When You Miss the SLA: Reviews, Renegotiation, and Not Panicking
The month will come when the report shows a red row, and the first feeling will be that something has gone badly wrong. Something has gone wrong, but not badly, and not in the way it feels. A target was written down, and a service fell short of it. Now there is a number on a page instead of a rumour in a corridor. That is the agreement working exactly as designed.
A miss — a month in which a service did not meet the target in the internal service level agreement — is information. It tells you either that the service needs work or that the target was wrong, and it tells you which one if you look properly. What it is not is a crime, a performance issue, or a reason to reword the exclusions. The teams that treat it as one of those stop learning anything from their own reports within a year.
By the end of this guide you will have a short review procedure that fits in half an hour. You will have a way to tell a bad month from a broken promise. You will have a plain-language grip on the error budget and how to spend it on purpose. You will have a method for proposing a new target without it sounding like surrender, and a template for the review itself.
In this guide
- A miss is information, not a crime
- The miss review, in five steps
- One bad month or a broken promise?
- The error budget in plain terms
- When to renegotiate, and how to propose it
- What to say in the miss month
- The miss review template
- A short story from Acme Fasteners
- Checklist after a miss
- Where to go next
A miss is information, not a crime
A target that is never missed is either too easy or not being measured honestly. Both are worse than a miss. A target set with any ambition at all will be missed occasionally, because outages are lumpy: a quiet year, then a Tuesday with a failed controller. The agreement was signed knowing that; its review clause exists for exactly this month.
The wrong reactions are all reactions to shame. Rewriting the exclusions so the month passes. Rounding 98.96% to 99.0%. Quietly dropping the service from the report. Each of these fixes the number and breaks the agreement, because the agreement's only value is that its numbers are believed. Once the reader suspects a miss can be made to disappear, every green row is a suspect too.
The right reaction is a short, dull procedure. Confirm the number, find out why, decide whether the target or the delivery was at fault, choose one change, write it down, move on. Half an hour, one page, no meeting with more than three people in it. The rest of this guide is that procedure and the judgement calls inside it.
The miss review, in five steps
A miss review is not an incident post-mortem. The post-mortem asks why a particular outage happened and how to stop it recurring. It goes deep: a timeline, contributing causes, actions with owners. The rules for doing that well, including the part where the actions actually get done, are in action items that happen. The miss review sits one level up. It asks why the month, as a whole, fell short of a promise, and whether the promise still makes sense.
The two documents point at each other. A miss review usually references one or two post-mortems, because most misses are one big outage. But a miss can also be twenty small outages with no single post-mortem. Or it can be one outage that was reviewed perfectly and simply cost more minutes than the target allowed. The miss review is where those cases get looked at, and it takes five steps.
- Step 1: Confirm the number. Re-run the month from the status history and the exclusion log. Check the service hours were applied, the exclusions were pre-announced, and the outage minutes are clipped correctly. A miss by two minutes deserves the same check as a miss by two hours. Either the figure is right or the specification has a gap, and both are worth knowing.
- Step 2: List what consumed the minutes. Every outage in the month, with its duration, in one table. Usually one row dominates. Occasionally none does, which is a different and more interesting problem.
- Step 3: Ask which was wrong, the target or the delivery. If the biggest outage was preventable and the post-mortem says how, the delivery was wrong. Suppose the outage was a normal failure of the architecture as it stands (a single server had to be rebooted; the only link went down). In that case, the target was wrong for that architecture. Write one sentence saying which.
- Step 4: Choose one change. Either an action on the service, taken from the post-mortem, or a proposal to change the target, taken to the business sponsor. Not both in the same month, and not a list of nine.
- Step 5: Write it up and file it. One page, using the template at the end of this guide, stored with the month's compliance worksheet. It becomes the evidence when the pattern question comes up.
Step three is the one people skip, because it is uncomfortable to write "the target was wrong" about a number you proposed. Skip it anyway and you get the other outcome. A team works hard to hit a target the architecture cannot deliver, month after month, describing each miss as bad luck. Bad luck does not have a trend line. Architecture does.
One bad month or a broken promise?
One miss is a month. Three misses in six is a pattern, and the difference matters because the two need different responses. A single bad month with a clear cause and an action gets a review, a paragraph in the report, and nothing else. A pattern means the promise itself is in question and someone should say so out loud, preferably you, preferably before the sponsor does.
The trend table in the monthly report is where the pattern shows first. If you have been reporting the count of months met, the reader has already seen the pattern. The table below gives a working rule for each shape. (The rule of three is not a law of nature. It is the point at which "coincidence" stops being a credible word.)
| Pattern over six months | Likely meaning | Response |
|---|---|---|
| One miss, single large outage, cause known | A bad month | Miss review; action from the post-mortem; no target change |
| One miss, many small outages, no single cause | Something underneath is degrading (a link, a disk, a flapping dependency) | Miss review; investigate the flapping; watch next month closely |
| Two misses, different causes | Possibly a bad half-year; possibly a target with no margin | Compare achieved against target: if every met month was within a few minutes of missing, the target has no room |
| Three or more misses | A broken promise: the target exceeds what the architecture, team, or vendors can deliver | Propose a new target, or a funded change to the architecture, at the next review |
| Met every month by a wide margin | The target is too easy, or the measurement is too kind | Check the vantage point and the exclusions; consider raising the target at the review |
The last row is the one nobody asks about, and it deserves the same attention. Suppose a service beats 99.0% by a full point every month. It has either been over-promised the wrong way or is being measured from a very comfortable place. Either way, the target is not telling anyone anything.
The error budget in plain terms
An error budget is the number of minutes the target allows you to lose in the period. It is just the target turned inside out: a 99.5% promise across 13,200 business-hour minutes is a permission slip for 66 minutes of downtime. Every outage spends some of it. At the end of the month, if you have spent less than 66, you met the target; if more, you missed. Nothing about that is new; the word "budget" only changes how you think about the minutes.
budget_minutes = measured_minutes × (1 − target)
= 13,200 × (1 − 0.995) = 66 minutes
spent_so_far = downtime minutes this month, inside service hours,
not inside an exclusion
remaining = budget_minutes − spent_so_far
Example on the 20th: spent 48, remaining 18.
A 15-minute patch reboot with no announced window would fit.
A 25-minute one would not; announce it or defer it.
The useful part is spending it on purpose. Change causes outages: patches, upgrades, moving a service to a new host. If the budget has room left in the last week of the month, that is the week to make a risky change. In that case, a short outage will not breach the promise. If the budget is nearly spent, defer the change or put it inside a properly announced maintenance window so it is excluded rather than counted. Planning windows against an availability budget is the subject of planned downtime. The error budget is the same idea, tracked through the month instead of planned at the start.
A budget that is never spent is also information. Suppose a tier-one service ends every month with sixty of its sixty-six minutes unused. Either the target is too loose or the team is deferring all change to avoid touching the service. Deferred change accumulates into the outage that spends a year's budget in a morning. Spend some of it. It was allowed for. A budget with nothing spent from it is not prudence; it is a queue.
When to renegotiate, and how to propose it
A target should be renegotiated when the thing it was derived from has changed. Three changes qualify. The architecture changed: a second server was added, or the only one was moved to a cloud host with a different failure pattern. The staffing changed: a person left, or the team gained an on-call rota. The business changed its hours: the warehouse now ships on Saturdays, or the office stopped. Any of those moves a layer under the promise, and the promise has to move with it.
A pattern of misses with no change underneath also qualifies, and it is the harder conversation, because it means the original number was wrong. Have it anyway. The rule from choosing what to promise still applies. The honest target is a little under what the service has actually achieved. Six months of reports now tell you exactly what that is. If the uptime figure alone is what needs resetting, internal uptime targets covers picking a percentage you can hit.
The proposal is one page and always contains the same five things. It contains the current target and what was achieved in each of the last six months. It contains what changed (or the pattern that shows nothing did). It contains the proposed new target and what it would cost to keep the old one. That last item is the one that makes the conversation a decision rather than a confession. "We can hold 99.5% with a second server and a failover test each quarter, or we can promise 99.0% on the current one." That gives the sponsor two honest options and a budget to attach to one of them.
Do not lower a target in the same month as a miss. It looks like what it is, which is a reaction, and a target that moves whenever it is missed is not a target. Propose it at the scheduled review, with the six-month table, having said in the miss month that you intended to. A target renegotiated on schedule is a plan. One renegotiated in a panic is a rumour with a spreadsheet.
What to say in the miss month
The monthly report carries the miss, in the plain three-part form — what happened, what it cost, what changes. That form is laid out in SLA reporting without spin. That paragraph is most of what management needs. What it needs on top, in the miss month specifically, is one sentence about the pattern and one about intent.
The pattern sentence puts the month in proportion. It might say "this is the first miss for this service in six months". Or it might say "this is the third miss in six months, and the review concludes the target exceeds what a single server can deliver." The intent sentence says what happens next and when. It might say "we will propose a revised target at the scheduled review on the fifteenth". Or it might say "the action from the post-mortem is due by the end of next month and we expect the service to meet its target from then." Two sentences. A manager reading them knows whether to worry and when to ask again.
Avoid three things in that paragraph. Do not apologise, because the agreement anticipated misses and an apology implies it did not. Do not list what went right, because the table already shows it. Do not use the word "unfortunately," which has never once shortened an outage. The tone you want is a colleague reporting weather: it happened, this is the damage, here is the plan. How to keep that tone across every number you put in front of management is the whole of monitoring data for management.
Tip: the moment the miss is confirmed, tell the sponsor by a short message before the report goes out. "The file server missed this month; the report on the fifth will have the detail." Nobody enjoys a red row, but everyone prefers one they were warned about, and the warning costs you one line.
The miss review template
One page, filled in within a few days of the report, stored alongside that month's compliance worksheet. Every bracketed item is a blank; the "target or delivery" line is one word and is not optional.
SLA MISS REVIEW Service: <name> Month: <month> Target: <xx.x%> during <hours> Achieved: <xx.xx%> Budget (minutes): <n> Spent: <n> Over by: <n> Reviewed by: <role> Date: <date> 1. NUMBER CONFIRMED Source re-run: yes / no Service hours applied: yes / no Exclusions pre-announced: yes / no Flap rule applied: yes / no Correction to figure: none / <describe> 2. WHAT CONSUMED THE MINUTES Date Start End Minutes Cause (one line) Post-mortem ref ---- ----- ----- ------- --------------------- --------------- <..> <..> <..> <n> <..> <id or none> <..> <..> <..> <n> <..> <id or none> 3. TARGET OR DELIVERY Which was wrong: target / delivery Reason (two sentences): <..> 4. PATTERN Misses in last 6 months: <n> of 6 Months met within 10% of budget: <n> Classification: bad month / degrading / no margin / broken promise 5. ONE CHANGE Action or proposal: <one line> Owner: <role> Due: <date> If a target change: proposed <xx.x%>, to be tabled at review on <date> 6. REPORT WORDING (copied into the monthly report) Pattern sentence: <..> Intent sentence: <..> Filed with: compliance worksheet <month>
A short story from Acme Fasteners
Acme Fasteners promised 99.9% around the clock on its order system. The system ran on one virtual machine on one host with one power feed. The promise came from a number in a slide about a different company. The service missed in nine of the first twelve months. Each miss review found a different cause — a patch reboot, a host failure, a UPS test — and recommended a different action. None of those actions changed the fact that one machine cannot deliver four minutes a month. The IT lead described the year afterwards as "a long time to notice that the target was the problem." At the next review he brought the twelve-month table and two options. They were 99.5% on the current host, or 99.9% with a second host and a funded failover. The finance director chose the first in about a minute. The service has met it every month since. The slide has not been seen again.
Checklist after a miss
Work through this in the week after the report goes out. It is short on purpose; a miss should cost an hour of process, not a week.
- The figure has been re-run from the status history, with service hours and pre-announced exclusions applied.
- No exclusion has been added, reworded, or reinterpreted since the month began.
- Every outage in the month is listed with its minutes and, where one exists, its post-mortem reference.
- The review states, in one word, whether the target or the delivery was wrong, and gives a reason.
- The six-month pattern has been classified using the table above.
- Exactly one change has been chosen, with an owner and a date.
- If the change is a new target, it is scheduled for the next review, not applied this month.
- The sponsor was told about the miss before the report arrived.
- The monthly report carries the three-part paragraph plus a pattern sentence and an intent sentence, and no apology.
- The remaining error budget for next month has been noted, and any risky change planned around it.
- The review is filed with the month's compliance worksheet.
Where to go next
Handled this way, a miss is the most useful month of the year. It is the one that tests whether the number, the measurement, and the conversation all hold up under weight. The number came from measuring SLA compliance from monitoring data. The promise it was measured against is described in what an internal SLA is. When the review recommends a change to the target, the method for picking a defensible one is back in choosing what to promise. If the outage behind the miss needs its own timeline first, incident timeline reconstruction covers building one from the monitoring record. All five guides are collected in Defining and Tracking Internal SLAs.
Put this into practice with SysPeep Monitor
A miss review starts by re-running the month, which needs a status history you can trust. The monitoring software behind this site keeps availability reports and state history for every service check in one web console. It has dependencies and alert rules that keep one upstream failure from looking like twenty. So the review is about the outage rather than the arithmetic. The free trial installs on a single Windows machine.
Download Free TrialLearn About SysPeep