What Auditors Actually Want from Your Monitoring
An audit is coming. A customer asked for one, or a contract requires one, or the business has grown into needing one. Either way, the word "monitoring" is on the list of things the auditor will look at. Your first instinct is to polish the dashboards: tidy the network map, fix the colors, make sure nothing is red on the day. I did exactly that before my first audit. It was a pleasant afternoon, and it produced nothing the auditor could use.
The auditor is not coming to admire your monitoring. They are coming to confirm that certain protective activities — called controls — happened, kept happening, and were noticed when they did not. A dashboard proves that a screen exists today. It says nothing about six months ago. What the auditor needs is a record: this check ran, this review happened, this alert fired and somebody acted. The record must show that week after week, across the whole period under review.
By the end of this guide you will know the six control families almost every framework shares. You will know what "operating" looks like for each in a small IT team, and which monitoring artifact proves it. You should be able to read your own audit request list without dread, and notice that you already produce most of what it asks for. You just have not been keeping it.
In this guide
- The mental shift: proof, not pictures
- A screenshot is not evidence
- The control families almost every framework shares
- The auditor's questions, and what answers them
- Periods, populations, and samples
- The small-team truth: you already make this
- Where to go next
The mental shift: proof, not pictures
Three definitions first. A control is any deliberate activity or mechanism that reduces a risk. "We review firewall logs weekly" is a control. "Only two named people can log in to the core switch" is a control. "Backups run nightly and somebody checks that they succeeded" is a control. A framework is a published list of controls that some organization has decided is the sensible minimum, grouped into families. Frameworks differ in numbering and vocabulary; underneath they ask remarkably similar questions.
An audit is a structured check that the controls you claim to have are real. Auditors split that into two questions. The first is design: is the control sensibly built? Would it actually reduce the risk if it worked as described? The second, and the one that causes the trouble, is operating effectiveness: did it actually work, consistently, over the whole period being audited?
Monitoring is unusual because it sits on both sides of that question. It is itself a control: you watch for failures and intrusions. And it is also the machine that produces evidence for other controls: it records who logged in, when the backup finished, when a configuration changed. That double role is why auditors care about it, and why a well-kept monitoring system makes the rest of the audit go faster.
Hold on to one sentence: the auditor wants proof that a control operated consistently over time. Every request traces back to it. When the spreadsheet asks for "evidence that the control operated" (the cell does not elaborate), it is not asking whether you have logs. It is asking you to show that somebody looked at them, on a schedule, for the whole period. Read every row through that lens and a hundred and forty rows collapse into six questions asked in different words.
A screenshot is not evidence
Picture two answers to the same question. The auditor asks, "Do you monitor the availability of your core servers?"
Answer one: you open the console, take a screenshot of the dashboard with all its green ticks, and paste it into a document. The auditor now knows that on the afternoon you took the screenshot, a dashboard existed and everything on it was green. It does not show that monitoring was running in the first month of the period, that alerts went anywhere, or that anybody responded to one. It is a photograph of a moment.
Answer two: you hand over a monthly availability report for each month of the period. The monitoring system generated it on the first of every month. You also hand over the alert history for the same period showing each outage, when it was detected, and who acknowledged it. The auditor now knows the control ran continuously, that it produced a result every month, and that humans were in the loop. That is evidence.
The difference is time and repetition. Auditors have a phrase for the screenshot answer: it proves "existence at a point in time." They will happily accept it for some things — a configuration setting, say, where a dated screenshot is reasonable. But for anything that is supposed to happen repeatedly, they want the record of it happening. A screenshot, however green, does not remember March.
Think of the temperature log on a restaurant fridge. The health inspector does not want to open the fridge and feel that it is cold today. They want the clipboard on the door showing that somebody wrote down the temperature every morning, with initials, all year. The cold fridge is the screenshot. The clipboard is the evidence. Your monitoring system is very good at being the clipboard, if you let it.
The practical rule: for any control that is a repeated activity, evidence has three parts. The parts are the output (the report, the log, the ticket), a date that places it inside the period, and a person who did or reviewed it. Output, date, person. If your artifact has all three for every occurrence, you are finished. If it is missing one, the auditor will ask for it, and you will end up reconstructing it from memory.
Northgate Retail learned this the night before their first audit. The request list wanted evidence of weekly log review across twelve months; the team had reviewed the logs every week and written nothing down. So two administrators stayed late and took fifty-two screenshots of the log console, each scrolled back to a different week. The auditor accepted them as proof that logs existed, and asked, politely, who had looked at them in February. The finding read "review not evidenced", which is the courteous form of "we believe you, and belief is not evidence".
The control families almost every framework shares
Nearly every framework contains the same six families in some form. Each gets a plain-language description, what "operating" looks like in a small team, and the evidence that proves it.
Logging and monitoring: you collect, and you review
What it asks: important systems must generate logs, and those logs must be collected somewhere central and kept for a defined time. A human (or a rule acting on a human's behalf) must look at them regularly for signs of trouble. Collecting is easy; any syslog server does it. Reviewing is the half that gets skipped, and it is the half auditors probe hardest.
Operating looks like: a log collector receives from the firewall, switches, servers, and directory. Alert rules fire on the events you care about. A recurring review — weekly is common in small teams — involves somebody scanning the alert summary and notable events and writing down that they did so. The note is the control. Scanning without the note is a habit, and nobody audits habits.
Evidence: this includes the list of sources sending logs, ideally with a "last received" timestamp proving each was live. It also includes the retention setting, the alert rules, and above all the review records. Those are a dated note, ticket, or sign-off for each review, with the reviewer's name. If your team already holds a regular look at which alerts fired and whether they were useful, the write-up from that is exactly this evidence. The alert review process guide describes how to run one.
Access monitoring: you watch who gets in and what admins do
What it asks: that you know who has privileged access, that logins (especially failed ones and administrative ones) are recorded, and that you would notice misuse.
Operating looks like: authentication logs from the directory, VPN, firewall, and network devices flow to the collector. There are alerts on repeated failures, logins at odd hours, or newly created admin accounts. There is a periodic review of who holds admin rights and a record of access being removed when someone leaves.
Evidence: this includes the authentication log stream (proving it exists and is retained) and the alert rules for suspicious authentication. It includes the alert history showing those rules actually fired and were handled. It also includes the periodic access review with the account list, the reviewer, and what changed as a result.
Change management: you track and review changes
What it asks: that changes to production systems are planned, approved, and recorded, and that an unauthorized change would be detected.
Operating looks like: each planned change has a change record (a ticket, a form, even a row in a spreadsheet), with an approval. Configuration snapshots or config-change alerts from monitoring show what actually changed on devices. There is a reconciliation, where somebody periodically compares detected changes against approved ones and investigates the leftovers.
Evidence: this includes the change log for the period and samples of approved changes with their records. It also includes the monitoring alerts or configuration diffs that detected changes, and a few examples where a detected change was matched to its ticket. That last item is gold: it shows the detection control is real, not decorative.
Availability and continuity: you know when things fail and recover
What it asks: critical services must be monitored, and outages must be detected and responded to. Backups must happen and be restorable, and you must have some idea what to do in a disaster.
Operating looks like: up/down checks on core services, alerts on failure, a record of each outage and its resolution, backup job results checked daily, and a periodic test restore.
Evidence: availability reports per period, the alert history for outages, backup job logs, and the restore test record. The uptime evidence file guide covers how to keep the availability side ready to hand over at any moment.
Incident management: you detect, respond, and learn
What it asks: that you have a defined way to recognize a security or operational incident, respond to it, and afterwards learn something from it.
Operating looks like: there are a short written incident procedure and alerts that feed into it. There is a record for each incident with a timeline and an outcome, and a post-incident review for the significant ones.
Evidence: this includes the procedure document, the incident register for the period, and a couple of incident records with their timelines. The timelines show the monitoring alert that started the incident, who acknowledged it, and when it was resolved. The evidence also includes post-incident notes. Summarizing detection and resolution times per period makes a natural report here, and the reporting incident metrics guide covers how to build one.
Asset inventory: you know what you have
What it asks: you maintain a list of the systems and devices you are responsible for. That list is reasonably accurate. The control also asks that an unknown device would be noticed.
Operating looks like: an inventory (from monitoring discovery, a spreadsheet, or a proper asset system) is reviewed periodically. There are new-device alerts and a routine for retiring devices from the list when they leave. Devices leave more often than the list is told.
Evidence: a dated inventory export per period, the review record, and the new-device alert history. Auto-discovery produces the first item almost for free; the keeping inventory current guide covers the habit behind the other two.
The auditor's questions, and what answers them
Auditors rarely say "please demonstrate operating effectiveness of the logging control family." They ask ordinary-sounding questions, and each maps to an artifact your monitoring can produce. The table collects the common ones.
| What the auditor asks | What they are really checking | The monitoring artifact that answers it |
|---|---|---|
| "Show me your logging." | Sources exist, are centralized, and are retained | Log source list with last-received timestamps; retention settings; a sample export |
| "Who reviews the logs, and how often?" | The review actually happens | Dated review sign-offs or alert review notes, one per occurrence |
| "How would you know if someone logged in to the firewall at 3 AM?" | Detection exists and works | The authentication alert rule, plus an example of it firing and being handled |
| "Give me a list of every change in the period." | A complete population to sample from | Change log export; configuration-change alert history |
| "Pick one outage. Walk me through it." | Detection and response chain | Alert record, acknowledgement, ticket, resolution time |
| "How do you know your backups work?" | Verification, not just execution | Backup job results per day; restore test record |
| "What devices are on your network?" | Inventory accuracy | Dated discovery or inventory export; new-device alert history |
| "Who has admin rights, and when was that last reviewed?" | Access review operates | Admin account list with review sign-off and resulting changes |
| "How long do you keep logs, and could they be altered?" | Retention and integrity | Retention policy; storage description; integrity or checksum report |
| "Show me this happened in the third month too." | Consistency across the whole period | The same artifact for every period, not just the latest one |
The last row separates teams that keep evidence from teams that generate it on demand. If you can produce the same report for every month without effort, the auditor relaxes. If the third month is missing, they start wondering what else is.
Tip: when a request is confusing, ask the auditor what risk the control is meant to address. "We want to know that unauthorized changes would be caught" is far easier to answer than a control number. Auditors are generally glad to explain; a team that understands the risk gives better evidence, and better evidence means a shorter engagement for everyone.
Periods, populations, and samples
Three auditor words explain half the requests you will receive and most of the friction when teams misunderstand them.
Period. The window under review; your engagement letter or scope document states it exactly. Evidence from before or after the period is not evidence for this audit. This is another reason "here is a screenshot from today" fails: today is often outside the period entirely.
Population. The complete list of instances of something during the period — every change, every incident, every new user account, every backup run. Auditors ask for the population first so they can choose from it. Handing over your best examples instead of the full list is the fastest way to lose an auditor's trust. That is because a curated list proves nothing about what was left out.
Sample. The handful of items the auditor selects from the population to test in detail. For each sampled item they will want the full chain: the record, the approval or review, the dates, the person. Keep the complete list, and make sure every item on it has its record attached. You will not know in advance which five they pick, and you should not need to.
This is why monitoring exports are such strong evidence. An alert history export is a population. So is a configuration-change log or an inventory export. They are generated by a machine, dated automatically, and awkward to cherry-pick from, which is precisely what an auditor is looking for.
The small-team truth: you already make this
Read back over the evidence lists and notice how much your monitoring already produces. It keeps an alert history. It can generate an availability report. It records who acknowledged what and when. Your backup software emails you every morning. Your directory records every login. Your firewall logs every rule change. Your discovery scan already knows what is on the network.
The problem is not production; it is retention and attribution. The alert email gets deleted. The availability report is never generated because nobody scheduled it. The weekly review happens in someone's head over coffee, leaves no trace, and so, for audit purposes, never happened. The change was detected, the admin recognized it as their own work, and nobody wrote that down.
The fix is smaller than the request list suggests. It comes down to three habits:
- Schedule the outputs. Every report your monitoring can produce for a control family goes on a schedule, lands in a folder, and stays there. If the tool can email it, email it to a mailbox that is never emptied. If it can save it, save it with a date in the filename.
- Leave a trace when you review. A dated one-line note is enough: "Reviewed alert summary for the week, nothing notable, opened ticket for the switch that flapped twice. — initials." A ticket, a shared document, a spreadsheet row; the medium matters far less than the habit.
- Keep populations, not highlights. At the end of each period, export the full list: all alerts, all detected changes, all incidents, the full inventory. Do not decide what is interesting. Let the auditor do that.
None of these takes more than a few minutes a week once set up. Together they turn the audit from a reconstruction project into a retrieval task. The rest of this pillar builds on that. It covers retention that satisfies your requirements without keeping everything and evidence packs that generate themselves. It covers a matrix that maps your monitoring to your framework's controls. And it covers the drift alerts and self-audits that keep it all true between audits.
A first week's worth of work
If the audit is close, this sequence gives you the most evidence for the least effort. Auditors are far more sympathetic to "the control started operating on this date and here is the unbroken record since" than to a vague claim that everything has always been fine.
- Create one restricted folder called something like
audit-evidence, with a subfolder per control family. Limit who can write to it. - Schedule the availability report and the alert history export to land in it monthly. Backdate nothing; just start.
- Export the current device inventory and the log-source list, date them, and file them. Repeat monthly.
- Start the weekly review note. Even if the first one says "first review; established baseline," it is a dated record with a name on it.
- Export the change log and the incident register for whatever part of the period you can still reconstruct honestly. Mark where records begin.
- Write down, in one page, what you keep, for how long, and where. This becomes the policy the auditor asks for next.
Where to go next
Audit readiness, as far as monitoring is concerned, fits in one idea. The auditor wants to see that a control operated consistently across the period. Your monitoring already produces the records that show it. Keep the outputs, date them, attach a name, and hand over populations rather than highlights. Polish the dashboard afterwards if you like; nobody will ask about it. Every other guide in Monitoring for Compliance & Audit Evidence is a practical elaboration on that idea. The starting point is how long to keep the logs that back it all up.
Put this into practice with SysPeep Monitor
Most of the evidence in this guide is output your monitoring already makes. SysPeep Monitor keeps alert history and generates availability and health reports from a single web console. It auto-discovers the devices that become your inventory export. So the record exists whether or not an audit is on the calendar. The free trial installs on a Windows machine in minutes.
Download Free TrialLearn About SysPeep