Procedural Drift: When the Written Process Stops Matching the Work

Updated on: 2026-08-28

What the written process misses

Most organisations run two versions of the same process.

There is the official one: the policy, the workflow, the checklist, the approved steps in the shared drive.

Then there is the version people use when the day gets busy. The shortcut learned from a colleague in the first week. The spreadsheet that sits outside the system because it is easier that way. The approval given over a quick message because the formal route takes too long.

None of this starts as defiance. It starts as people trying to get their work done.

That is why Procedural Drift is easy to miss. No one calls a meeting to announce the process has changed. It moves through an organisation in small ways: one altered handover, one skipped check, one workaround good enough to be taught to the next person.

Eventually the organisation has a documented process and a practised one, and they are no longer quite the same thing.

What Procedural Drift is

Procedural Drift is not about whether the organisation has good policies. That is a governance question.

It is about whether the steps people actually take still match the process those policies rely on.

A procedure exists to turn intent into action. It tells people what happens next, who owns a decision, when something needs escalating, what gets recorded, and where accountability sits. It makes work repeatable enough for a colleague, a new starter, an auditor or a partner team to follow.

When that shared route starts splitting into local versions, the risk is not always visible straight away. The work can even look faster. A team handles a case in half the time because someone knows which stage can be skipped. A manager approves an exception with a quick yes rather than asking for the usual evidence. A colleague keeps a private tracker because the official platform does not match the reality of the job.

Each choice can make sense on its own. Together, they leave an organisation unable to say with any confidence how a process really works.

This pattern has a name in organisational research. Harvard researcher Scott Snook studied how U.S. military units, all following sound orders, gradually drifted from a formally agreed no-fly procedure toward locally sensible shortcuts, and called it practical drift: the slow uncoupling of local practice from written procedure [1]. Procedural Drift describes the same mechanism at work in an office rather than an operations room. Reasonable people, reasonable adjustments, and a growing distance between the policy and the job.

3. Why it is easy to miss

One of the clearest signs shows up during handovers.

Someone leaves. Someone moves teams. Someone is off unexpectedly. A new person opens the documented procedure, follows it carefully, and discovers halfway through that it does not quite get them where everyone else ends up.

The missing detail is rarely deliberate. It is just held somewhere else: a manager's judgement, a colleague's inbox, a chat thread from months ago, a spreadsheet with a name that tells you nothing, the memory of whoever has “always handled that bit.”

The organisation has not lost the process. It has lost the shared version of it.

That distinction matters. When know-how sits with a few people instead of the organisation, routine work gets fragile. Onboarding takes longer. Outcomes vary. Investigations slow down because no one is quite sure which version of events reflects what actually happened.

There is a reason this tends to happen quietly rather than as one dramatic failure. Researcher John Banja, writing on patient safety, described how small deviations from a standard stop registering as deviations once they have been repeated often enough without obvious harm, a process he called the normalisation of deviance [2]. A workaround does not need approval to become permanent. It just needs enough repetition and enough silence.

How AI accelerates Procedural Drift

AI can speed this up, not because it is inherently disruptive, but because it quietly redistributes work.

A team starts using a tool to draft a response, prepare a summary, route a query or flag a priority. At first it saves time. Then it becomes part of the rhythm. The original human step stays on the process map, but in practice it gets lighter, later, or purely symbolic.

None of this started with generative tools. A 2022 review of workaround research in electronic health record systems found that almost every new piece of workplace technology produces its own fresh set of workarounds, each with a different cause: a step that is too complex, a gap in training, a habit that outlived the reason for it [3]. AI is simply the newest system doing this, faster. A 2026 analysis of enterprise AI rollouts found that most organisations had not documented their workflows in enough detail for an AI system to run them without a person quietly filling the gaps, which means the tool often ends up automating whichever version of the process it happens to be shown rather than the one written down [4].

That change might be perfectly reasonable. It might even improve the process. But it still needs naming. Who checks the output now? What does review actually mean in practice? Where does the decision sit if the tool is uncertain or wrong? What gets recorded when someone overrides it? Has the procedure itself been updated to describe what is really happening?

These aren't questions about whether AI is good or bad. They are questions about whether the organisation still has an honest, usable account of how its own work gets done.

Early signals to watch for

You do not need an incident to spot this. Start with the ordinary moments.

• Three people explain the same task three different ways, each one capable and sincere, each one on a slightly different route without realising it.
• The formal process gets pulled out for audits, not used for the actual work.
• A temporary workaround has quietly survived long enough to become tradition, and no one has decided whether to approve it, fix it, or retire it.

None of this is about stamping out adaptation. A procedure that never changes will stop serving the work fairly quickly. The point is making adaptation visible before it turns into dependency on knowledge nobody wrote down.

A practical way to check for it

Pick one process that matters: a customer complaint, an onboarding journey, an incident response, an approval route, an AI-assisted decision. Ask whoever does it every day to talk you through it, step by step. Then hold that account up against the written procedure.

Skip “are you following the process?” and ask instead:

• Where do you need to use judgement?
• What do you do when the standard route does not fit?
• Which part takes longer than it should?
• What have you changed to make the work possible?
• What would a new colleague need to know that the document does not say?
• Which steps have changed since AI or automation entered the workflow?

That conversation tends to be worth more than any compliance reminder. It shows the distance between a process designed at one point in time and the work people are actually doing now. Some of what you find may deserve to become the new formal process. Some may need a safeguard. Some may just show that a system or workload is creating friction that no one has named yet.

Why this matters for AI governance

Procedural Drift produces a particular kind of false reassurance. The policy exists. The procedure is filed. The training record is complete. The audit trail looks tidy.

But the actual work may be moving through informal channels, patchy checks and decisions nobody wrote down.

That is why it matters for AI governance specifically. An organisation can have a solid policy on oversight, accountability or human review while the real workflow tells a different story. A review step gets rushed. An override happens informally. A tool ends up doing work that no longer has a named owner. None of that shows up on a dashboard. It shows up when people compare notes, when something gets investigated, or when the one person who “just knows” how it works is no longer around to ask.

Where to start this week

You do not need to rewrite every procedure this week. Start with one process where outcomes differ between people or teams, where workarounds are common knowledge but rarely discussed openly, where a handover would be genuinely difficult tomorrow, where AI has changed a meaningful step, or where nobody has reviewed the procedure since the work itself changed.

Map the real route. Name the exceptions. Decide who owns the procedure. Write down why a change happened. Give people a straightforward way to say, “this isn't how the work happens anymore.”

That is not bureaucracy. It is how an organisation keeps its ability to work reliably while it keeps adapting.

Reflection questions

Before the week wraps up, sit with these:

1. If you compared the documented process with what actually happens, how far apart would they be?
2. Which workarounds have quietly become “the way we do it”?
3. Would this process survive the person who currently runs it leaving tomorrow?

A procedure that only exists on paper is not a procedure. It is a record of what the organisation used to do.

About the AIR Clarity Check

The AIR Clarity Check is a complimentary 30-minute conversation through The Clarity Line, a space to look at where your organisation is adapting well, where practice has moved beyond documentation, and how to bring adaptability, innovation and resilience back into the same picture.

Book a complimentary AIR Clarity Check: Book your free 30-minute conversation

References

Snook, S. A. (2000). Friendly Fire: The Accidental Shootdown of U.S. Black Hawks over Northern Iraq. Princeton University Press. https://press.princeton.edu/books/paperback/9780691095189/friendly-fire

Banja, J. (2010). The normalization of deviance in healthcare delivery. Business Horizons, 53(2), 139. https://doi.org/10.1016/j.bushor.2009.10.006

Blijleven, V., Hoxha, F., & Jaspers, M. (2022). Workarounds in electronic health record systems and the revised sociotechnical EHR workaround analysis framework. Journal of Medical Internet Research, 24(3), e33046. https://doi.org/10.2196/33046

Estes, P. (2026). Why AI automation strategy fails: 84% haven't documented their workflows. Virtasant. https://www.virtasant.com/ai-today/ai-automation-strategy-starts-with-your-sop

About the Author

CKC Cares | Community, Tools & Services supports practical capability for ethical, human-centred leadership. Our work helps leaders turn thoughtful governance into everyday operational action, with a focus on digital resilience, decision clarity and the human realities of AI-enabled change.

Disclaimer: This article provides general educational information on digital governance and leadership practices. It does not constitute legal, regulatory, or professional advice. Leaders should apply these ideas in alignment with their internal policies, operational context, and risk responsibilities.

The content in this blog post is intended for general information purposes only. It should not be considered as professional, medical, or legal advice. For specific guidance related to your situation, please consult a qualified professional. The store does not assume responsibility for any decisions made based on this information.