Home › Topics › Capacity Planning From Monitoring History › Lead Time and Triggers

Lead Time: Buying Before the Pain Starts


A forecast tells you when a resource will be full. It does not tell you when to order more. The gap between those two dates is where most capacity plans quietly fail. You know the disk fills in May. Have you ever bought anything through a purchasing department? Then you know that "we should order a disk" and "the disk is in service" are separated by a chain. That chain includes quotes, approvals, deliveries, and change windows that nobody has ever measured. That chain is the lead time. It is the number this guide is about.


By the end you will be able to find the real lead time for anything you buy, from a licence key to a circuit. You will measure it rather than guess. You will set a trigger date that leaves a margin for the things that always slip. You will put that trigger into the monitoring system as a threshold so it arrives in your inbox instead of your memory. You will keep a register that shows at a glance which purchases should already be moving. You will also see what the pain phase costs when you miss the trigger.


In this guide

  • What lead time really measures
  • The chain, link by link
  • Time-to-full versus time-to-buy
  • The trigger point and the safety margin
  • Putting the trigger into monitoring
  • The pain phase and what it costs
  • Renewals and licence counts are capacity too
  • The register with trigger dates
  • The circuit that lost a race with a disk
  • Checklist
  • Where to go next

What lead time really measures


Lead time is the elapsed time from the moment you decide to buy to the moment the new capacity is doing its job. It starts when you send the email asking for a quote, not when the order is signed. It ends when the data has moved, the traffic is flowing, or the licence is applied, not when a box arrives at reception. Both ends matter, and both are routinely measured wrong: the start too late and the end too early.


The reason the definition is strict is that every link in the chain takes calendar time. Calendar time is what the disk is filling in. Suppose a four-week delivery is preceded by three weeks of approval and followed by two weeks waiting for a change window. That is a nine-week lead time. If your register says four, you will order five weeks too late, with perfect confidence. (Confidence is the one thing the register never runs out of.)


Lead time is also different for every kind of resource, and often for the same resource in different seasons. A licence key can arrive the same day. A disk shelf takes weeks. A circuit takes months, and a circuit ordered just before the year-end freeze takes months plus the freeze. The guide on capacity planning basics introduced lead time as a column in the register. This guide is about what to write in it.


The chain, link by link


Every purchase walks the same chain. The way to know your lead time is to know how long each link takes in your organisation. Here are the seven links. For each one, the honest answer to "how long?" is "ask the people who do it, and check the last three times it happened."


  1. Quote: you describe what you need, a supplier prices it. Ask your usual supplier how long a quote takes, and look at the dates on the last three you received.
  2. Approval: someone with budget authority says yes. Ask finance what the approval path is for the amount involved and when the approving meeting sits. A purchase that just misses a monthly meeting waits a month.
  3. Purchase order: the approved request becomes an order the supplier will act on. Ask procurement how long from approval to a raised order, and whether new suppliers need to be set up first.
  4. Delivery: the supplier ships or provisions. Ask the supplier for the current delivery estimate, and check what the last delivery actually took against what was promised.
  5. Installation: the thing is racked, cabled, configured. Ask whoever runs the change process when the next available window is, and how far ahead a change must be booked.
  6. Migration: data moves, traffic cuts over, devices are relicensed. Ask yourself, honestly, how long the last migration of the same kind took including the weekend it overran.
  7. Verification: the monitoring graph confirms that headroom increased. Allow at least a week of normal use so the new numbers are real, not a quiet Sunday.

The best source for every link is history, the same as for the forecast. Your purchasing system holds the dates of the request, the order, and the delivery for everything you have ever bought. Your change system holds the dates changes were requested and performed. Pull the last three purchases of each kind and read the gaps between the dates. That is your measured lead time. It is usually longer than anyone in the room would have guessed, and it is the only figure that will survive the next review.


Here is a worked example for a fictional organisation, Acme, with every duration measured from its own last three purchases. Yours will differ, which is the point of measuring:


LinkStorage shelfWAN circuitMonitoring licences
Quote1 week2 weeks (site survey first)2 days
Approval2 weeks (monthly meeting)4 weeks (two signatures)1 week
Purchase order3 days1 week1 day
Delivery4 weeks10 weeks (provisioning)1 day (key by email)
Installation1 week (next change window)2 weeks (engineer visit)same day
Migration1 week (data copy)1 week (cutover and rollback window)none
Verification1 week1 week1 day
Measured lead timeabout 10 weeksabout 21 weeksabout 2 weeks

Notice that delivery, the number everyone quotes, is less than half of the total for the shelf and about half for the circuit. The rest is your own organisation's process, and it does not appear on any supplier's website.


Time-to-full versus time-to-buy


Capacity planning is a race between two clocks. Time-to-full is how long until the resource reaches its limit, from the forecast. Time-to-buy is the lead time, from the chain above. If time-to-full is longer than time-to-buy, you have a plan. If it is shorter, you have an incident with a delivery date. The whole point of the register is to compare the two clocks every month. Start the chain while the first is still longer than the second.


A disk with four months of headroom is comfortable if the shelf takes ten weeks and alarming if it takes twenty. A circuit with a year of headroom is comfortable until you learn that the last circuit took twenty-one weeks plus a freeze. Time-to-full comes from the forecast. Time-to-buy comes from the purchasing records. The register is where they meet.


The trigger point and the safety margin


The trigger point is the date you must start the chain. It is arithmetic: the projected full date, minus the lead time, minus a safety margin. The safety margin is the extra time you allow because both of the other numbers are estimates. Both tend to err in the same unhelpful direction. Forecasts run early more often than late, because new load arrives more often than load disappears. Lead times run long more often than short, because nobody has ever been surprised by a delivery arriving early.


trigger date = projected full date - lead time - safety margin

safety margin = the larger of:
   one month
   (most likely full date - earliest full date from the forecast range)
plus any known freeze or holiday period inside the lead time

Setting the margin is a judgement, but it is a bounded one. Start with one month. It covers the meeting that gets cancelled and the delivery that slips a fortnight. Widen it by the uncertainty in the forecast. If the forecast says "between March and June, most likely May", the gap from the earliest date to the likely one is two months. So the margin becomes two. Then add any period in which nothing moves, such as a change freeze or the weeks when the approver is away. Write the margin in the register with its reason, so the next person does not shrink it because it looked generous.


If the trigger date is in the past when you compute it, the arithmetic has told you something important. In that case, you are already late. Do not recompute it with a smaller margin to make it come out in the future. Start the chain today, and read the pain-phase section below to plan for the gap. In a spreadsheet, put the projected full date in D4, the lead time in days in D6, and the margin in days in D7. The trigger is =D4-D6-D7, formatted as a date. A conditional format that turns the cell red when it is earlier than TODAY() is the cheapest early-warning system you will ever build.


Putting the trigger into monitoring


A trigger in a spreadsheet fires only when someone opens the spreadsheet. A trigger in the monitoring system fires on its own, and that is where it belongs. There are two ways to express it, depending on what your tool can alert on.


The first way works everywhere: convert the trigger date into a utilisation level. The trend tells you how much use to expect on the trigger date. That value divided by the limit is a percentage the monitoring system can watch. Using the slope per day from the forecast:


use at trigger   = use today + (slope per day x days until trigger)
trigger percent  = use at trigger / limit x 100

Example (invented numbers):
  use today 1,620 GB, slope 1.5 GB/day, trigger in 90 days, limit 2,000 GB
  use at trigger  = 1,620 + (1.5 x 90) = 1,755 GB
  trigger percent = 1,755 / 2,000 x 100 = 87.75, so alert at 88%

Set a threshold at that percentage on the daily peak of the resource. Use a "sustained for" condition of several days so a single busy afternoon does not fire it. Route the notification to the register owner rather than to the on-call phone. This is a planning alert, not an incident. The guide on static thresholds done right covers how to keep it from becoming noise. The threshold must be recomputed whenever the forecast changes, so put that in the monthly register hour.


The second way is available in some monitoring tools: an alert on the projected days-to-full itself. The projection is computed by the tool from its own trend. If yours has it, the rule is simply "alert when days to full falls below lead time plus margin". Check the product documentation for whether the projection is built in and what window it fits over. If the window is short, it will inherit every seasonal lie the forecasting guide warns about. Either way, keep the spreadsheet trigger as well. Two independent reminders of the same date cost nothing and have saved me twice.


Tip: name the planning threshold so its purpose is obvious in the alert subject: "CAPACITY TRIGGER fs01 D: order storage now". An alert that says "disk 88%" at two in the afternoon gets triaged as an incident and closed as "not urgent". That is exactly the wrong outcome.


The pain phase and what it costs


The pain phase is the period between the resource becoming a daily problem and the new capacity arriving. It is what you are paying the safety margin to avoid. It has a predictable shape. First comes the clean-up: an evening spent finding and deleting the largest files you can justify removing. That buys a few weeks and one complaint. Then the daily check, in which someone reads the free-space number every morning like a weather report. Then the workaround: a temporary volume, a redirected share, a throttled backup, each of which becomes a permanent feature of the environment. Then the explanation meeting.


The cost is not mainly money, though rushed purchases are rarely cheaper. The cost is effort and attention. Every hour spent nursing a full disk is an hour not spent on the thing you were supposed to be doing. The nursing lasts the entire lead time, because starting the chain late does not shorten the chain. Meanwhile, the incident that finally happens gets a review. The review discovers a graph that had been predicting the date for months. I have sat in that review. The graph is always very clear in hindsight.


If you are already in the pain phase, start the chain today. Then buy time with the cheapest reversible measures. Then tell the people who will be affected what the date is. Do not spend the lead time optimising the workaround. The workaround is not the project. The delivery is.


Renewals and licence counts are capacity too


Licences fill up in two directions, and both have lead times. A licence count is a slot resource. The product allows two hundred devices, you have a hundred and eighty-six, and every discovered switch takes one. The licence count's trend comes from device growth. Its limit is the count you bought. Its lead time is the quote-to-key chain in the table above. The guide on licence growth traps explains how counts creep and where the tier boundaries bite. For planning purposes, a licence count is a row in the register like any other.


A renewal is capacity with a fixed full date. The support contract, the subscription, the certificate: each expires on a day already printed on a document. After that day the thing stops working or stops being supported. The forecast is trivial, since the full date is known. The lead time is the entire difficulty. Renewals that need approval, a purchase order, and a supplier's processing time need the same trigger arithmetic as a disk. Put every renewal in the register with its expiry as the full date, its measured lead time, and a margin. Then the licence that "just expired" stops being a category of incident. The licence tracking guide covers keeping the inventory that feeds this.


The register with trigger dates


The starter register from the basics guide had a trigger column. This version expands it so that the trigger is traceable. The register shows where the date came from, how wide the margin is and why, and where the purchase currently sits in the chain. The status column is the one people look at in the monthly hour, because it tells them what is moving and what is stuck. Every number below is an invented example.


ResourceFull (likely)Full (earliest)Lead time (measured)Margin and reasonTriggerStatusOwner
fs01.example.com D:8 months out7 months out10 weeks6 weeks: 1-month floor + forecast gap4 months outWaitingJ. Ortiz
Head office WAN14 months out11 months out21 weeks4 months: forecast gap + year-end freeze5 months outWaitingJ. Ortiz
Core switch access ports4 months out3 months out5 weeks1 month: floorPastTriggered, quote requestedM. Adeyemi
Monitoring device licences5 months out4 months out2 weeks1 month: floor3 months outWaitingJ. Ortiz
Firewall support renewalexpiry dateexpiry date6 weeks1 month: floor10 weeks before expiryOrdered, awaiting keyM. Adeyemi

Use exactly five status words, in this order, and move each row forward as the chain progresses: Waiting, Triggered, Ordered, In service, Verified. A row that stays at Triggered for longer than the quote link should take is a stuck purchase. The monthly hour is where you chase it. A row that reaches Verified goes back to Waiting with a new forecast, because the loop never ends. It only gets calmer.


The circuit that lost a race with a disk


Bellweather Logistics had two forecasts in the same month. The depot file server would be full in five months, and the depot's WAN link would be full in seven. The disk seemed more urgent, so the disk was ordered first. It arrived in three weeks and was in service in four. The circuit was ordered when the file server was safely sorted, with what felt like plenty of time. The carrier's survey then found the duct into the building was full. The install would take five months after the order. The link hit its limit a month before the new circuit went live. For those four weeks the depot's afternoon transfers stalled so badly that drivers carried drives between sites in their cabs. The disk never once got close to full. The circuit was the emergency all along, because nobody had compared time-to-buy with time-to-full.


Checklist


  • For each kind of resource you buy, pull the dates of the last three purchases from the purchasing and change systems. Measure the lead time from request to in service.
  • Record each link of the chain separately, so you know which one is slow.
  • Ask procurement and the approver when the approving meeting sits and whether any freeze periods are coming.
  • Compute the trigger for every register row: projected full, minus lead time, minus margin.
  • Set the margin at one month or the width of the forecast range, whichever is larger, plus any freeze, and write down the reason.
  • Convert each trigger date to a utilisation percentage and set a planning threshold in the monitoring system, routed to the register owner.
  • Name the planning alert so it cannot be mistaken for an incident.
  • Add every renewal to the register with its expiry as the full date.
  • Give every row a status word and chase any row that sits at Triggered for longer than a quote should take.
  • Recompute triggers and thresholds in the monthly register hour whenever a forecast moves.

Where to go next


You now have both clocks. The forecast gives time-to-full. The purchasing records give time-to-buy. The trigger is where they meet, with a margin for the month that always goes missing. The rest of Capacity Planning From Monitoring History continues from here. The guide on capacity beyond bandwidth lists the resources nobody thought of as capacity. The guide on justifying an upgrade with graphs turns a triggered row into the one-page proposal that starts the approval link. Start with the purchasing records this week. The lead time is already in there. It is just waiting for someone to subtract two dates.


Put this into practice with SysPeep Monitor

A trigger only helps if something fires it. The second half of this guide depends on a threshold on utilisation that reaches the right person. SysPeep Monitor lets you set thresholds and alert rules on the metrics it collects over SNMP, WMI, and SSH. It sends them as email or other notifications. It keeps the long-term graphs and reports that give you the projected full date in the first place, all from one web console. The free trial installs on any Windows machine.

Download Free TrialLearn About SysPeep