared inbox vs help desk.
FAQ
What are the three types of SLA? IBM groups them as customer-level, service-level and multilevel. A customer-level SLA covers one customer group across the services it uses. A service-level SLA covers one service for every customer. A multilevel SLA splits the terms into layers.
What does SLA mean in a job description? Usually an internal target. A support job ad that says you will “work to SLAs” means the team has reply and resolution goals it measures itself against.
What is the difference between an SLA and a KPI? The agreement states the level you have agreed to provide, and the KPI measures whether you are providing it. A KPI obliges nothing on its own.
Does “SLA” mean the same thing in healthcare and finance? The acronym is shared and the document is not. It stands for service level agreement, and what sits inside one differs by industry.
Next step
If you have decided what to commit to, the next question is how the number gets measured and how it gets gamed. What first response time is defines the clock and what a small team should do with it.
