Home › Topics › Monitoring Your ISP & Holding Them Accountable › Reading Your SLA

Reading Your SLA: What Your Provider Actually Promised


Somewhere in the paperwork for your internet circuit there is a document almost nobody reads until they are angry. It is the SLA, or service level agreement. It is the provider's formal answer to "what exactly am I paying for?". Most people read it for the first time in the middle of an outage, hoping it says something it does not. I did. It did not.


Read calmly, on a good day, it is one of the most useful documents you own. It tells you which promises exist, which numbers were never promised at all, and what happens when a promise breaks. This guide walks through the anatomy in plain language. It covers availability targets and what the nines really allow. It covers performance commitments and the crucial question of where they are measured. It covers repair-time promises and how credits typically work. Everything is described generically, because the only SLA that binds anyone is the one attached to your contract. By the end you will be able to read yours in twenty minutes and map every promise to a check you can actually run. You will be able to walk into the next contract conversation with better questions than "is it fast?"


One framing note. An SLA is a commercial document, and this is an engineer's reading of one, not legal advice. Your job is to understand the promises and measure against them, which is covered here and in measuring your circuit yourself. What a given clause is worth in a dispute is a question for whoever handles your contracts.


In this guide

  • What an SLA is — and what it is not
  • Availability targets: what the nines actually allow
  • Latency, loss, and jitter: where they measure
  • Repair promises: response vs. restore
  • Credits and remedies, generically
  • Marketing numbers vs. contract numbers
  • Mapping each promise to a check you can run
  • Questions to ask before the next contract

What an SLA is — and what it is not


A service level agreement is a set of measurable targets the provider commits to, plus defined remedies when a target is missed. That is the whole machine: targets, measurements, remedies. Availability is the headline target; latency, loss, jitter, and repair times are the supporting cast. The remedy is almost always a service credit — a slice of your monthly fee returned, not compensation for what the outage cost your business.


Finding yours is the first small quest. The promises rarely live in the signed cover agreement itself; they hide in attachments. One is a service schedule describing your specific circuit and its grade. Another is an SLA document the schedule points to, sometimes hosted on the provider's site and revised from time to time. Ask your account contact, in writing, for the exact SLA version that applies to your circuit, and file the copy you receive with the contract. "The one on the website" can change; the one in your files, with a date on it, cannot.


Now the "is nots," which prevent most of the disappointment. An SLA is not a guarantee the service will work — it is a price list for the ways it may fail. It is not automatic. In many agreements, you must notice the miss, report it within a claim window, and ask for the remedy. That is why this pillar exists. And it is not measured by your experience. It is measured by the provider's definitions, on the provider's instruments, unless your own records give you standing to argue. The cynical summary is that an SLA is a bet the provider expects to win. The practical summary is better: it is a list of exactly which numbers the provider considers themselves accountable for. That list is where accountability conversations get traction, because you are no longer arguing about feelings — you are arguing about their own document.


Availability targets: what the nines actually allow


Availability is promised in nines: a percentage of time the service is supposed to be up, usually per calendar month. The arithmetic is where intuition fails. A month holds roughly forty-three thousand minutes. So 99.9% — "three nines," a common business-circuit target — still allows about forty-three minutes of downtime in a month with no promise broken. A 99.5% target allows several hours. Two circuits with targets one notch apart live in genuinely different worlds. You can find the full ladder in the nines explained if you want the table. The point here is narrower: convert your target into minutes per month and write that number down. It is the yardstick every outage gets measured against.


Check the accounting period and scope, too. "Per calendar month" means the counter resets on the first. An outage straddling midnight at month-end splits into two smaller misses, each possibly inside its month's allowance. And confirm the target applies to your circuit rather than to the provider's network in aggregate. A network-wide availability figure can be very high while your particular line has a bad quarter. Only a per-circuit commitment obliges anyone to care.


Then read how availability is defined, which matters more than the number. A composite example, of the flavor you will actually meet:


Availability Target: 99.9% per calendar month.
"Unavailability" means a total loss of Service, measured from
the time a trouble ticket is opened by the Customer until the
Provider restores the Service, excluding Scheduled Maintenance,
Force Majeure events, and faults in Customer equipment.

Read it slowly and three doors quietly close. "Total loss": a circuit limping along with heavy packet loss is not "unavailable"; it is degraded, and this clause owes you nothing for it. "From ticket opened": the outage clock starts when you call, not when the circuit died. So twenty minutes of "maybe it will come back" is twenty minutes the scorecard never sees. That is why your monitoring should open the ticket, not your patience. "Excluding Scheduled Maintenance": maintenance windows do not count against the target, so a provider can be down regularly at 3 AM and remain contractually perfect. Fine for most businesses, worth knowing for a shop that runs around the clock. None of these clauses is unusual or underhanded. They are simply the difference between the promise you assumed and the promise that exists.


Ashgrove Housing learned the ticket-clock clause the slow way. Its circuit dropped for twenty or thirty minutes a few times a month. The help desk's habit was to wait for it to come back before calling anyone, because it usually did. Over a year, Ashgrove's own logs recorded a circuit that missed its target in four separate months. The provider's records showed a circuit that had missed it once, because every outage had officially begun when someone finally picked up the phone. At renewal the account manager described the service as "broadly compliant", and Ashgrove had nothing on paper to say otherwise. The monitoring now opens the ticket, and the help desk has been asked to stop being patient.


Latency, loss, and jitter: where they measure


Better-grade circuits often carry performance commitments alongside availability. These include a ceiling on latency (round-trip delay) and packet loss (the fraction of packets that vanish). They sometimes include a ceiling on jitter (the variation in delay that real-time audio and video hate). The numbers look reassuring. Then comes the clause that gets missed on a first reading and read first on every reading after: where those numbers are measured.


Almost universally, performance commitments apply within the provider's own network. That means between their core locations, or from their core to the far side of your access line. Those commitments do not cover the journey from your users to the internet destinations your business actually depends on. Most of that journey crosses networks the provider does not own and sensibly refuses to promise on. This is the classic gap: a circuit can meet every performance commitment it has while your video calls fall apart. The commitment covers one segment of the path and your experience covers all of it. The commitments are also typically averages over a month. A monthly average is a forgiving judge: an hour of trouble every afternoon disappears into the mean.


The practical response is not outrage; it is measurement layered to match. Your own probes toward the provider's gateway watch the segment they promised. Your probes toward external anchors watch the experience nobody promised but everybody feels. When users suffer while the promised segment stays clean, you have learned something real. In that case, the SLA is being kept and the problem lives elsewhere. That distinction saves whole afternoons of arguing with the wrong party.


Repair promises: response vs. restore


When you report a fault, two very different clocks can be promised, and the difference between them is the most common SLA misreading in the business. Response time is how quickly a human acknowledges your ticket and begins work. Restore time (or repair time, time-to-restore) is how quickly the service is actually working again. A commitment that reads "we respond to critical faults within the hour" promises a conversation within the hour — not a working circuit. Many consumer and mid-grade services promise only response. A hard restore commitment is the more expensive, more meaningful promise. It often arrives softened as a "target" or "objective" rather than a credit-backed guarantee. The words matter: a target missed is a disappointment, a guarantee missed is a credit.


Repair promises are also gated by severity classification — total outage versus degraded service versus "informational" — with the strong commitments reserved for the top tier. Read who assigns the severity (usually the provider) and what counts as each level. A circuit at 40% packet loss is unusable in practice but may not be "Severity 1: total loss" on paper. Note also the service hours attached to the promise. A repair commitment that runs "business hours, business days" is a different product from one that runs around the clock. The gap between those two is precisely what the pricier support tier is selling. Your own contribution to every repair clock: timestamps. When you report, log when it broke and when you called; when it is fixed, log when service actually returned versus when the ticket closed. Those pairs of times — kept in your circuit evidence file — are how repair promises get audited by anyone other than the provider.


Credits and remedies, generically


The remedy for a missed target is nearly always a service credit: a portion of that circuit's monthly charge. It is scaled by how badly the target was missed — more downtime, larger slice. It is capped, commonly at some fraction of one month's fee. Three mechanics deserve your attention. First, the claim window: many agreements require you to request the credit within a set number of days of the incident. Miss the window and the credit evaporates regardless of merit. Second, the evidence standard: the provider will check your claim against their records. As the rest of this pillar explains, that is exactly why you keep your own. Third, the exclusivity language: credits are often defined as the remedy for service failures, which is a commercial way of saying "this is the arrangement; plan accordingly."


Suppose a circuit with a 99.9% monthly target — about forty-three allowed minutes — suffers one four-hour outage. The target is missed by a wide margin. So a tiered credit schedule might return some percentage of that month's fee for that band of downtime. A second outage the same month pushes the total into a higher band, up to the cap. Now rerun the example with the definitions from earlier. Suppose your team waited half an hour to open the ticket. The outage is four hours long in your logs and three and a half on the provider's scorecard. That is possibly a whole credit band's difference. The arithmetic is trivial; the definitions move the money.


Be clear-eyed about scale: a credit on a circuit's monthly fee is small money next to what an outage costs a business that depends on it. So why claim at all? Because a claim is a formal, dated admission that the target was missed, on the record, in the provider's own system. One credit is a small line on an invoice. A file of them is a documented pattern of missed commitments. That is exactly the material that powers renewal negotiations and the chronic-problem escalations covered in escalating with data. Claim the credit for the paper trail; the money is a bonus. And look for a chronic-trouble clause while you are in there. Some agreements let you exit without penalty after repeated qualifying failures in a defined period. If yours has one, the evidence file is how you would ever prove you qualify.


Marketing numbers vs. contract numbers


The brochure said "blazing speeds up to" some number, "rock-solid reliability," "guaranteed uptime." The contract's service schedule says what you bought. These are different documents written by different departments for different purposes, and only one of them is enforceable. "Up to" is the load-bearing phrase: an up-to rate is a ceiling, not a floor. On shared access technologies the delivered rate legitimately varies with the neighborhood's evening habits. Dedicated-grade services instead specify a committed rate — the throughput the provider stands behind — and that number lives in the contract, not the ad.


The reading rule: any number that appears only in marketing material effectively does not exist. Find the service schedule — the attachment naming your specific circuit, its committed characteristics, and its SLA tier. Treat it as the single source of promised truth. This is also where the grade of service shows its real difference. Generically: consumer-grade service is best-effort — up-to rates, often asymmetric, support when the queue gets to you, and no meaningful SLA. Business-grade adds some availability commitment, defined support access, and sometimes performance commitments. Dedicated-grade adds committed symmetric rates, the strongest repair targets, and the contractual right to a technician at 3 AM. The monthly price climbs accordingly — and what you are buying at each step is mostly the promises, not the bits. Knowing which promises your organization actually needs is how you avoid paying for nines nobody would notice.


Mapping each promise to a check you can run


An SLA you cannot measure is a bedtime story. The final move of a proper SLA reading is to walk its clauses and attach a check of yours to each one. The provider audits their promises with their instruments, and symmetry demands you bring your own. The standard mapping:


The promiseTypical shapeYour checkEvidence to keep
AvailabilityNines per month, defined exclusionsPer-minute probe through the circuitMonthly uptime rollup; outage timestamps
Latency / loss / jitterMonthly averages inside their networkContinuous probes to gateway + anchorsGraphs at incident time; monthly medians and p95s
Committed rateThroughput the circuit should sustainScheduled off-hours throughput samplesSample trend vs. ordered rate
Response timeAcknowledgment within N minutes/hours by severityTicket log with call and reply timesEvidence-file entries per incident
Restore timeService back within N hours (target or guarantee)Your probes mark actual restorationBreak/restore timestamp pairs vs. ticket close
Maintenance conductAdvance notice, defined windowsCalendar of notices vs. observed outagesNotices filed with the affected nights' graphs

Everything in the third column is a few minutes of setup in ordinary monitoring, and the first article in this pillar builds the whole kit. The fourth column is the habit that turns measurements into leverage; the evidence file guide gives it a template.


Tip: after reading your SLA, write a one-page cheat sheet. Include the availability target in minutes per month, what starts the outage clock, and response and restore commitments by severity. Include the credit claim window and the support phone number that honors it all. Tape it inside the comms room door. During an outage, nobody has ever regretted the tape.


Questions to ask before the next contract


The best time to read an SLA is before signing it, when every answer is still negotiable. Questions that sort providers quickly:


  • 1. Where is availability measured, and what starts the outage clock — the failure itself, or my ticket?
  • 2. Does anything short of total loss — sustained heavy packet loss, say — count as unavailability?
  • 3. Which performance commitments exist, over what segment of the path, averaged over what period?
  • 4. Is the repair commitment response or restore? Per severity? Who assigns severity? What are the service hours?
  • 5. Are credits automatic or claimed? What is the claim window and the cap? Is there a chronic-trouble exit clause?
  • 6. How much advance notice do maintenance windows get, and through what channel?
  • 7. What is the committed rate — not the up-to rate — and is it symmetric?
  • 8. Will you accept my monitoring data as supporting evidence in a dispute?

Ask them in writing and keep the answers with the contract. A provider comfortable with these questions is telling you something; so is one who is not. Question eight is the one to watch: the answer is revealing either way, and a pause before it is an answer too.


Where to go next


You now know what the document actually says: which numbers are promised, where they are measured, what the clocks mean, and what a broken promise pays. The promises only bite when your own records can show a miss. So the companion work is running the checks in the mapping table and filing what they catch. The next time you open the SLA in the middle of an outage, you will already know what it says, which is a better position than hoping. The rest of the program lives in Monitoring Your ISP & Holding Them Accountable: independent measurement, the evidence file, and escalation that leans on all of it.


Put this into practice with SysPeep Monitor

An SLA is a list of promises; monitoring is how you audit them. SysPeep Monitor runs the checks this guide maps to each clause, all from one Windows machine. It provides per-minute availability and latency probes per circuit, dashboards, and reports you can set beside the contract numbers. The free trial is at the download page.

Download Free TrialLearn About SysPeep