Keeping the Queue Honest: Metrics, Hygiene, and Reviews
A monitoring-to-ticketing bridge is never finished. On the day you switch it on, the rules are right for the network as it was that day. Six months later the network has changed and three new device classes have appeared. Someone adjusted a threshold without telling anyone. The queue has drifted from "the record of work" to "the place monitoring tickets go to be ignored." Nobody decided this. It happened one unread ticket at a time. Every queue I have inherited had one monitoring ticket, years old and technically still open, that nobody could explain and nobody would close.
The bridge stays honest the same way any other system does. Somebody looks at it regularly, with a short list of questions and the authority to change things. This guide gives you the list. It names the four ways a queue goes bad and the hygiene rules that catch each one early. It covers the metrics you can compute from the tickets you already have — without inventing a single number. It also covers a monthly review that takes an hour and pays for itself the first time it runs.
It also closes a loop that the earlier guides in this series opened. Tickets are created from alerts; what the tickets reveal should change the alerts. A ticket that was never useful is a reason to stop creating it. A ticket that recurs every week is a reason to fix something. The queue is not only where work goes; it is where the monitoring system finds out how well it is doing.
In this guide
- The four ways a queue goes bad
- Hygiene rules that catch drift early
- Closing codes: the smallest useful vocabulary
- Feeding lessons back into monitoring
- Metrics you can compute today
- The monthly queue review checklist
- A worked review
- Where to go next
The four ways a queue goes bad
The landfill: auto-tickets nobody works
Tickets are created, nobody touches them, and after a while nobody expects to. The cause is almost always volume: too many auto-tickets for the team to keep up with, so they stop trying. The symptom is a pile of tickets in "new" with no assignee and a creation date weeks old. The dangerous part is that a landfill hides real work, because the one ticket that needed hands is sitting between forty that did not. You detect it with an aging report, described below, and you fix it by cutting volume at the source, not by closing tickets faster. Closing tickets faster gets you a faster landfill.
Closed without a cause
Tickets get closed with "resolved," "fixed," or nothing at all. The work may have been done perfectly. But a ticket closed without a cause teaches nobody anything. The next technician facing the same failure cannot learn what fixed it last time. The enrichment lookup described in ticket enrichment returns an empty "last time" field. The monthly review cannot tell a false alarm from a hardware failure. The symptom is a closing-note field that is blank or single-word on most auto-tickets.
Tickets that outlive the alert
The alert cleared days ago; the ticket is still open. Sometimes the bridge failed to append the recovery note. Sometimes it did, and the ticket is waiting for human confirmation that never comes. Either way the queue now contains work that does not exist. Technicians learn to assume that an open monitoring ticket probably is not real. You detect it by comparing open tickets against the monitoring system's current alert state; any open ticket whose alert is not active is a candidate.
Category drift
The bridge assigns a category from the device class. A human re-categorises the ticket to something else. A new device class appears that maps to nothing and lands in "other." Over months the categories stop meaning anything, and the reports that group by category stop being trustworthy. The symptom is a growing "other" or "uncategorised" bucket, and categories whose names nobody can define. (There is always one called "Network — Other".)
Hygiene rules that catch drift early
Four standing rules, each aimed at one of the failure modes above. They are cheap individually and, in combination, they make the monthly review possible. One of them counts by dedup key: the device-plus-check string the bridge stamps on every ticket to identify one incident.
- Every auto-ticket has a category and a closing code. The bridge sets the category at creation; the ticketing tool requires a closing code before a monitoring ticket can be resolved. Make the code a pick-list, not free text, so that it can be counted. The next section proposes the list.
- Auto-closed tickets are sampled weekly. Pick a handful of tickets the bridge resolved on its own and read them. Was the recovery real? Would a human have wanted to add something? If the sample keeps turning up tickets that should have been held for confirmation, the resolution rules need a tighter table. Those rules are in the sibling guide on auto-closing and updates.
- An aging report runs automatically. A saved query or a scheduled report listing every open monitoring ticket older than a few days, grouped by assignment group. It should go to the people who own those groups, not to a mailbox nobody reads. This is the landfill detector.
- The top sources get a "why was this ticketed?" review. Once a month, list the ten dedup keys that created the most tickets, and for each one ask whether the tickets led to action. This is where most tuning decisions come from.
If you adopt only one, adopt the closing code. It is the input to everything else.
Closing codes: the smallest useful vocabulary
A closing code is a single pick-list value that says what the ticket turned out to be. The temptation is to build a taxonomy of fifty. Resist it, because a code that technicians have to think about is a code they will pick at random. Eight cover nearly every monitoring ticket, and each one points at a specific action for the monitoring side.
| Closing code | Meaning | What it feeds back |
|---|---|---|
| Fixed: root cause | The underlying fault was found and corrected | The alert worked; nothing to change |
| Fixed: symptom only | A restart or workaround cleared it; cause unknown | Candidate for a problem ticket if it recurs |
| Self-recovered | It came back before anyone acted | Consider a longer confirmation delay or no ticket |
| False alarm: threshold | The condition was real but not a problem | Tune the threshold or the baseline |
| False alarm: check | The check itself was wrong (bad credential, wrong target, stale inventory) | Fix the check or the inventory |
| Duplicate | Another ticket already covered it | Fix the dedup key or dependency rules |
| Expected: maintenance | Planned work that should have been suppressed | Fix the maintenance window or its scope |
| No action needed | Real, but nobody would ever act on it | Stop ticketing this alert; possibly stop the alert |
Six of the eight codes in the right-hand column are instructions to change something in monitoring. That is the point. The closing code is how a technician, in the two seconds it takes to pick a value, tells the monitoring system what it got wrong.
Feeding lessons back into monitoring
Once closing codes exist, the feedback loop almost runs itself, because each pattern in the codes maps to one of a small number of responses.
A ticket that recurs has the same dedup key producing a ticket week after week, closed each time as "fixed: symptom only" or "self-recovered". It has three possible answers, and the review should pick one deliberately. Tune the threshold if the condition is real but the line was drawn in the wrong place. For example, a disk that sits at 86% forever is not a problem at 85%. Fix the root cause if the condition is a genuine fault that keeps returning. In that case, open a problem ticket, link every incident to it, and give someone the afternoon to find out why. Build a remediation if the fix is always the same safe action and the cause cannot be removed. In that case, the service that needs restarting after every nightly job is a candidate for a script rather than a technician. The Auto-Remediation guides explain how to do that without creating new problems.
A ticket that was never useful — closed as "no action needed" most of the time it appears — is a reason to stop creating it. That does not necessarily mean deleting the alert; it may still belong in a log or a daily digest. It means removing that alert type from the bridge's allow-list, which is the filter described in the first guide of this series. Every such removal makes the remaining tickets more trustworthy.
A ticket that was a false alarm goes back to whoever owns the check. "False alarm: check" almost always means the inventory and the monitoring system disagree about something. Examples include a decommissioned server still being polled or a renamed interface. In that case, the fix is in the inventory. "False alarm: threshold" is a tuning task and should be tracked as one, so it actually happens rather than being noted and forgotten.
A duplicate is a bridge bug, and the only item on this list you can fix without a meeting. Either the dedup key was built wrongly for that device class, or the monitoring system's dependency rules did not recognise a parent-child relationship. In the latter case, forty children came through. Both are fixable in an afternoon, and the sibling guide on auto-creating tickets covers where each rule lives.
Tip: keep a short "monitoring changes made because of the queue" log — date, dedup key, what was changed, why. It takes a minute per entry. After a few months it is the most persuasive document you have when someone asks whether the ticket integration is worth maintaining.
Metrics you can compute today
You do not need a dashboard product or a target from a benchmark to measure the queue. Every metric below is a query against ticket fields you already have, and each one detects a specific failure mode. Trend them month over month; the direction matters far more than the absolute value. What a "good" number looks like is entirely specific to your team and your network.
| Metric | How to compute it | What a rising value suggests | What a falling value suggests |
|---|---|---|---|
| Tickets created by monitoring vs by humans | Count by the source field, per month | Bridge volume is growing; check whether the work is | Either quieter infrastructure or alerts being removed — find out which |
| Time to first touch | Created time to first human note or assignment, median per group | Landfill forming; too many tickets or too few eyes | The queue is being read |
| Reopen rate | Tickets reopened, divided by tickets resolved, per month | Fixes are not sticking, or grace periods are too short | Recoveries are real |
| Auto-close rate | Tickets resolved by the bridge, divided by all monitoring tickets | More tickets need no human; consider not creating them at all | More tickets need hands, or the resolution rules tightened |
| Open beyond a set age | Count of open monitoring tickets older than N days, by group | Landfill, or tickets outliving their alert | Hygiene is working |
| Closing-code mix | Count of each code, per month | Depends on the code: "no action" rising means over-ticketing; "root cause" rising means the bridge is carrying real work | Likewise, by code |
| Top sources | Tickets per dedup key, top ten | A few devices dominating the queue; problem-ticket candidates | Recurring faults are being fixed |
| Enrichment "unknown" count | Tickets whose enriched fields say unknown, by field | Inventory is drifting | Inventory is being maintained |
Two cautions. The auto-close rate is the metric most often misread. A high value can mean the bridge is doing its job beautifully, or that it is creating tickets for conditions that never needed one. Only the weekly sample of auto-closed tickets can tell you which. And time to first touch is a measure of the queue's health, not of any individual's speed. The moment it is used to rank people, technicians will start touching tickets to stop the clock rather than to work them. Wider incident measurements — detection and resolution times and how to present them — are the territory of the MTTD and MTTR guides. The metrics here are narrower, and they exist to keep the bridge honest.
At Bellweather Logistics the auto-close rate had climbed steadily for a year. The monthly report presented it as good news: the bridge was resolving most of what it created. The first time anyone sampled the auto-closed tickets, they found two hundred and eight from a single quarter. Those had been created, recovered, and resolved without a human opening one of them. Most were a depot's wireless controller rebooting at two every morning, which turned out to be a scheduled job nobody remembered scheduling. The report had been right about the bridge and wrong about the work.
The monthly queue review checklist
One hour, once a month, with the person who owns the monitoring system and someone who works the queue. Bring the aging report, the closing-code counts, and the top-ten sources. Work through the table and leave with a list of changes.
| Item | Question | Action if the answer is bad |
|---|---|---|
| Aging report | Are there open monitoring tickets older than the agreed age? Whose are they? | Assign or close each one; find the volume source if there are many |
| Outlived alerts | Do any open tickets have no active alert behind them? | Close with the right code; check the bridge's recovery handling |
| Auto-closed sample | Of the sampled auto-closed tickets, would a human have wanted to add anything? | Move that ticket type to "hold for confirmation" |
| Top ten sources | For each, did the tickets lead to action? | Tune, root-cause, remediate, or stop ticketing |
| Closing-code mix | Which code grew most this month? | Follow that code's feedback action from the table above |
| Blank closing notes | What share of monitoring tickets closed with no note? | Make the note field required for the monitoring category |
| Category health | Is "other" growing? Are there new device classes with no mapping? | Add mappings; retire categories nobody can define |
| Reopen rate | Which keys were reopened, and why? | Problem ticket, or a longer grace period |
| Enrichment gaps | Which enriched field says "unknown" most? | Send the inventory owner the list |
| Rules changed | What did we change in monitoring or the bridge since last month, and did it help? | Record it in the changes log; revert if it did not |
If you are a team of one, the weekly version is ten minutes. Scan the aging report, read two auto-closed tickets, and look at the single busiest source. The monthly hour is for the rest.
A worked review
This is a review as it might go, on a modest network, using nothing but the queries above.
Top sources this month (tickets per dedup key): fs01/disk-D 9 closing codes: symptom-only x8, self-recovered x1 branch-rtr-2/unreachable 4 closing codes: root-cause x1, self-recovered x3 ups-b/on-battery 3 closing codes: root-cause x3 lab-sw-3/unreachable 3 closing codes: no-action x3 prn-wh-01/unreachable 2 closing codes: no-action x2 Aging: 6 open monitoring tickets older than 7 days, all lab-sw-3 or prn-wh-01. Auto-close sample: 5 read; 1 (branch-rtr-2) should have been held for a cause. Reopens: fs01/disk-D reopened twice.
Walk the findings. fs01/disk-D is nine tickets, all closed "symptom only" with notes that say some variation of "cleared temp files." That is a recurring fault with a known symptom fix and an unknown cause. Open a problem ticket, and the afternoon of investigation finds an application whose log rotation was never configured. Fix the rotation; the tickets stop. Had the cause been unfixable, the same review would have concluded that clearing the temp directory is a safe scripted remediation.
branch-rtr-2 recovered by itself three times and was auto-closed each time. The sampled ticket shows nobody ever wrote down that the carrier had an outage. Change the rule: site routers are core tier, and core-tier recoveries are held for human confirmation. ups-b is three real incidents, three root causes, all closed with useful notes. The alert is doing exactly its job and nothing changes. lab-sw-3 and prn-wh-01 are the landfill: five tickets, all "no action," six of them still open past the aging limit. The lab switch is switched off on weekends, and the warehouse printer is a printer. Remove both from the bridge's allow-list; keep the lab switch alert in the daily digest, and let the printer be somebody else's spiritual burden.
One hour, four decisions, and next month's queue is smaller and more truthful than this month's. That is what a review is for.
Where to go next
A queue stays honest when tickets carry a category and a closing code, and when auto-closed tickets are read occasionally. That also depends on an aging report reaching the people who can act on it, and the busiest sources being questioned once a month. The closing codes turn the queue into feedback for monitoring. The metrics turn it into evidence. The review turns both into changes. This is the last guide in Connecting Monitoring to Ticketing and ITSM, and it is the one you will come back to most often. That is because the other four describe something you build once and this one describes something you keep doing. Keep doing it, and the ticket that has been open for years finally gets closed, with a code and a cause. When an incident needs more than a ticket — a coordinated response with roles and a timeline — the Incident Response guides take over from here.
Put this into practice with SysPeep Monitor
Half of the closing codes in this guide are instructions to tune monitoring, and a system built for tuning makes that loop short. SysPeep Monitor's thresholds, dependency rules, and alert rules are designed to keep alert noise down. That means fewer tickets that close as "no action". Its reports and dashboards show what actually went down and when. So the monthly review argues from evidence. It does not run your queue. Check the product documentation for its notification options, and try the free trial on any Windows machine.
Download Free TrialLearn About SysPeep