Practical guide

How to design a standard operating procedure with AI

Use AI to organize an SOP, not to invent the safe way to perform a task. Begin with the governing policy, an observed process map, named controls, and the needs of the people who perform and supervise the work. The useful result is a testable SOP draft with prerequisites, ordered actions, decision points, stop conditions, owners, and a revision trail.

By Two Prune

Use AI to organize an SOP, not to invent the safe way to perform a task. Begin with the governing policy, an observed process map, named controls, and the needs of the people who perform and supervise the work. The useful result is a testable SOP draft with prerequisites, ordered actions, decision points, stop conditions, owners, and a revision trail.

Decide whether an SOP is the right control

A procedure is only one part of a work system. Before drafting, identify the task, the consequence of failure, existing technical controls, required competence, and the situations in which work must stop. The UK Health and Safety Executive advises that risk assessment should inform procedure design and warns against relying on procedures as the sole defence against human error.

Create a one-page design record:

| Question | Synthetic warehouse example | | --- | --- | | Task | Restore a stalled packing line after a sensor fault | | Users | Shift technician and line supervisor | | Preconditions | Line stopped, fault code recorded, area controlled | | Decisions | Reset permitted or maintenance escalation required | | Primary failure | Restart while a person is inside the guarded area | | Independent controls | Interlock, isolation device, supervisor authorization | | Evidence of completion | Reset log, test cycle result, owner sign-off |

Ask AI to identify missing fields and inconsistent terms. Do not ask it to determine the actual isolation method or whether a control is sufficient. Those decisions belong to qualified owners using the applicable rules, equipment information, and risk assessment.

Convert the real process into decision steps

Walk the task with the people who do it. Record normal flow, variants, handoffs, unavailable-system cases, and recovery paths. Then separate three kinds of content: actions the user performs, conditions the user checks, and information that explains why a condition matters.

For the synthetic line-reset example, a compact sequence might be:

The text deliberately does not supply equipment-specific safety instructions. A real SOP must obtain those from the responsible organization. HSE guidance also recommends using task analysis, involving users, matching detail to the task and consequences, and placing warnings immediately before the relevant action.

  • Record the displayed fault code and current line state.
  • Confirm the guarded area is clear using the site-approved check.
  • If the code is on the authorized reset list, request supervisor authorization; otherwise stop and contact maintenance.
  • Apply the approved reset action for that equipment model.
  • Run one empty test cycle and compare the result with the acceptance condition.
  • If the fault returns, stop, preserve the log, and escalate. Do not repeat the reset.

Mark ownership and evidence inside the procedure

Give each consequential step an owner and an observable completion record. “Check the system” is not testable. “Line supervisor confirms the guarded area check and records initials in the reset log” can be observed and reviewed.

AI can transform approved source notes into a consistent table, flag undefined acronyms, compare terminology with a process map, and produce a change summary between versions. It should not silently reconcile conflicts. If the policy calls a role “maintenance controller” and the process map says “engineering lead,” the draft should surface the mismatch for resolution.

Test edge cases before approval

Run a tabletop walkthrough with at least one operator and each control owner. Test normal completion plus cases such as missing authorization, a different equipment model, an unavailable log, an interrupted handoff, and a repeated fault. For every branch, ask whether the user knows what to do, who decides, what evidence to retain, and when to stop.

A common failure is a polished generic instruction such as “ensure all safety checks are complete.” It hides which checks apply, who owns them, and what happens after a failed check. Repair it by naming the approved check, decision owner, record, and stop condition. If those facts are unknown, the procedure is not ready.

Before release, verify that:

  • the procedure reflects the observed task and current approved controls;
  • users and control owners walked through normal and exceptional paths;
  • prerequisites, warnings, decision points, and stop conditions are visible;
  • equipment, role, and record names match their source systems;
  • every critical action has an owner and completion record;
  • the revision owner and review trigger are named.

Sources and scope

This is an educational drafting method. It is not equipment-specific instruction, a risk assessment, or a determination that a procedure makes work safe.

  • UK Health and Safety Executive, Procedures: used for procedure design principles, task analysis, user involvement, risk context, and the limits of procedures as controls.
  • UK Health and Safety Executive, Operating procedures: used for guidance on prerequisites, role clarity, warnings, consistent terminology, quality control, and keeping procedures current.

Sources: [1] [2]

Sources

  1. 1.Procedures · www.hse.gov.uk
  2. 2.Operating procedures · www.hse.gov.uk