What an Internal SLA Is (and Why You Want One Even If Nobody Asked)
Somebody has probably told you that "the network is always down." It was not always down. It was down for forty minutes on a Tuesday, and again for ten minutes during a firmware update you announced twice. But you had no number to hold up against the complaint, so the complaint won. That gap — between what happened and what people remember — is the gap an internal SLA fills.
An internal service level agreement (internal SLA) is a short written promise from the IT team to the rest of the organisation. It says what "working" means for each service and how much of the time that will be true. It also says how fast you will react when it is not. Nobody may have asked you for one. That is fine. The first version is mostly for you, and it will be the most useful page you write this year.
By the end of this guide you will know what the three sets of letters mean (SLA, SLO, and SLI). You will know how they fit together and how an internal agreement differs from the one your ISP sells you. You will know what belongs in the document and what does not. And you will have a one-page template you can fill in this afternoon.
In this guide
- Three sets of letters: SLA, SLO, and SLI
- Internal versus vendor SLAs
- What the document buys the IT team
- What it buys the business
- What goes in one
- What stays out
- The one-page internal SLA template
- A short story from Meridian Parts
- Checklist before you publish
- Where to go next
Three sets of letters: SLA, SLO, and SLI
The SLA is the agreement. It is the document itself. It includes the services it covers, the hours it applies to, the targets, and the exclusions. And it includes the promise to report on all of it. One line of an SLA looks like this: "IT will keep the file server available for 99.5% of business hours and will respond to a priority-one ticket within fifteen minutes."
The SLO is the service level objective — the target inside the agreement. It is a number with a unit and a period attached. "99.5% availability during business hours, measured monthly" is an SLO. So is "priority-one tickets acknowledged within fifteen minutes." One SLA usually holds several SLOs, one per promise.
The SLI is the service level indicator — the measurement that proves whether you hit the target. It names the source, not just the number. "The percentage of five-minute service checks against files.example.com that returned success, taken from the monitoring system's status history" is an SLI. Without a named source, a target is a hope with a percent sign on it.
A road analogy holds the three apart. The agreement is the rule that you will obey the speed limit. The objective is the number on the sign. The indicator is the speedometer. It only counts if it is the one in your car, not the one you remember from the last car. In meetings, everyone will say "SLA" for all three. Let them. In the document, do not.
| Term | What it is | One-line example |
|---|---|---|
| SLA (agreement) | The written promise, including scope, hours, targets, exclusions, and reporting | "IT keeps the file server at 99.5% during business hours and answers P1 tickets in 15 minutes" |
| SLO (objective) | One target inside the agreement: a number, a unit, and a period | "99.5% availability, business hours, per calendar month" |
| SLI (indicator) | The measurement, with its source, that shows whether the target was met | "Share of successful 5-minute service checks in the monitoring status history" |
Internal versus vendor SLAs
A vendor SLA is a contract with penalties. Your ISP's agreement promises a percentage, defines an outage in its own favour, and pays a service credit when it misses. Reading one properly is a skill of its own, covered in understanding vendor SLAs. The short version is that the exclusions section is where the promise actually lives.
An internal SLA is a shared expectation with a review. Nobody pays a credit when IT misses. The consequence of a miss is a conversation: what happened, was the target realistic, what changes. That sounds softer than a penalty, and it is, on purpose. A penalty makes people argue about the measurement. A review makes people improve the thing being measured.
The difference in purpose changes the tone of the writing. A vendor's document is written to limit what the vendor owes. Yours is written to be understood by a warehouse manager reading it on a phone. Short sentences, named services, plain numbers. (A vendor SLA's exclusions often run longer than its promises. Yours should not.)
The one thing both kinds share is the need for a measurement. A vendor SLA without your own evidence is just their word; an internal SLA without a monitoring record is a mission statement. Every clause you write needs a check somewhere that produces the number. That is the subject of measuring SLA compliance from monitoring data.
What the document buys the IT team
It buys you a written definition of "good enough." Without one, the target is perfection, because nobody has said otherwise, and perfection is unfunded. Once the document says 99.5% during business hours, an outage of twenty minutes in a month is inside the promise. You can say so, calmly, with the page open.
It buys you a defence against "the network is always down." The reply is no longer "no it isn't," which is an argument. Instead, it is "it was available 99.7% of business hours last month, here is the report," which is a fact. If you have never seen what a percentage like that means in minutes, the nines explained turns each one into a number of minutes you can picture.
It buys you a basis for budget. When someone wants the mail server at 99.9% around the clock, the SLA lets you answer with a list instead of a sigh. In that case, the list is a second server, a second internet link, and a person on call at three in the morning. The promise becomes a line item, and line items get decided by people with budgets. That is a better place for the decision to live than your inbox.
It also buys you a monitoring plan. Every clause names a service. Every service needs a check. And every check needs an alert rule, so the scope of your monitoring writes itself from the document. A good internal SLA gives you a definition, a defence, and a shopping list.
What it buys the business
It buys them predictability. A manager who knows that a priority-two ticket gets a response within an hour can plan around it. A manager who does not know will either escalate everything or assume nothing will happen. And both of those cost you more time than the ticket did.
It buys them a place to complain, with rules. A miss against a written target is a legitimate complaint with a legitimate process behind it. An outage that stayed inside the target is a bad morning, not a failure of IT. Both are easier to talk about when the line between them is on paper. People are far calmer about a rule they agreed to than about a standard that lives in your head.
It buys them a voice in the target. Writing the SLA is the first time many departments are asked which services matter most. The answer is rarely what IT assumed. The warehouse cares about the scanners, not the intranet. Finance cares about the last three days of the month more than the other twenty-seven. Sorting services into levels of importance is covered in criticality tiers. The SLA is where those tiers turn into promises.
What goes in one
An internal SLA has eight parts, and a good one fits on a page. The table below names each part and the question it answers. Everything after the table is about the parts people get wrong.
| Part | Question it answers | Example |
|---|---|---|
| Services covered | Which named things does this promise apply to? | Email, file server, ERP, internet access, Wi-Fi |
| Hours | When does the promise apply? | Mon–Fri 07:00–19:00 local time, excluding public holidays |
| Availability target | How much of those hours will each service work? | 99.5% per calendar month, per service |
| Response and resolution by priority | How fast will IT react, and how fast is a fix likely? | P1: respond 15 min, resolve target 4 h; P2: respond 1 h, resolve target 1 day |
| Exclusions | What does not count against the target? | Announced maintenance windows, ISP outages, power loss to the building |
| How it is measured | Which check, from where, how often, what counts as down? | HTTP check every 5 min from the monitoring server; down after 3 failures |
| How it is reported | Who sees the numbers, how often, in what form? | One-page monthly report to department heads by the 5th |
| Review date | When is the document re-read and revised? | Every six months, or after three misses in a row |
Hours are the clause everyone skips
The hours clause decides whether a night-time outage counts at all. A service promised "during business hours" can be rebooted at midnight without touching the number. A service promised around the clock cannot. Write the hours per service, not once for the whole document, because the warehouse system and the marketing website do not keep the same hours.
Response is a promise; resolution is an estimate
You control how fast a person looks at a ticket, so you can promise it. You do not control how long the fix takes, because the fix depends on what broke, so you can only estimate it. Write the two words differently in the document and people will read them differently. How to pick both numbers for a real team is the subject of choosing what to promise.
Exclusions must be written before the month, not after
An exclusion is a category of downtime the target does not count. Announced maintenance is the usual one. The rules for planning it and the idea of an availability budget are in planned downtime. The rule that keeps exclusions honest is simple: an exclusion added after the outage is an excuse. Write them in advance, keep the list short, and resist adding to it in a bad month.
"Down" needs its own definition
Every availability target depends on what counts as down, and "the server answers a ping" is not it. A web server can answer pings while its application returns errors to every user. Choosing the moment an outage starts and stops is covered in defining down. The SLA should point at that definition rather than repeat it.
The arithmetic behind the target
A percentage only becomes real when you turn it into minutes for your own hours. The calculation is one line. You should do it before you write any target down, because a number that sounds strict can be generous. And one that sounds relaxed can be brutal. Suppose a month of 22 business days at 10 hours each:
minutes_in_period = 22 days × 10 hours × 60 = 13,200 minutes allowed_downtime = 13,200 × (1 − 0.995) = 66 minutes per month
So 99.5% during business hours allows about an hour of trouble a month. The same 99.5% around the clock, on a 30-day month, allows 216 minutes, because the period is bigger. Hours change the meaning of the number more than the number does. The full arithmetic, including how to handle partial months and several services, is in honest availability math.
What stays out
Leave out anything you cannot measure. "IT will provide excellent service" is not a clause, because no monitoring check can return "excellent." If you cannot name the indicator, the sentence belongs in a mission statement, and mission statements do not get reviewed monthly.
Leave out promises about things you do not control, unless they are worded as estimates. Your internet availability cannot be higher than your ISP's, and your cloud email cannot be more reliable than the provider's. You can promise to detect their outage and tell people about it; you cannot promise it will not happen.
Leave out penalties. An internal penalty — a budget clawback, a name on a slide — teaches everyone to argue about whether the outage counted. That argument is slower than the outage. The consequence of a miss is a review, and the review is the point.
Leave out the runbook. The SLA says what you promise; the procedures for delivering it live elsewhere. A document that tries to be both becomes forty pages, and forty-page documents are stored, not read.
Leave out individual names. Write "the on-call engineer" and "the head of finance," not "Priya" and "Tom." People leave. Roles are slower to.
Leave out the maintenance schedule itself. Refer to it: "announced maintenance windows, as published in the IT calendar." Keep the actual calendar where it can change without a document revision. The whole practice of planning windows and suppressing alerts during them has its own pillar at maintenance windows and suppression.
The one-page internal SLA template
Below is the whole document, ready to fill in. Every angle-bracketed item is a blank. Keep the finished version to one page; if it grows past that, you have started writing the runbook.
INTERNAL SERVICE LEVEL AGREEMENT Version: <n> Effective from: <date> Owner (IT): <role> Business sponsor: <role> Review date: <date>, or after three consecutive monthly misses 1. SERVICES COVERED Service Description Tier ------------ ----------------------------- ---- <name> <what users do with it> <1/2/3> <name> <what users do with it> <1/2/3> 2. SERVICE HOURS (per service) Tier 1: <days> <start>-<end> <time zone> Tier 2: <days> <start>-<end> <time zone> Tier 3: best effort, business hours only 3. AVAILABILITY TARGETS (per calendar month, within service hours) Tier 1: <xx.x>% Tier 2: <xx.x>% Tier 3: not measured 4. RESPONSE AND RESOLUTION Priority Definition Response (promise) Resolution (estimate) -------- ------------------------------ ------------------ --------------------- P1 <whole service down for many> <n> minutes <n> hours P2 <degraded, or down for a few> <n> hour(s) <n> business day(s) P3 <single user, workaround exists> <n> business day(s) <n> business days "Response" means a named person has acknowledged and begun work. 5. EXCLUSIONS (do not count against availability) - Maintenance windows announced at least <n> business days ahead - Outages caused by <ISP / power to building / named upstream provider> - Faults on user-owned equipment 6. MEASUREMENT Source: monitoring system status history, <server or probe name> Check type: service check (<HTTP / SMB / SQL / DNS>), not ping only Interval: every <n> minutes Down means: <n> consecutive failed checks Up means: <n> consecutive successful checks Response time: timestamp of ticket acknowledgement or alert acknowledgement Resolution: timestamp of alert clear or ticket closure 7. REPORTING Monthly, by the <n>th, to <audience>, one page: per service: target, achieved, met/missed, worst incident, note 8. REVIEW Both parties re-read this document on the review date. Any month with a miss gets a written miss review within <n> days. Targets change only by agreement of both signatories. Signed (IT): ______________ Signed (business): ______________
Tip: print the finished page and pin it near the monitoring screen. An SLA that lives in a shared folder gets read once. One on the wall gets pointed at, which is the whole reason it exists.
A short story from Meridian Parts
Meridian Parts ran a warehouse, a sales office, and a two-person IT team with no written agreement of any kind. Every complaint arrived with the same urgency, because nothing said otherwise. A slow intranet page and a dead order-picking system were both "IT emergencies." Both got the same panicked response. The team spent an afternoon rebuilding a manager's browser bookmarks while the barcode scanners in dispatch quietly lost their wireless connection for two hours. When the warehouse lead finally rang, the only defence available was "we were busy." He described that afterwards as "not the number I was looking for." The SLA they wrote the following week was one page long and put the scanners in tier one and bookmarks nowhere. Nobody has asked them about bookmarks since.
Checklist before you publish
Run down this list with the draft open. Every item is a yes or a no; a no means the document is not ready, not that it is bad.
- Every service in the document has a name a non-technical reader recognises.
- Every service has its own hours, written with a time zone.
- Every availability target has been converted to minutes per month for its hours. You have looked at that number and agreed with it.
- Response targets are worded as promises; resolution targets are worded as estimates.
- Every priority level has a one-line definition a manager could apply without calling you.
- Every exclusion is a category, not an incident, and the list fits in three lines.
- Every availability clause names its check type, interval, source, and what counts as down.
- A monitoring check exists today for every measured service. If not, the clause waits until it does.
- The reporting clause names an audience, a frequency, and a day of the month.
- The review date is in a calendar, not just in the document.
- The whole thing fits on one page and contains no individual names.
- Someone from the business side has read it and signed it, even if the signature is an email saying "fine."
Where to go next
You now have the vocabulary, the parts, and a page to fill in. The filling is where the judgement lives. The rest of Defining and Tracking Internal SLAs walks through it in order. Start with choosing what to promise so the targets match your team and your hours. Then comes measuring compliance from the monitoring data you already collect. Next come reporting the result without spin and what to do when you miss. If your uptime targets alone need work first, internal uptime targets covers picking a number you can hit. This pillar takes that number and builds the rest of the promise around it. And once the monthly report exists, monitoring data for management is about making it land with the people who read it.
Put this into practice with SysPeep Monitor
An internal SLA is only as strong as the record behind it. The monitoring software behind this site keeps availability reports and status history for every service check in one web console. So the number you write in the document is a number you can show. The free trial installs on a single Windows machine.
Download Free TrialLearn About SysPeep