Home › Topics › Justifying Monitoring Spend to Leadership › Downtime Math

Downtime Math: The Monitoring Dividend in Minutes and Money


At some point a finance director will ask the only question that matters to a finance director: "What does it pay back?" You can answer with adjectives, and the meeting will be polite, and the line will be cut. Or you can answer with a number you built from your own incident log, show the working, and watch the conversation change shape. This guide is about the second answer.


The number is what I call the monitoring dividend: the outage minutes that monitoring removed this year, multiplied by what a minute costs your organisation. It is not a vendor's return-on-investment figure and it does not come from a survey. Every input is a timestamp you can open, a count you can check, or a cost figure the business supplied. That is what makes it survive a sceptic, which is the whole point.


By the end you will be able to count the three sources of the dividend from records you already have. You will be able to work the arithmetic on a fictional example and lay it out in a spreadsheet. You will also be able to build a sensitivity table that halves every assumption and shows whether the dividend still clears the spend. This guide also tells you what to say when it does not, because sometimes it does not. Saying so is how you keep the right to be believed next year.


In this guide

  • The dividend, defined
  • Source one: detection minutes saved
  • Source two: repair minutes saved
  • Source three: incidents avoided
  • A worked example: Meridian Parts
  • The sensitivity table
  • The rules of honest arithmetic
  • The spreadsheet layout
  • A war story
  • The dividend checklist
  • Where to go next

The dividend, defined


The monitoring dividend is minutes times cost. The minutes are outage minutes that did not happen because monitoring existed, counted from your own incident history. The cost is your organisation's cost of one minute of a given service being down. Multiply, add up across the year, and compare with what the monitoring cost you. That is the entire calculation, and everything below is about getting the two inputs honestly.


The cost per minute is not this guide's job. The guide on framing findings as risk and cost shows how to build it from the number of people affected, their loaded hourly cost, and any lost transactions. It also shows how to state the cost as a range rather than a false-precision point. Build it there, once, per service, and bring it here as a finished input. You will need at least two figures: a cost class for a whole-site outage and one for a single-department outage. A dead core switch and a dead label printer are not the same minute.


The minutes are this guide's job, and they come from three places. Detection: the time between the alert and the moment a user would have reported it. Repair: the diagnosis time saved when the alert named the cause. Avoidance: the outages that never happened because a trend was caught first. Each has its own arithmetic, its own evidence, and its own level of softness. And you keep them separate all the way to the final table so that a sceptic can discount the soft one without discarding the hard two.


Source one: detection minutes saved


Detection minutes are the gap between the alert firing and the first user ticket, counted only when the alert came first. The logic is simple: without monitoring, the repair clock would have started at the ticket. With it, the clock started at the alert. The outage was shorter by the difference. You are not guessing how long users would have taken to notice, because for every incident you have a timestamp for when they actually did.


Pull two timestamps per incident from your records. The alert time comes from the monitoring system's alert log. The first ticket time comes from the help desk, or from the email inbox if that is your help desk. It is the earliest user-raised report for that incident. Subtract. If the result is positive, that is the lead in minutes. If it is negative or zero, the users beat the alert. The saving for that incident is zero, not negative. You do not owe the business minutes; you simply did not earn any on that one.


Two honesty adjustments. First, suppose an incident happened out of hours and nobody would have reported it until the morning. In that case, the true lead is from the alert to the start of the working day, not to the eventual ticket. Cap it there, because the outage would not have cost much overnight either. Second, if no user ever raised a ticket because the fix landed first, you have no user timestamp. In that case, the honest lead is the shortest lead you recorded for a similar incident in the same year, not the longest. The incident clock guide defines the timestamps precisely if your log does not already keep them.


The formula, per incident, and then summed:


detection_minutes(i) = max(0, first_ticket_time(i) - alert_time(i))
detection_dividend   = sum over incidents of detection_minutes(i) * cost_per_minute(class(i))

Source two: repair minutes saved


Repair minutes are the diagnosis time saved when the alert told you what was wrong. An alert reading "disk C: on fileserver01.example.com at 97 per cent" skips the part of the repair where someone works out which server, which disk, and why. An alert reading "host unreachable" does not; it started the clock but left the diagnosis to you. Count only the first kind, and count it conservatively.


The arithmetic needs a baseline, and you get it from your own tickets. Start with incidents where the alert did not name the cause (or there was no alert). For those incidents, measure the time from the ticket being opened to the first note that identifies the cause. Take the median, the middle value when they are sorted, so one nightmare afternoon does not drag the number. Do the same for incidents where the alert did name the cause. The difference is the repair minutes saved per named-cause incident. Multiply by the count of such incidents and by the cost per minute of each.


If your tickets do not record when the cause was found, you cannot count this source this year. Write "not measured" in the table, start recording it, and count it next year. A blank row you can explain is worth more than a filled one you cannot. (Spreadsheets are where optimism goes to be multiplied, and this is the cell where it usually starts.)


Source three: incidents avoided


Avoided incidents are the outages that never happened because a trend was caught and fixed first. Examples include the disk extended at eighty per cent, the certificate renewed a fortnight early, and the uplink upgraded before it saturated. This is the softest source, because you are pricing something that did not occur. It must be counted so conservatively that a sceptic would count it higher. Three rules keep it honest.


  • Count only catches with a projection. The trend line had to show the limit being reached within the quarter, and the fix had to land before that date. A disk extended at sixty per cent "just in case" is good practice, not an avoided incident.
  • Price each one at a real outage you already had. Find the shortest comparable outage in your own log — the last time a disk actually filled, the last time a certificate actually expired — and use its duration. Never use the longest, and never estimate one you have no precedent for; that row gets zero.
  • Degradation is not an outage. A link that would have been slow is not a link that was down. Count it at zero in this table and mention it in prose.

Keep the avoided total on its own line all the way to the summary, and always show the dividend with and without it. Most of this work is invisible by nature. And the pillar on making invisible work visible is about getting it recorded at the time. That is a great deal easier than reconstructing it in budget week.


A worked example: Meridian Parts


Meridian Parts is fictional, and every number below is invented for the example. The costs are in cost units, an abstract unit standing in for whatever currency and cost-per-minute figures your own organisation supplies. Meridian's finance team had provided two cost classes: forty units per minute for a whole-site outage and eight per minute for a single department. The year's incident log held fourteen user-affecting incidents.


Detection. The alert beat the first ticket on eleven of the fourteen. Three of those were whole-site, with leads of 22, 30, and 45 minutes: 97 minutes at forty units, 3,880 units. Eight were single-department, with leads of 8, 12, 15, 18, 25, 35, 40, and 60 minutes: 213 minutes at eight units, 1,704 units. The three incidents where users called first scored zero. Detection dividend: 5,584 units.


Repair. The alert named the cause on six of the fourteen. From the tickets, the median time to identify the cause was 35 minutes when the alert did not name it and 10 minutes when it did. That is a saving of 25 minutes per named incident. Two named incidents were whole-site (50 minutes at forty, 2,000 units) and four were single-department (100 minutes at eight, 800 units). Repair dividend: 2,800 units.


Avoided. Four trend catches had a projection and a fix before the date: two file-server disks, one certificate, one uplink. The shortest full-disk outage in Meridian's own history was 90 minutes, department-level: 720 units each, 1,440 for the pair. The last certificate expiry had taken the customer portal down for 100 minutes, whole-site: 4,000 units. The uplink was degradation, not outage: zero. Avoided dividend: 5,440 units.


SourceIncidents countedMinutesDividend (cost units)Softness
Detection11 of 143105,584Hard: two timestamps per row
Repair6 of 141502,800Medium: median from tickets
Avoided3 of 4 catches2805,440Soft: priced on precedent
Total74013,824Shown with and without the soft row
Total without avoided4608,384The number a sceptic will use

Meridian's full annual monitoring cost, built the way the budget line guide describes, came to 6,000 units. The dividend exceeded it with the soft row and without. That is a good year, and the next section is about proving it was not a lucky one.


The sensitivity table


A sensitivity table shows what happens to the dividend when each assumption is made worse, one at a time and then together. Its purpose is to answer the sceptic's questions before they are asked: "What if your cost per minute is too high? What if users would have noticed sooner? What if none of those disks would have filled?" You answer each with a row. And the last row is the one where you halve everything and drop the soft source at the same time. If the dividend still exceeds the spend, say so. If it does not, say that, and say which assumption tipped it.


Assumption changedDetectionRepairAvoidedDividendSpendVerdict
As calculated5,5842,8005,44013,8246,000Exceeds
Cost per minute halved2,7921,4002,7206,9126,000Exceeds, narrowly
Detection leads halved2,7922,8005,44011,0326,000Exceeds
Repair saving zeroed5,58405,44011,0246,000Exceeds
Avoided incidents zeroed5,5842,80008,3846,000Exceeds
Cost halved and avoided zeroed2,7921,40004,1926,000Does not exceed

Meridian's sentence for the meeting was: "The dividend exceeds the spend under every single assumption we halved, and only fails if the finance team's own cost per minute is wrong by half and none of the four capacity catches would have failed." That is a sentence a finance director can repeat upstairs. It is also, and this matters, a sentence that admits a way to be wrong, which is why it was believed.


When your own last row fails more easily than that, the sensitivity table is still the right thing to show. "The case holds on the hard numbers alone and needs the avoided incidents to hold at half cost" is honest and still persuasive. "The case only holds if every assumption is right" is honest and tells you to spend the year improving the inputs rather than the argument. Both are better than a single confident number that one question can knock over.


The rules of honest arithmetic


The dividend survives scrutiny only if every rule below is kept, because a sceptic needs to find just one broken rule to discount the whole table. I have broken most of these at one time or another, which is how I know which ones get found.


  • Never count an outage that would not have happened. A catch without a projection and a fix date is good housekeeping, not a dividend.
  • Never use an industry cost per minute. The figure comes from your own finance team or it does not appear. If they will not supply one, present minutes only and let them price it.
  • Show your working. Every minute traces to two timestamps, every count to a list of incident numbers, every cost to a named source. The spreadsheet, not the summary, is the deliverable.
  • Use the median for any "typical" value. One bad afternoon does not get to set the average.
  • Never count the same minutes twice. Detection minutes end at the ticket; repair minutes start at diagnosis. If your baseline for repair already includes noticing time, remove it.
  • Keep the soft source separate. Show the total with and without avoided incidents, every time, without being asked.
  • Round down. Always. Even when it hurts. Especially when it hurts.
  • Include the zeros. The three incidents where users called first stay in the table at zero. They are the coverage gaps, and they are next year's improvement plan.

Tip: if the alert log and the ticket system disagree about an incident's timeline, use whichever gives the smaller lead and note the discrepancy. The note costs you a few minutes of dividend and buys you every other row.


The spreadsheet layout


One workbook, four sheets, no macros. The layout below uses ordinary spreadsheet functions and a date-difference multiplied by 1440 to turn days into minutes. Adjust the syntax for your spreadsheet if it differs. The point is that a reviewer can click any cell in the summary and follow it back to a timestamp.


Sheet "Rates"
  A: cost_class        B: cost_per_minute      C: source
  whole-site           40                      finance, memo dated
  department           8                       finance, memo dated

Sheet "Incidents"   (one row per user-affecting incident)
  A: incident_id     B: service      C: cost_class
  D: alert_time      E: first_ticket_time
  F: cause_named_by_alert (Y/N)
  G: lead_minutes        = IF(E>D, (E-D)*1440, 0)
  H: repair_saved        = IF(F="Y", Baseline!B2 - Baseline!B3, 0)
  I: cost_per_minute     = VLOOKUP(C, Rates!A:B, 2, FALSE)
  J: detection_units     = G * I
  K: repair_units        = H * I

Sheet "Baseline"
  B2: median minutes to identify cause, alert did not name it   = MEDIAN(range)
  B3: median minutes to identify cause, alert named it          = MEDIAN(range)

Sheet "Avoided"   (one row per trend catch with a projection)
  A: catch_id   B: what was caught   C: projected fail date   D: fix date
  E: precedent_minutes (shortest comparable real outage, incident_id noted)
  F: cost_class   G: cost_per_minute = VLOOKUP(F, Rates!A:B, 2, FALSE)
  H: avoided_units   = E * G

Sheet "Summary"
  Detection dividend     = SUM(Incidents!J:J)
  Repair dividend        = SUM(Incidents!K:K)
  Avoided dividend       = SUM(Avoided!H:H)
  Total                  = sum of the three
  Total without avoided  = Detection + Repair
  Annual spend           = from the budget line
  Sensitivity rows       = each of the above with the changed input

Fill the Incidents sheet as the year goes, one row per incident at the time of the post-incident review. That is when the timestamps are fresh and the cause is known. Filling it in budget week from memory is how a table ends up with round numbers, and round numbers are the first thing a finance analyst notices.


A war story


Bellweather Logistics took a dividend table to their finance director, who brought an analyst, who brought a highlighter. The analyst picked a row at random and asked where 22 minutes had come from. The administrator opened the alert log and the ticket side by side, and the timestamps were 22 minutes apart. The analyst asked how anyone knew the second disk would have filled. The administrator opened the graph, with the trend line crossing the limit on a date eleven days after the extension. Three more rows were checked the same way. Then the director asked what the number was with the avoided incidents removed. The administrator turned the page to the row that already said so. The line was approved with a handwritten note in the margin reading "arithmetic checked", which the administrator has kept.


The dividend checklist


Work through this before the table leaves your desk. Every item is a yes or no, and any no is a row a sceptic will find.


  • Every cost per minute comes from your own finance team, with the source named on the Rates sheet.
  • Every incident in the year is listed, including the ones where users called first, at zero.
  • Every detection lead is the difference of two timestamps you can open.
  • Out-of-hours leads are capped at the start of the working day.
  • Repair savings use medians from your own tickets, or the row says "not measured".
  • Every avoided incident has a projection, a fix date, and a precedent outage from your own log.
  • Degradation is counted at zero.
  • The total is shown with and without the avoided row.
  • The sensitivity table has a row for each assumption halved and a row for the worst combination.
  • You have a one-sentence verdict for the meeting that names the assumption that would break the case.
  • Every number is rounded down.

Where to go next


The monitoring dividend is minutes times cost. And the whole discipline is in the minutes. Detection leads come from two timestamps, and repair savings come from ticket medians. Avoided incidents are priced on your own precedents and kept on their own line. Build the table as the year goes. Halve everything before anyone asks you to, and lead with the sentence that says where the case would break. A number that admits how it could be wrong is the only kind a finance director trusts.


The dividend is the arithmetic half of the case; before-and-after stories is the half that persuades the room. What monitoring spend actually buys supplies the plain sentences each row of the table supports. And the monitoring budget line supplies the spend figure the dividend is compared against. All of them are part of Justifying Monitoring Spend to Leadership.


Put this into practice with SysPeep Monitor

Every timestamp in the dividend table comes from the alert history. SysPeep Monitor keeps that history in the same web console as its dashboards and reports. So the two rows a finance analyst wants to check are a screen away rather than an export away. The free trial at /download/ runs on one Windows machine if you want a lab install. You can see what its alert log records before you build the spreadsheet against it.

Download Free TrialLearn About SysPeep