The Weekly Trend Review: Thirty Minutes That Prevent Outages
Everything else in this pillar is a way of seeing: a longer graph, a named shape, a catalogue of slow metrics, a rule that watches a slope. None of it works unless someone looks. Nobody looks at a graph that is not urgent unless there is a time in the calendar with their name on it. The weekly trend review is that time. It is thirty minutes, once a week, in front of one page. It is the single habit that turns monitoring history from a disk full of numbers into outages that did not happen.
I have run this review in three different organisations, with two people, with six, and alone. The version that survives is the small one. Thirty minutes is short enough to defend against every other meeting. One page is few enough graphs to finish. One line per finding is little enough writing that it gets written. Everything larger than that has, in my experience, been cancelled by spring.
This guide gives you the ritual in full. It covers who attends, the agenda with minutes against each item, and the three graphs to bring for every metric. It shows how to record a finding so it leads to something, with a findings log template you can copy. It also covers how to keep the review alive through holidays, migrations, and the weeks when everything is on fire.
In this guide
- The ritual, defined
- Who attends
- The agenda
- The graphs to bring
- How to record a finding
- The findings log template
- Keeping it to thirty minutes
- Surviving holidays and busy weeks
- The fourth week
- A first-review starter checklist
- Where to go next
The ritual, defined
The weekly trend review is a fixed thirty-minute slot, on the same day at the same time every week. The people responsible for the network look at a single page of long-range graphs. They note anything that has moved and write down what they will do about it. It is a ritual rather than a task because a task gets done when there is time and a ritual gets done on Tuesday. The distinction is the whole design. Nothing about creep is ever urgent, so nothing about creep ever wins a fight for attention against something that is. A calendar slot does not fight. It just arrives.
The review is not a troubleshooting session, a status meeting, or a place to discuss alerts that fired overnight. Those have their own homes. This one looks only at slow change: what is different this week from four weeks ago, and from a year ago. The guide on what degradation creep looks like explains why that question needs its own meeting. And the threshold review ritual is its natural sibling, held less often, for tuning the alerts rather than reading the graphs.
Who attends
Two people is enough, and one is fine. The ideal pair is the person who owns the network and the person who owns the servers. Most creep lives on the boundary between them. A retransmit count on a server might really point to a switch port. A firewall memory leak might really be a change in what the servers are sending. Two sets of eyes see two halves of the picture. Two people also keep each other honest about turning up, which is worth more than it sounds.
If you are the whole IT department, run it alone and run it anyway. Book the slot, open the page, and write the findings log as if someone else would read it. In six months the someone else is you. You will not remember what you thought a graph looked like in March. The IT department of one pillar has more on rituals that work for a single admin. What does not work, at any size, is inviting the whole department. Six people in a trend review produce one finding and forty minutes of discussion about it.
The agenda
The agenda is five items, with minutes against each, and the minutes are not decorative. The review fits in thirty minutes only because every item has a budget and the budget is enforced. Here it is as a block you can paste into the calendar invitation.
WEEKLY TREND REVIEW (30 min, same slot every week)
1. The weekly-look page ............ 10 min
Every panel, 90-day view. Anything higher, lower, or steeper than last week?
2. Anything that moved .............. 5 min
For each panel that changed: name the shape, guess the cause, assign an owner.
3. Trend rules that fired ........... 5 min
Each trend alert since last review: was it right, what was done, does the rule need tuning?
4. Capacity register ................ 5 min
Any item whose projected date moved closer? Any item due for its review date?
5. Last week's findings ............. 5 min
Read each open line in the findings log. Done, in progress, or overdue?
Item one is the look itself, using the page and the checklist from the guide on early-warning metrics. Item two is where a change becomes a sentence: "linear, probably the new scanning project, server team owns it." The guide on reading growth curves is the vocabulary for that sentence. Item three closes the loop on the rules from alerting on trends, because a trend alert nobody reviews drifts out of tune. Item four is a glance at the capacity register, the list of things with a projected date. Item five is the one that makes the other four worth doing.
Run the items in that order and do not reorder them to start with the interesting one. Item five goes last because it is the item people skip when time runs out. Putting it last with a five-minute budget means it gets skipped less than you would expect. (It still gets skipped. Just less.)
The graphs to bring
For every metric on the page, the review needs three views. The weekly-look page should show at least one of them, with the other two a click away. The year view shows the shape: whether the line is flat, rising, stepped, or bending. The month view shows what changed recently and whether the last week is part of the pattern or a break from it. The daily maxima, the highest value each day plotted as its own line, show creep in the peaks that both of the other views average away. For anything measured in errors, milliseconds, or queue depth, the daily-maximum line is the one that matters; for free memory it is the daily minimum.
Three views for twelve metrics sounds like thirty-six graphs, and it is. That is why the page shows the month view by default. The review only opens the other two for a panel that has moved. A panel that looks the same as last week gets a glance and a tick. A panel that has changed gets the year view to name its shape and the daily maxima to check whether the peaks are ahead of the average. Most weeks, that is two or three panels. And the other thirty graphs stay unopened, which is the correct number of graphs to leave unopened.
Bring last week's page too, as a saved report or a screenshot. The review is a comparison, and a comparison needs two things to compare. If your tool can email the page as a scheduled report each Monday morning, that email is last week's page, already in a folder, with a date on it. The dashboard design pillar covers building the page. The review only demands that every panel has the same time range and the same fixed axis starting at zero. Each also needs a label saying which direction is bad.
Tip: do not open the monitoring tool itself during item one. Open the page. The tool is where troubleshooting lives, and the moment someone clicks into a device to "just check something", the thirty minutes are gone. Note the device in the findings log and check it after the review.
How to record a finding
A finding is one line: the date, the metric, what changed, what was decided, who owns it, and when it will be looked at again. One line, because a finding that takes a paragraph does not get written, and the log's value is entirely in being complete. If the review notices twelve things and writes down four, the other eight are not findings. They are things someone once said in a meeting.
The "decision" column has four legal values, and "monitor" is not one of them. Investigate means someone will find out why, by the review date. Fix means the cause is known and someone will change it. Plan means it goes on the capacity register with a projected date. Accept means the change is understood and intended. In that case, the metric should be rebaselined so the alerts stop arguing with it. The guide on rebaselining explains how. "Monitor" is what you write when you have not decided, and a finding that says "monitor" will say "monitor" again next week and the week after. The action items that happen guide has more on why a decision needs a name and a date. And it applies here word for word.
The review date is the day the finding comes back to item five. Two weeks is the usual answer. A week is right for anything with a runaway shape. And a month is right for a plan item whose date is a year away. Never leave it blank. A finding without a review date is a finding that has been quietly accepted.
The findings log template
The log is a spreadsheet, or a table in whatever your team already uses for shared documents, with six columns and one row per finding. Keep it in the same place as the weekly-look page so the two are always opened together. The rows below are examples of the shape each line should take.
| Date | Metric | What changed | Decision | Owner | Review date |
|---|---|---|---|---|---|
| Week 14 | CRC errors, core-sw1 port 48 | Daily count rose from 0 to about 40 over 3 weeks, linear | Investigate: inspect cable and transceiver | Network admin | Week 15 |
| Week 14 | Free memory, main firewall | Daily minimum falling 0.3 percent per day since last reboot, sawtooth | Fix: scheduled reboot in next window; vendor case opened | Firewall owner | Week 16 |
| Week 14 | Disk used, mail store volume | Linear, about 1 percent per week; time to full 34 weeks | Plan: added to capacity register, extension by week 40 | Server admin | Week 22 |
| Week 14 | WAN utilisation, branch link | Step up on Tuesday week 13; new scanning project confirmed | Accept: rebaseline; new threshold agreed | Network admin | Week 18 |
| Week 15 | CRC errors, core-sw1 port 48 | Patch lead replaced week 14; count back to zero | Fix: closed | Network admin | Closed |
Notice that the last row is the first row, one week later, closed. That is what the log is for. A finding opened, investigated, fixed, and closed in eight days is a graph that never reached the review that follows an outage. Keep closed rows in the log; they are the evidence, months later, that the review is worth its thirty minutes. Keep real dates in the review date column, and replace the date with the word Closed when a finding is done. Then one spreadsheet formula tells you how many findings are overdue: =COUNTIF(F:F, "<"&TODAY()), with review dates in column F. Put it at the top of the sheet. If it is not zero at the start of item five, item five starts there.
Keeping it to thirty minutes
The review stays short because of three rules, and it grows long the week any of them is dropped. The first is no troubleshooting. The moment someone says "let me just log in and check", the finding gets a line in the log with "investigate" and a name. The review moves on. The second is the page, not the tool. Item one is read from the weekly-look page with its fixed panels, not from the monitoring console with its infinite drill-down. The third is the timer: someone watches the clock. When an item's minutes are up, the review moves to the next one regardless of how interesting the conversation has become.
Interesting conversations are the main threat. A trend review that finds something genuinely surprising, a runaway curve or an unexplained step, will want to solve it on the spot. Solving it on the spot takes an hour. Write it down, assign it, and solve it at eleven. The review's job is to find things, not to fix them. A review that fixes one thing and skips the rest of the page has traded eleven findings for one. I have made that trade. It was a good fix and a bad week.
Surviving holidays and busy weeks
The review will be skipped. Someone is on leave, a migration is running, an audit is in the building, or the week is simply on fire. The design question is not how to prevent skips but how to make the ritual survive them. And there are four habits that do it.
First, the slot is permanent and recurring in the calendar, and a skip is a declined instance, never a deleted series. Second, the weekly-look page is delivered automatically as a scheduled report. So the page exists for the skipped week even though nobody looked at it. The next review reads two weeks' worth. Third, a skipped week gets its own line in the findings log that says "skipped" and why. A log with a gap looks like a log that was abandoned. A log that says "skipped: migration" looks like a log kept by adults. Fourth, two skips in a row trigger a double-length review the following week. That rule is written at the top of the log where nobody can claim not to have seen it.
If the review has one attendee, the same habits apply, and the scheduled report matters most. A solo admin's review is read from an email on a Monday, and the email arrives whether or not the admin does. That is the closest thing to a ritual that survives being the only person in the building, and it works better than it should.
The fourth week
Acme skipped its trend review three weeks running. One was for a storage migration, one for a bank holiday, and one because the two attendees were both in an audit meeting that overran. In the fourth week the review opened the page and found three things at once. The office DHCP scope was at 96 percent after a delivery of handheld scanners. The main firewall's free memory was on course to reach zero in five days. And the mail store volume had been growing at one percent a week for a year. It had started growing at six percent a week after a retention policy change and had nine days left. The scope and the firewall were an afternoon's work each. The mail store needed an emergency storage purchase with expedited shipping and a Saturday. The ordinary purchase would have taken three weeks, and the disk did not have three weeks. The three skipped reviews had saved the team ninety minutes; the Saturday cost rather more than that. And the log now has a line at the top about what two skips in a row trigger.
Every one of those three findings had been on the page for at least a fortnight before the fourth week. Two of them would have been trivial at the first skipped review. The review did not fail; it was not held, and the graphs went on quietly doing what graphs do, which is wait.
A first-review starter checklist
The first review is different from every review after it, because the page does not exist yet and the log is empty. This is the list for the first one. Expect it to take an hour; every review after it takes thirty minutes.
- Book a recurring thirty-minute slot on the same day and time every week, and invite one other person if there is one.
- Build the weekly-look page with the early-warning metrics you can find today, every panel at 90 days with a fixed axis from zero.
- Set the page to arrive as a scheduled report each week, to everyone who attends.
- Create the findings log with the six columns from the template, and put the overdue-count formula at the top.
- Write the two-skips rule at the top of the log.
- In the first review, set every panel to twelve months once, name the shape of each, and log a finding for anything rising.
- Pick the two or three metrics that deserve a trend rule and write them into the trend alert spec.
- Save a copy of the page as the first "last week" for next week's comparison.
- End on time, even if the page is not finished; carry the remaining panels to next week.
Where to go next
Thirty minutes, one page, one line per finding, and a rule for the weeks it does not happen. That is the whole ritual, and it is the part of this pillar that does the work. The long graphs, the curve shapes, the metric catalogue, and the trend rules are all ways of seeing. The review is the appointment to look. Start with the first-review checklist this week. The rest of the Trend Analysis and Degradation Creep pillar explains what you will see. The capacity planning from history pillar takes the "plan" findings from the log through to a signed purchase.
Put this into practice with SysPeep Monitor
The review only survives if the page arrives whether or not anyone opens the tool. A central Windows service keeps long-term graphs from SNMP, WMI and SSH collectors. It puts them on one dashboard beside your thresholds and alert rules. It emails that dashboard as a scheduled report from the same web console. That turns the Monday look into something that happens on its own. The free trial installs in minutes.
Download Free TrialLearn About SysPeep