Client Reporting and SLAs: Proving the Service
Monitoring has an awkward property as a service: when it works, nothing happens. The client's network stays up. The alerts you handled at two in the morning never reach them. Then, at renewal time, somebody in their finance department asks what, exactly, they have been paying for. The honest answer is "a great deal, quietly." I have watched a good provider lose a renewal because "quietly" was the only part the client could see.
A report is how you say "a great deal" in a form a client can hold. Reporting is not paperwork bolted onto monitoring. It is the part of monitoring the client actually sees, and for most clients it is the only part.
This guide covers the whole loop: what clients genuinely want to know and a monthly report that fits on a page. It covers availability figures with the definition of "down" written down so nobody can argue with them, plus response and restore times per incident. It explains how to talk to a client while something is broken and covers the quarterly review meeting. It also covers the discipline that keeps your report from becoming the one nobody opens. An SLA (service level agreement) runs through all of it. It is the written promise about what is monitored, how fast you respond, and how available things will be. The report is the evidence that the promise was kept.
In this guide
- What clients actually want to see
- The monthly report: a page, not a binder
- SLA evidence: availability with "down" written down
- Response and restore times per incident
- Talking to clients while something is broken
- The quarterly review meeting
- Avoiding the report nobody reads
- The mirror image: your own providers
- Where to go next
What clients actually want to see
Ask a client what they want in a monitoring report and they will say "everything." Watch what they actually read and it comes down to four questions, in this order.
- Was the stuff I care about working? Not "were the 214 monitored objects up," but "was the internet up, could people log in, did the phones work, was the warehouse system available." Availability of services in their words, not devices in yours.
- What broke, and how fast did you deal with it? The incidents, with when they started, when you knew, when you responded, and when things were back.
- Are we about to run out of anything? Disk, bandwidth, licenses, the age of the firewall. The capacity trend that turns into next quarter's budget conversation.
- What did you do for us? The work behind the quiet: alerts handled, changes made, problems caught before anyone noticed. This is the part the finance department is asking about.
Everything in a good report answers one of those four questions. Anything that answers none of them — the CPU graph of a switch that was fine all month — belongs in an appendix nobody is obliged to open, or nowhere.
The monthly report: a page, not a binder
The monthly report is a single page (two at most, with the second being detail). It is sent on the same day every month, in the same format every month. So the client learns where to look. The report's structure follows the four questions, with the headline first. Here is a template that has served well; the rows are the sections, in order.
| Section | What it contains | Source |
|---|---|---|
| Headline | One sentence: "All contracted services met their availability targets; two incidents, both resolved within target." Or the honest opposite. | You write it last, from the rest of the page |
| Service availability | One line per contracted service: name in the client's words, measured availability, target, met or missed. Maintenance windows listed, not hidden. | Availability data from the monitoring platform's history |
| Incidents | One line each: what, when detected, when responded, when restored, cause in a phrase. Link or reference to the post-incident note for anything significant. | Alert history plus your ticket timestamps |
| Capacity and trends | Up to three things approaching a limit, each with the number now, the number three months ago, and when it becomes a problem at the current rate. | Performance history; disk, bandwidth, license counts |
| Work done | Alerts handled (count, with how many were caught before users noticed), changes made, maintenance performed, recommendations closed. | Ticket system and the alert log |
| Recommendations | At most three, each with why, what it costs in effort, and what happens if it waits. | Your judgment |
| Next month | Planned maintenance windows, scheduled changes, anything the client needs to decide. | Your calendar |
The report is generated by the monitoring platform wherever possible and edited by a human always. Automated reports are accurate and unreadable; human reports are readable and late. The combination is what a client will actually read. The platform produces the numbers; a technician writes the headline, the incident phrases, and the recommendations. This takes under an hour per client once the template is stable.
Decide the delivery as deliberately as the content. The report goes to the named contacts in the client's record and nobody else. It goes on the agreed day, in a format they can open without your platform. "Nobody else" is not politeness: a report that leaves the platform is one client's data in the open. The tenant separation guide in this pillar is about what a misdirected one costs. Pair the report with the live site dashboard where the client has one. The report is the month's story, the dashboard is today's status, and a client who has both stops phoning to ask either question.
SLA evidence: availability with "down" written down
An availability figure is only evidence if both sides agree on how it was measured. The number "99.6%" on its own invites argument: measured how, from where, and does the Tuesday maintenance count? The way to make the number unarguable is to write the definition of "down" into the SLA. Then measure exactly that, every month, the same way. The full discipline of building an availability record that stands up to scrutiny is covered in The Uptime Evidence File. That includes what to log, how to keep it, and how to present it. Here is what belongs in the definition.
- Which check decides. "Internet service" is down when the WAN link check from the site's probe fails. "Email" is down when the mail flow check fails, not when one user cannot send. Name the check.
- How many failures make an outage. Three consecutive failed polls at one-minute intervals, say — so a single dropped packet is not an outage and a genuine outage is counted within minutes.
- From where. The probe inside the site, the central platform, an external checkpoint, or some combination. Where you measure from changes what you measure.
- The measurement period and hours. Calendar month; all hours, or business hours only, as the contract says.
- What is excluded, and how. Exclusions include announced maintenance windows, the client's own changes, and events outside the scope of the service. Each exclusion is listed in the report with its times, so "excluded" never means "hidden."
Availability, one service, one month (illustrative)
Period ................ calendar month, all hours = 30 days x 24 h = 720 h
Down (per definition) . 3 consecutive failed WAN checks from the site probe
Outages recorded ...... 07th 09:14-10:02 (48 min) circuit fault, carrier
19th 02:10-02:31 (21 min) router reboot, unplanned
Maintenance excluded .. 12th 22:00-23:00 (60 min) announced 7 days prior
Measured time ......... 720 h - 1 h maintenance = 719 h
Downtime .............. 48 + 21 = 69 min = 1.15 h
Availability .......... (719 - 1.15) / 719 = 99.84%
Target ................ 99.5% -> met
Note .................. the router reboot was inside the maintenance window's
WEEK but not its HOUR; it counts
The last line is the honesty test. Maintenance windows are excluded when they were announced in advance, happened inside the announced hours, and did the announced work. An unplanned reboot on the same day is an outage. A client who catches you sliding an outage into a maintenance exclusion will never trust another number you send. So the report lists each exclusion with its times, and the technician who writes the report checks that each outage sits where it belongs. This is also why a probe that buffers during a WAN outage matters for reporting. It shows the site's internal services stayed up while the circuit was down. That is the difference between "your network failed" and "your carrier (the company that provides the circuit) failed."
When a target is missed, say so in the headline. A missed month hidden on page two is discovered anyway, and then the story is about concealment rather than the outage. State the figure, the target, and the incidents that caused the gap. Add any service credit the contract provides (the refund or discount promised for a missed target). State what has changed so the problem does not repeat. Clients are far more tolerant of a missed target reported plainly than of a met target they cannot verify.
Response and restore times per incident
Availability tells the client how much was lost; response and restore times tell them how you behaved while it was being lost. Both are SLA terms, and both need timestamps rather than impressions. Five moments define an incident, and your ticket should record each one.
| Moment | Definition | Where the time comes from |
|---|---|---|
| Started | When the failure actually began | Monitoring history (often earlier than detection) |
| Detected | When the monitoring alert fired — or, honestly, when the client called if that came first | Alert log, or the ticket's creation time |
| Responded | When a person acknowledged and began work | Alert acknowledgement or ticket assignment |
| Restored | When the service was usable again, even if by workaround | Monitoring shows the check passing |
| Resolved | When the underlying cause was fixed and the ticket closed | Ticket closure |
The SLA clock usually runs from Detected to Responded (response time) and from Detected to Restored (restore time). Two honesty rules. First, when the client told you before your monitoring did, record that. A report that shows "detected by monitoring" for an incident the client phoned in is a report the client knows is wrong. A column for "detected by: monitoring / client / staff" turns an embarrassment into a metric. The fraction caught by monitoring is one of the best numbers you can show improving. Second, "restored" is the moment the check passes, not the moment someone remembers to close the ticket. Let the monitoring history set the time. The wider craft of measuring detection and repair speed is in MTTD and MTTR.
Response targets normally depend on severity: a site-wide outage gets a faster promise than a single printer. The SLA should say which incidents fall into which class. The classes should be the same ones your alerting already uses. That way, the severity on the alert, the severity on the ticket, and the severity in the report are one fact rather than three opinions. The reasoning behind a small, stable set of levels is in Severity Design. Report response and restore times per class, and the client can see at a glance that the serious things were treated seriously.
Talking to clients while something is broken
During an outage, the client's experience of your service is not the fix; it is the communication. A client who hears from you before they notice the problem, and then hears from you again on a predictable rhythm, will forgive a long outage. A client who has to call you to find out what is happening will not forgive a short one. The rules are short, factual, and on time.
- First message quickly, as soon as you know a contracted service is affected, even if all you know is that. "We are aware" beats a perfect message an hour later.
- Say what is affected in their words, what you are doing, and when the next update will come. Then send the next update at that time, even if the update is "no change."
- No speculation, no blame. "The carrier is investigating a circuit fault" is a fact; "the carrier has messed up again" is a fight you are starting on the client's behalf.
- Close it explicitly. A restored message with the time, and a short follow-up within a few days with the cause and what changes.
Incident update template (three messages, same shape)
Subject: [Client code] Internet at Riverside branch — update 2 of N
WHAT Internet access at the Riverside branch is unavailable.
Internal systems (file server, phones between floors) are working.
SINCE 09:14 today, detected by our monitoring at 09:17.
CAUSE Circuit fault reported to the carrier; carrier reference 12345.
DOING Carrier engineer dispatched; we are checking the backup path.
NEXT UPDATE 11:00, or sooner if anything changes.
CONTACT Name, phone.
--- restored message adds ---
RESTORED 10:02. Total impact 48 minutes. Monitoring confirms normal.
FOLLOW-UP A short summary with the confirmed cause will follow by Friday.
The same three messages, copied into the ticket, become the incident line in the monthly report and the timeline in the post-incident note. Write them once, in the right shape, and the reporting takes care of itself.
The quarterly review meeting
The monthly report is a document; the quarterly review is a conversation, and it is where the relationship is actually maintained. Keep it to an hour, with the client's IT lead and whoever owns the budget. And give it a fixed agenda so that both sides know what to bring.
- The quarter in one page. The three monthly headlines side by side, availability per service across the quarter, and the incident count with the fraction caught by monitoring. Five minutes; the point is the trend, not the detail.
- Incidents worth learning from. One or two, walked through honestly: what happened, what changed afterward. Include the one that made you look bad. Clients notice when the review only features your victories.
- The capacity conversation. The trends from the monthly reports, now with a recommendation and a timeframe: "the file server reaches 90% disk in about four months at the current rate; here are two options." This is where monitoring data turns into budget, and it is the most valuable twenty minutes of the meeting for both sides.
- Recommendations, open and closed. What you suggested last quarter, what was done, what is still waiting and why. A recommendation that has waited three quarters is either not important or not being heard; find out which.
- Scope check. What has changed on their side — a new office, a cloud migration, a new application — that monitoring should cover next. This is how the SLA stays current instead of describing a network that no longer exists.
Tip: bring the site dashboard to the review and show it live for two minutes. A client who has seen their own status page — in their own words, with their own sites — understands the service in a way no report achieves. They will often ask for a login. That request is the best possible outcome of a review.
Avoiding the report nobody reads
Every provider has sent a report into silence. The usual cause is not that the client does not care. It is that the report made them work to find the thing they care about. A few habits fix most of it.
Saltmarsh Technology sent every client the same twelve-page monthly report: forty graphs the platform generated, a cover page with the client's logo, and no sentences. It went out on the first of every month for two years. Then a template change left the cover reading "CLIENT LOGO HERE" for four months. Nobody at any of the nineteen clients mentioned it, because nobody had opened the report since the first month. The review recorded "low engagement with the reporting deliverable". The report is now one page with a headline sentence, and three clients have replied to it, which is three more than before.
Lead with their question. The headline sentence goes first, and it answers "was everything fine?" in plain words. If a client reads one line, that is the line. Use their vocabulary. "Internet at Riverside," not "WAN-RVS-01"; "phones," not "SIP trunk registration." The mapping from your device names to their service names is decided once, at onboarding, and used everywhere. Give every graph a sentence. A chart without a "so what" is decoration. "Bandwidth use is up by a third since spring and will need attention before the busy season" is information. Keep the shape constant. Same sections, same order, same day of the month, so the reader's eye learns where things live. Say when nothing happened. A quiet month is a result, not an absence. "No incidents; 61 alerts handled before any user was affected" tells the story of the service better than any outage could. And ask. Once a year, ask the client which parts of the report they read and which they skip, then cut the skipped parts. A shorter report that is read beats a complete one that is filed.
The mirror image: your own providers
Everything in this guide runs in the other direction too. Your client's internet outage was your carrier's circuit fault; your restore time depended on their engineer; your availability figure inherits theirs. The same evidence you assemble to prove your service to a client lets you hold your own providers to their promises. That evidence includes timestamps, a written definition of down, and a per-incident timeline. And the monitoring you already run is the record. The upstream version of this discipline, from measuring a circuit honestly to making a credit claim stick, is the subject of ISP Accountability.
There is a quiet benefit in seeing both directions at once. A provider who has argued with a carrier about whether a forty-minute blip was inside a maintenance window writes clearer maintenance clauses for their own clients. A provider who has waited on a carrier's "next update at eleven" that never came sends their own updates on time. Being a client of someone else is the best training there is for reporting to your own.
Where to go next
Reporting closes the loop that the rest of this pillar opens. The architecture collects the data and the site rollups summarize it. Tenant separation keeps each client's slice private, and the probe fleet keeps it flowing. The report is where all of that becomes something a client can hold, question, and eventually renew. Keep the report to a page and define "down" before you measure it. Timestamp every incident and communicate on a rhythm. And use the quarterly review to turn trends into decisions. Do that, and "a great deal, quietly" becomes a sentence the finance department can read for itself. The other guides in Monitoring Across Multiple Sites and Clients cover each of the upstream pieces. The evidence side of reporting continues in Compliance Evidence.
Put this into practice with SysPeep Monitor
The numbers in this guide come from monitoring history, and the report is only as good as the record behind it. SysPeep Monitor keeps that record in a central Windows service fed by lightweight remote probes at each site. It produces availability and performance reports alongside its live dashboards and network maps from a single web console. The free trial installs in minutes on any Windows machine.
Download Free TrialLearn About SysPeep