Choreography: the living SOP
The procedure you wrote once and never updated again doesn't have to die in a drawer. This lesson shows the ladder that turns a stalled SOP into a procedure that updates itself with reality, always with human approval before anything changes.
The receiving SOP was written two years ago by an analyst who no longer works here. It describes a process with three manual checks. Except in practice, the team found a faster way to check using the barcode reader, and nobody updated the paper. Today, whoever is new reads the SOP, tries to follow it to the letter, gets stuck, and asks the coworker next to them "is this still how it's done?". The answer is almost always no. The SOP isn't wrong by accident: it's a snapshot of a moment that passed, and nobody took a new photo.
The store's cash closing procedure still asks for a paper double count, but the team already uses the POS app that adds it up automatically. Nobody updated the text, and every internal audit flags the same "non conformity" that in practice was already solved a different way, just never documented.
The contract approval checklist still requires a physical signature the company stopped asking for once it adopted electronic signatures. The formal procedure, if followed to the letter, would delay every contract by two days for a step that no longer exists in practice.
The ad asset approval guide still describes a flow with three sequential reviews the team abandoned a while ago, because today they approve in a single meeting. Whoever is new loses a whole day trying to follow a step by step nobody follows anymore.
The offboarding SOP still says to print a form the HR system now fills out on its own. The team actually uses the system, but the compliance auditor only sees the written paper, and every review turns into a discussion about a step that no longer exists in practice.
The feature launch guide still describes a manual QA checklist the team swapped for automated tests three months ago. Whoever follows the guide to the letter loses a day testing by hand what already runs on its own, and nobody flagged the step as obsolete.
The lead qualification script still asks a budget question the team stopped asking, because the CRM already pulls that data from the website form. A new salesperson asks the customer again, who already answered that, and the script has stayed the same for a year.
At the distribution hub, the order picking SOP still tells the operator to check each item against a printed list. It's been eight months since the team switched to the scanner with automatic reading, which already checks on its own, and the paper still says to print the list. Every new shift learns the wrong way first, because that's what's written, and only afterward does someone experienced teach the real way. A manager asked AI to compare last month's execution records against the SOP's text, and it found: in 91% of picks, the "print list" step simply doesn't happen anymore. The official procedure describes work nobody has done in months, and it's still what gets taught to the new team.
The data incident response procedure still names a ticketing system the company replaced six months ago. If a real incident happens tomorrow, the official procedure's first step points to a tool that no longer exists.
The deploy runbook still describes three manual steps the CI/CD pipeline automated a while ago. In the middle of the night during an incident, the on call engineer follows the runbook to the letter, wastes twenty minutes running a manual command, and only afterward remembers that it already runs on its own.
The usability testing guide still asks for recording in software the team switched vendors for two cycles ago. Whoever is new tries to install the old tool, can't find it, and loses the morning until someone explains the process changed and the paper didn't keep up.
The quarterly goal review procedure still asks for a spreadsheet the team swapped for an automatic dashboard two quarters ago. Every cycle, someone new tries to fill out the old spreadsheet, and the time lost on that outweighs the time it would take to explain the dashboard.
Everyone's lived this: you open the official procedure to do something you did last month, and it describes a way of working nobody uses anymore. The team found a shortcut, a new system came in, an exception became routine, and the paper stayed frozen on the day it was written. Think about it: how much time have you already lost following a procedure that was wrong not because it was badly written, but because it was out of date?
The core idea of this lesson. Every SOP written once goes stale on its own, because the operation changes and the paper doesn't. The way out isn't writing the perfect SOP, it's building a system that detects when real practice has drifted from the written procedure, and tells you. You climb a ladder: from by hand, to a fixed template, to a workflow that compares real execution against the text and flags the drift, up to an SOP that receives the update draft almost on its own. But there's one rung that never gets automated: approving a change to the official procedure is always a person, because a wrong SOP in operations isn't just a broken report, it's the risk of an accident and of a quality failure.
01Why the SOP never gets updated
Every SOP has three parts, and confusing them is the root of the problem. The fixed structure: the same sections, purpose, steps, owners, exceptions, this practically never changes. The real practice: what the team actually does today, which changes all the time, new equipment, new system, a new exception that became routine. And the reading of what changed: why practice drifted from what's written, and what to do about it. This third part is yours.
An SOP written once is a photograph. The operation is a video. Nobody takes a new photo every month, so the photograph keeps getting further from the video, until it stops being a guide and becomes documented fiction: everyone follows what actually works in practice and pretends the paper doesn't exist, until the day an auditor, a new hire, or an incident investigation goes looking for the official procedure and finds it describes a different company.
02The ladder: four rungs from a dead SOP to a living SOP
Think of a four rung ladder, each one costing less maintenance effort than the one below it.
Rung one is where most SOPs live today: a stalled document, revised only when someone remembers or an auditor demands it. The next three rungs are the subject of the following two sections. Notice the design: maintenance effort drops at each rung, but who approves the procedure's change stays at the top, always.
03Rungs two and three: the template and the workflow that detects drift
Rung two is the template. You stop rewriting the structure from scratch every time you need to revise an SOP. A good SOP template fixes: the sections in the right order (purpose, numbered steps, who's responsible for each step, what counts as an exception and what to do about it), the expected level of detail, and where the date of the last real revision gets logged. That alone already cuts the friction of keeping the procedure up to date, because editing an existing section is fast, writing a new document isn't.
Rung three is the drift detection workflow. This is where AI comes in with real force. You feed it recent execution data (shift records, tickets, the logs of the system the operation actually uses) and the current SOP's text, and ask it to compare the two: "flag every step in the SOP that real practice no longer follows, and every step practice does that the SOP doesn't describe". The output is a concrete list: "the SOP says to print the picking list; over the last thirty days, 91% of picks used automatic scanning, with no printing". This isn't opinion, it's comparing data against text, and AI does this fast, at a volume nobody would review by hand every month.
04Rung four: almost one click, never one click
The top of the ladder is where the SOP starts receiving, on its own, an update draft whenever the detected drift is large and stable (not an isolated case from one bad shift, but a pattern that repeats). The workflow runs, flags the drift, and already proposes the new text for the section that changed.
The rule is strict: you automate detecting the drift and drafting the update, never approving the new procedure. In operations, this isn't excessive caution. An SOP describes, among other things, how to handle equipment, how to deal with hazardous cargo, how to act in a safety exception. A procedure that changes on its own, with no accountable owner reviewing every word, isn't agility, it's an accident waiting for the right moment. AI flags what changed and proposes the text; the person accountable for the operation reads it, adjusts it, and signs off.
05The math that justifies keeping the SOP alive
Do the simple math. Every time someone follows an outdated SOP, or wastes time trying to understand why the paper doesn't match practice, or worse, makes a mistake because they followed a step to the letter that already changed, that costs time and, sometimes, costs an incident. Multiply that by the number of people who go through that procedure in a year, new hires, new shifts, audits. Add it up, and the number is usually big enough to justify the few minutes a month rung three takes to run.
Keeping the SOP alive isn't extra bureaucracy. It's the difference between a procedure that protects the team and one that only exists to pass an audit while everyone, in practice, works a different way under the table.
Do it now
Take a real SOP you use, your real task or another one (receiving, picking, support, deploy, whatever it is).
- Write its structure as a template: which sections, in what order, who's responsible for each step, what counts as an exception.
- Calculate the cost of it being outdated: how many times a month does someone follow a step that already changed in practice, and how much time that costs each time. Multiply by a year.
- Describe, in one sentence, how you'd ask AI to compare recent execution data (tickets, shift records, logs) against this SOP's text, to find where practice drifts from what's written.
- Mark the red line: who, by name or by role, is the only one who can approve a change to this procedure's official text. Without that person's approval, the SOP doesn't change, no matter how confident AI is about the update.
Practice
1. What's the correct ladder sequence to turn a stalled SOP into a living SOP?
2. At the ladder's highest rung, when the SOP already receives the update draft almost on its own, what never gets automated?
For the board
On the frozen documentthe team found the shortcut, the system changed, the exception became routine, and the procedure stayed on the day it was written.
On the laddereach step shortens the distance between the paper and the practice. Approving the change stays at the top, with a person.
On the riskin operations a wrong SOP is not a flawed report. It is an accident or a quality failure.
Thanks for the feedback. It helps sharpen the next lesson.