Home › Topics › Monitoring Scope: What to Watch (and Skip) › Criticality Tiers

Criticality Tiers: Ranking Systems So Monitoring Matches What Matters


Somewhere in your monitoring system right now, a test virtual machine and the server that runs the business are being treated as equals. Same checks, same interval, same alert severity. If either one dies at three in the morning, the same email goes to the same inbox with the same urgency. I ran a monitoring system like that for a couple of years and thought of it as fair. It was not fair; it was flat. Attention is a budget, and spreading it evenly means spreading it thin. The ten systems that deserve intense care get diluted care, and the two hundred that deserve a shrug get checks they never earned.


The fix is old, boring, and works. You rank your systems into criticality tiers — a small number of buckets, each defined by what happens when a system in it fails. Then you let the tier drive every monitoring decision: how deeply you check, how often, who gets alerted, and how fast anyone is expected to react. That is all tiering is. By the end of this guide you will have a three-tier model with clear membership rules. You will have a worked example of a real-world company estate sorted into it, arguments included. An estate is the full set of servers, devices, and applications you are responsible for. You will have a table showing exactly what each tier changes, and an agenda for doing the whole exercise in one afternoon.


In this guide

  • Why treating everything equally fails
  • The three tiers, defined by consequence
  • The four sorting questions
  • A worked example: sorting a 150-person company
  • What a tier actually changes
  • Who assigns tiers, and keeping them honest
  • Edge cases: shared gear and seasonal systems
  • The one-afternoon tiering workshop
  • Where to go next

Why treating everything equally fails


A monitoring system with no tiers is a flat list, and a flat list makes two opposite mistakes at once. First, it over-serves the trivial. The intern's sandbox VM gets a one-minute ping check and a wake-someone-up alert, because that is the default and nobody changed it. Second, it under-serves the vital. The database behind your order system gets exactly the same shallow "is it pingable?" check as the lobby display. Deep checks take effort, and effort was spread evenly across two hundred devices instead of concentrated on the ten that matter.


The damage shows up at the worst moment: alert time. When every alert arrives with the same severity, the ranking still has to happen. It just happens in the head of whoever is on duty, at 3 a.m., half asleep. They are deciding whether this down notification is a shrug or an emergency. Tiering is simply doing that triage in advance, in daylight, with coffee and colleagues, and writing the answer down. The phrase to keep in mind is the old one: if everything is critical, nothing is.


The three tiers, defined by consequence


Rank systems by one thing only: the consequence of failure. Not by how much the hardware cost, how new it is, how complicated it was to set up, or how much anyone loves it. A cheap old box can be the most critical thing you own, and an expensive new one can be a toy. Three tiers is the sweet spot. Two is too coarse to be useful. Five invites an hour-long debate about whether something is a 3 or a 4. Seven means somebody has been to a conference. Here are the three, with membership criteria.


Tier 1: work stops, or money or safety is at risk

A Tier 1 failure means a large share of the company cannot do their jobs, revenue stops flowing, or something safety-related is compromised. There is no acceptable workaround, or the workaround collapses within an hour. Typical members: the internet link, the firewall, the core switch, authentication (your domain controllers), the main line-of-business application, and anything customers touch directly. The test for membership is blunt: would you be willing to be woken up at 3 a.m. for this? If the honest answer is no, it is not Tier 1.


Tier 2: work degrades, but workarounds exist

A Tier 2 failure makes the day worse, slower, and more annoying, but the business limps along. A department is inconvenienced rather than stopped, or a workaround exists that people can actually use for a day or two. Typical members include the Wi-Fi (where wired fallback exists), secondary file shares, the intranet, and department-level applications. They also include the backup server (its failure risks the future rather than stopping the present). Tier 2 problems deserve a same-business-day response, not a midnight one.


Tier 3: annoyance

A Tier 3 failure inconveniences an individual or is purely cosmetic. Someone re-sends a print job, a meeting room display shows the wrong calendar, a test VM is unreachable. Nobody's day is ruined, no money is lost, and the fix can wait for a quiet moment. Tier 3 systems get light monitoring or none at all. And "none at all" is a legitimate, documented choice, which is exactly what the not-to-monitor list is for. The word "documented" is doing the work in that sentence. An unmonitored system with a written reason is a decision; an unmonitored system without one is a thing you forgot.


The four sorting questions


When you sit down to classify a system and the answer is not obvious, four questions sort almost everything. Ask them in order and let the answers argue it out.


QuestionWhat it revealsHow it moves the tier
How many people stop working?Blast radiusMost of the company → Tier 1. One department → Tier 2. One person → Tier 3.
Is it on the revenue or safety path?Business exposureIf orders, payments, shipments, or safety depend on it, push toward Tier 1 regardless of headcount.
Is there a workaround people will actually use?SurvivabilityA real workaround pulls a system down a tier. "Theoretically we could use paper" is not a workaround.
How hard is recovery?Time at riskSlow, fragile, or nobody-knows-how recovery pushes a system up a tier — early warning matters more when the fix is long.

The recovery question surprises people. Two systems with identical failure impact can deserve different tiers. One might restart itself in two minutes while the other needs a day-long restore from backup plus a vendor call. The longer and scarier the recovery, the more valuable it is to catch trouble early — which is what a higher tier buys you. When the four answers disagree, the worst answer usually wins, but only if someone can say why in one sentence. "It feels important" is not a sentence.


A worked example: sorting a 150-person company


Theory is easy; the arguments are where tiering gets real. Here is a typical estate for a 150-person distribution company. It has an office, a warehouse that ships orders all day, and a handful of remote salespeople. The estate is sorted into tiers, with the one-line reasoning that settled each row. Your business will differ, and that is the point: the reasoning column, not the tier column, is the part you are copying.


SystemTierThe one-line reasoning
Internet link & firewall1Everything cloud, VPN, and customer-facing dies with it. No workaround.
Core switch1Carries every Tier 1 service; inherits their tier automatically.
Domain controllers1No logins, no file access, no printing — "everything is broken" in one box.
ERP / order system1Orders and invoicing stop. This is the revenue path.
Warehouse label printer & its print path1No labels, no shipments. A printer, yes — but on the revenue path.
Email1Argued hard. Customers order and complain by email; an hour of silence costs real money here.
VPN gateway1Remote sales team stops working entirely; no workaround.
The fax line1The largest customer sends purchase orders by fax; it is on the revenue path. See the argument below.
Main file server1Quotes, specs, and shipping docs live there; the whole office touches it hourly.
Office Wi-Fi2Painful, but desks have wired ports and the warehouse scanners have their own APs (which are Tier 1).
Backup server2Its failure stops nothing today but quietly cancels the safety net; same-day fix, never a 3 a.m. page.
Intranet wiki2Promoted from Tier 3 when someone pointed out the runbooks live on it — it matters most mid-outage.
Payroll server2*Workarounds exist for 27 days a month; the asterisk is the seasonal clause covered below.
Server room air conditioning & UPS2Failure starts a countdown clock rather than an outage — but the clock is short, so alerts are immediate even though the tier is 2.
Badge / door system2Doors fail to a safe state; the workaround is keys and patience.
Office printers3Someone re-sends the job to the printer next door. Including the CEO's — see below.
Meeting room displays3Annoying, never urgent.
Test & sandbox VMs3By definition nobody depends on them. One of them did not match its name; see below.

Three rows in that table caused actual arguments, and each argument teaches something. One of them is the fax line. Yes, really.


The CEO's printer debate

Someone will suggest, half-joking, that the CEO's printer is Tier 1. It is a real test of whether your tiers mean anything. The consequence-of-failure logic is unmoved by job titles: one person re-sends a document to the device down the hall. What the CEO actually wants is not a 3 a.m. page to the on-call admin. It is a fast, cheerful fix during business hours, which is a support-process promise, not a monitoring tier. The printer stays Tier 3. If you cave here, you have announced that tiers are negotiable by seniority, and the whole list becomes politics.


The test server that turned out to be production

The box named test-app-02 was waved through as Tier 3. Then the warehouse manager mentioned that the shipping-label service "temporarily" deployed to it two years ago never moved. It was serving the Tier 1 label printer all along. The name was a lie, and it had been a lie for two years. This is one of the quiet payoffs of the tiering exercise: it interrogates reality. Tier by what runs on the system today. Never tier by its name, its VLAN (the network segment it sits on), or what the wiki says it was built for. Every estate has at least one of these; the workshop is where you find it, instead of during the outage.


Why the ancient fax line is Tier 1 here

The company's largest customer — a hospital group — sends purchase orders by fax, because their procurement system has always done so. That one wheezing analog line carries a double-digit percentage of revenue. In almost any other business it would be Tier 3 or unplugged; in this one it is unambiguously Tier 1. This is the deepest lesson of tiering: tiers are business knowledge, not tech specs. You cannot read criticality off a spec sheet, a rack diagram, or an asset database — you learn it by asking what the business would feel. That is also why tiering is really a scope decision in disguise. It is why tiering belongs alongside the goals conversation covered in monitoring goals and scope.


What a tier actually changes


A tier that changes nothing is a sticker. The entire value of the exercise is that tier membership mechanically drives concrete monitoring settings. When someone asks "why does this box page you and that one doesn't?", the answer is "Tier 1 versus Tier 3," not a shrug. Here is the mapping that makes tiers real.


TierCheck depthCheck frequencyAlerting & routingResponse expectation
1Full stack: reachability, resources, the actual service, and its dependenciesEvery 1 minuteImmediate, high severity, notifies a human who is expected to act — around the clockStart within minutes, any hour
2Standard health: reachability, disk, CPU/memory, key servicesEvery 5 minutesNormal severity to the team channel or inbox, business hoursSame business day
3Reachability only — or deliberately nothingEvery 15–60 minutesDashboard status or a daily digest; never an interruptionBest effort, when convenient

Check depth deserves a word. For a Tier 1 system, "is it up?" is nowhere near enough. You want the layered checks described in the first ten checks applied in full. That means ping, resource metrics, and a check that exercises the real service (a real DNS lookup, a real login page fetch). It also means checks on the things it depends on. For Tier 3, a ping is plenty, and for some Tier 3 systems even that is more than the box deserves.


Notice what else quietly falls out of the same list: maintenance priority. The tier column tells you which systems get patched first, which get spare hardware on the shelf, which earn redundancy budget, and which wait. One afternoon of classification ends up steering monitoring, alerting, patching, and purchasing — a remarkable return on a one-column spreadsheet.


Who assigns tiers, and keeping them honest


Here is the part IT teams get wrong by default: tiers are not IT's decision to make alone. IT knows what breaks and how; the business knows what a failure costs. The fax-line story proves it — no admin would have ranked that line correctly without the sales conversation. Classify together with someone who owns business outcomes: an operations manager, a department head, the owner in a small firm. Then get an explicit sign-off. The sign-off is not bureaucracy; it is armor. When someone later demands to know why their pet system did not page anyone at midnight, the answer is "here is the tier list you approved." The conversation then becomes a calm review instead of a blame hunt.


Tiers also rot, because the business moves. The pilot app becomes the company standard; the "critical" legacy system quietly loses its last user and keeps its tier for another three years. Put a tier review on the calendar — thirty minutes once or twice a year. Also review after any major change like an acquisition, a new product line, or a big migration. The review is cheap because you wrote down the reasoning the first time: you re-read each one-line justification and ask "still true?"


The chronic disease of tier systems is inflation. Every owner believes their system is critical, and every promotion request sounds reasonable in isolation. Eighteen months later two-thirds of the estate is Tier 1. You are back to the flat list you started with — just with extra paperwork. Resist with two tools. First, price it: every Tier 1 addition means more interruptions and more midnight duty for a real human. So each request must name the consequence that justifies waking someone. Second, ask the forcing question below.


Meridian Parts showed me their tier list with some pride, because every system on it had a tier. Tier 1 had ninety-one entries. Among them were a label printer in a depot that had closed, three test VMs, and the server that hosted the tier list. Tier 3 had four. The on-call engineer had stopped reading the severity field months earlier, because it no longer said anything. A genuine core router failure sat in the same queue as a paper jam for forty minutes. The review described the list as "comprehensive". It was rebuilt in one afternoon with the question in the tip below, and Tier 1 came out at eleven.


Tip: the tier-inflation antidote is one question, asked kindly: "What are you willing to be woken up at 3 a.m. for?" When the person requesting Tier 1 status understands that the promotion means someone's phone — possibly theirs — goes off at night for this system, most requests politely reconsider themselves. Tier 1 should be a small minority of your estate. If it is more than roughly the count you could recite from memory, inflation has already started.


Edge cases: shared gear and seasonal systems


Two situations trip up every first-time tiering effort, so handle them explicitly.


Shared infrastructure inherits the highest tier it carries

What tier is a switch? Wrong question — a switch has no tier of its own. It inherits the highest tier of anything that depends on it. The access switch feeding only meeting-room displays is Tier 3. The identical model carrying the ERP server's uplink is Tier 1, full stop, even if it also serves the lobby TV. The same inheritance applies to hypervisor hosts (the physical machines your virtual machines run on), storage arrays, UPS units, and the server room's air conditioning. A dependency of a Tier 1 system is a Tier 1 system. The air conditioning does not know it is Tier 1. It finds out in August. Tracing those threads — finding out what actually sits underneath each critical service — is a discipline of its own, covered in dependency thinking.


Seasonal tiers

Some systems change criticality on a schedule. Payroll is the classic: for most of the month a failure is a Tier 2 inconvenience. But in the three days before pay day it is as Tier 1 as anything you own. Nothing tests company morale like late paychecks. Retail systems spike before holidays; school systems spike at term start. You do not need anything fancy. Either schedule stricter alerting for the known window if your tooling supports it, or simply write the clause into the tier list. The clause is "Tier 2; treat as Tier 1 from the 25th to month-end", so the on-call human knows the rules changed. The written clause is the important part; tribal memory about seasonal criticality is how payroll outages become legends.


The one-afternoon tiering workshop


You do not need a project, a consultant, or a framework with a trademark. You need one afternoon, the right three or four people, and a list. Expect at least one fax-line surprise and one test-server confession along the way; every estate has them. Here is an agenda that works.


  • Before the meeting (30 min, IT): export a flat list of every server, network device, and major application from your inventory or monitoring tool. Do not pre-sort it — pre-sorting smuggles in IT's assumptions, and the assumptions are what the meeting exists to test.
  • Invite (5 min): one or two people from IT, plus someone who owns business outcomes — operations manager, finance lead, or the owner. Small on purpose.
  • First pass (45 min): walk the list fast — thirty seconds per item, gut-call the tier using the four questions. Anything contested goes to a debate pile instead of eating time.
  • The debate pile (45 min): now argue, item by item. Apply the 3 a.m. question to every proposed Tier 1.
  • Write the reasons (30 min): one sentence per system, in the reasoning column. This sentence is what makes next year's review take thirty minutes instead of another afternoon.
  • Map tiers to policy (20 min): agree on your version of the tier-mapping table above — frequencies, alert routing, response expectations per tier.
  • Sign-off and schedule (5 min): the business attendee approves the list; a calendar entry books the next review.

Then spend the following days making your monitoring configuration match the table. Deepen the Tier 1 checks, calm the Tier 3 alerts, and route notifications by tier. An imperfect tier list applied today beats a perfect one that never leaves the whiteboard.


Where to go next


A criticality tier is one column in a spreadsheet, and that column decides check depth, alert routing, response promises, and even patch order. It is the cheapest piece of monitoring work you will ever do. It is also the one most estates have not done, which is why the test VM and the order server are still being treated as equals. One natural companion is dependency thinking, which finds the unglamorous systems that silently inherit Tier 1 status. Another is the not-to-monitor list, which documents the other end of the scale. To keep the list alive, fold tier assignment into every new deployment's intake so nothing goes live unranked. For the wider discipline of deciding what deserves attention at all, the rest of Monitoring Scope: What to Watch (and Skip) picks up from here. Book the afternoon. The flat list is not going to sort itself.


Put this into practice with SysPeep Monitor

A tier list only matters once your tools enforce it. SysPeep Monitor lets you set check intervals, thresholds, and alert rules per device. That lets Tier 1 systems get one-minute checks that notify a human immediately while Tier 3 stays a quiet line on a dashboard. Its auto-discovery builds the estate list your tiering workshop starts from. The free trial installs on any Windows machine.

Download Free TrialLearn About SysPeep