Three Ways to Handle Patching, Judged Like Maintenance Plans

Three Ways to Handle Patching, Judged Like Maintenance Plans

Anyone who buys, stocks or ships aircraft parts has a strong instinct for maintenance philosophy. Fix it when it breaks, replace it at fixed intervals, or watch its condition and act on evidence: these are three distinct schools, and each has a cost profile that shows up in the stockroom. The same three schools exist in the way organisations deliver software fixes, and putting them next to each other is the fastest route to understand what PatchOps is really for. This article sets the strategies against each other using criteria a parts business already trusts: cost of surprise, predictability, evidence and effort.

A note on terms before we start. The Ops Atlas has an entry on PatchOps as the Atlas defines it, and it is honest that the label is informal, closer to a nickname for enterprise patch management than to a standard. The three strategies below are my own framing for a maintenance-minded audience, not categories taken from that page, though the Atlas supplies the details I lean on.

Strategy one: repair on failure

In hardware terms this is run-to-failure. You leave the component alone until it stops working. For a cabin light that is sensible. For a flight control it would be unthinkable.

In software, repair on failure means updates are applied when someone notices a problem, or when a machine happens to update itself. There is no calendar, no inventory discipline and no proof. The appeal is low effort today. The cost arrives later and in lumps: a flaw sits open for months, then everyone scrambles at once.

The Atlas offers a widely cited cautionary case. In 2017 a large credit-reporting firm was hit after a flaw in a web framework went unaddressed on one portal; the official review said the flaw “was not properly identified as being present” on that system. The lesson is not that the firm was careless in general. It is that a single unlisted machine behaved like a part nobody knew was fitted. Without an inventory, repair on failure becomes repair after exposure.

Strategy two: calendar-based replacement

This is the fixed-interval school: every part comes off at a set number of hours or months whether or not it looks fine. It is predictable, easy to budget and easy to audit.

The software equivalent is a regular cycle: updates released on a known day are applied in a known window. The Atlas describes a monthly rhythm in which the first step is triage on release day, followed by a daily look at rollout gates, a weekly nudge for laggards, then a monthly write-up and a plan check every quarter. For a parts counter used to cycle-time planning, that will feel familiar.

The weakness is the same as in hardware. Calendar logic treats every unit alike. An update that suits the office desktop may be risky on the machine that runs the warehouse scanner, and a fixed date does not know the difference. A purely scheduled approach also tends to count a job as complete when the installer has run, not when the fix is verified working.

Strategy three: condition-based and staged delivery

Condition-based maintenance watches evidence: vibration, wear, trend data. You act when the data says so, and you test changes on a small subset first. The software parallel is a staged rollout with measured gates. A tiny first ring tries the update. If the install failure rate stays under a threshold, nothing new breaks and enough machines are reporting, the next ring goes. If a gate trips, the rollout halts.

The Atlas gives example gates from a commonly used management product: a failure ceiling on installs of 2%, 90% or more of devices still checking in, and zero fresh incidents. It also describes the staged ring idea itself, from canary groups through pilot and broad stages to a sensitive final group handled node by node. Real organisations tune those numbers; the point is the habit of letting data decide when to move on.

The three on one scorecard

Scoring them against the same four yardsticks gives this picture.

  • Cost of surprise. Repair on failure carries the highest, because problems surface during operations. Calendar cycles lower it. Staged, condition-driven delivery lowers it further because bad changes are caught on a few machines.
  • Predictability. Repair on failure has none. Calendar delivery is very predictable. Staged delivery is predictable in rhythm but flexible in order, since a halted ring delays later ones.
  • Evidence. The first strategy produces almost none. The second produces completion records. The third produces outcome records: which machines work after the change.
  • Effort. Lowest for the first, moderate for the second, highest at the start for the third, though automation brings it down.

Where the numbers mislead

All three strategies can be undermined by the way success is counted. The Atlas walks through a small lab exercise: fifteen patchable machines showed a dashboard score of 100 against a truthful 53, because the dashboard counted expired exceptions as fine and ignored a machine whose reporting agent had gone stale. Anyone who has reconciled stock records with a physical count will recognise the pattern. A ledger can say a bin is full while the shelf is empty.

The Atlas also flags pending restarts as a leading cause of false compliance and points to coverage that plateaus in the high eighties because a few machines never come online to receive the fix. Those are the unreachable units of the software world, the equivalent of an aircraft that is on the books but away in another country.

What the sourcing mindset adds

People who buy parts have a useful reflex: they check applicability before they order. Which tail number, which serial range, which revision. That reflex is exactly what staged delivery needs, and it is missing from lazy approaches. The Atlas puts applicability at the centre of the discipline: knowing which machines a given update actually affects before deciding where it goes.

The same attitude shows up in procurement advice, such as our notes on common mistakes when ordering avionics parts, where verifying fit and paperwork before purchase is the recurring theme. Hardware care and software care are siblings. The buyer who insists on traceability will understand why an inventory and a record matter, and our rundown of aircraft fasteners and hardware shows how small, unglamorous items do the most damage when nobody tracks them.

Trade-offs the staged approach cannot hide

It would be dishonest to present the third strategy as free. Staged delivery needs a good inventory, working reporting and someone with authority to halt a ring. It slows the last ring by days or weeks, which matters when a threat is active. The Atlas handles that with an emergency lane: a faster path with a named approver and a short test on a canary group, measured in minutes to hours rather than weeks.

Rollback is another limit. In maintenance you can often reinstall the previous part. In software the practical route is frequently to move forward with a corrected update, and the Atlas notes some tools offer no undo at all. That is a reason to test early in small rings, not a reason to skip testing.

Which strategy fits which operation

A one-person parts shop with two laptops does not need ring gates. A fixed monthly calendar, a written list of devices and an honest restart check will cover most of the risk. A mid-sized distributor with warehouse scanners, a customer portal and several offices benefits from the staged model because a bad update on a shared system can stop shipments for a day. A large operation with regulated customers should add the measurement habits: median days until a fix is truly active, how many machines went silent, and the age of pending restarts.

In every case, keep two rules. Never let a dashboard define done without checking the denominator, and never leave an exception without an expiry date. Those two controls cost nothing and catch the most common failures from all three strategies.

Bottom line

Repair on failure is the cheapest this month and the most expensive over a year. Calendar cycles bring order but not proof. Staged, evidence-led delivery brings both, at the price of some setup. For a business that already respects traceability in its stockroom, the third approach is the natural extension to its computers, and the comparison above should make the case for taking the first small step: write down every device, pick a window, and count honestly.