Why operational memory matters

Reusing expert judgement across sites and shifts requires more than storing documents. It requires lessons people can trace, challenge and apply to the work in front of them.

A contractor submits risk assessments and method statements (RAMS) for maintenance work. The document looks complete. An experienced reviewer notices that the emergency arrangements refer to another site.

The reviewer asks for a correction. The work can then proceed through the organisation's approval process. But what happens to the lesson? If it stays in an email exchange, another team may overlook the same problem next week.

The organisation has corrected one document. It has not necessarily improved the next review.

From a stored record to a usable check

Reports, procedures and previous reviews preserve information. Operational memory connects relevant information to the moment someone needs to act on it.

In this article, operational memory means retaining practical checks, the evidence behind them and the conditions under which they apply. It may include reasons for returning a contractor document, escalation criteria, examples of acceptable work or lessons from an incident.

AI can help retrieve those lessons and compare new material against approved checks. The organisation still needs to decide what is authoritative, who can use it and when it should be reviewed.

How a lesson becomes operational memory

A useful memory needs more than a saved AI response. It needs a route from observation to approved guidance, followed by review as circumstances change.

1. Capture the finding and its evidenceRetain the relevant extract, the reviewer's observation and the action taken. Separate what was observed from what was inferred.
2. Verify the lessonA responsible person checks whether the finding is valid and whether it reflects a recurring issue, a specific circumstance or an individual preference.
3. Define where it appliesRecord the relevant work, sites and conditions. A lesson from one location should not silently become a requirement everywhere.
4. Approve and make it retrievableAssign an owner and version. Connect the check to the relevant workflow so it can be surfaced with its source and limitations.
5. Review, revise or retire itRevisit the lesson when procedures, site conditions or evidence change. Feedback from later use should inform a reviewed update.

Suggested operating model. Approval and review requirements should reflect the consequences of the workflow.

What another reviewer should be able to see

Suppose a team at another site receives a similar contractor document. A useful output would identify the relevant check, show the supporting text and make any missing information visible.

The following example shows the information a reviewer would need to assess a finding. The wording deliberately asks for verification rather than declaring the work safe or unsafe.

This is how a lesson can travel without losing its context. Another team can use the check while still examining whether it applies to their work and location.

Retained judgement still needs scrutiny

Experience is valuable, but it is not automatically correct. A long-standing practice may be outdated. A precaution may be appropriate for one site and excessive or insufficient for another. A previous decision may have been made with incomplete information.

Operational memory should therefore distinguish formal requirements, approved internal guidance, local practice and lessons still under review. Presenting all four as equally authoritative would make the system harder to trust.

Access matters too. Reusing a lesson does not necessarily require exposing the names, sensitive incident details or commercial information in the original record. The guidance and its supporting evidence need appropriate access controls.

Finally, an AI-generated finding should not automatically become a new rule. Reviewers need a way to challenge it, record an exception and refer a proposed change to the person responsible.

Does it improve the next piece of work?

The value of operational memory should be tested in the workflow it supports. For a RAMS review, compare work performed with and without the retained checks using representative documents and an agreed expert assessment.

Examine whether reviewers identify more valid issues, raise fewer irrelevant queries and complete the review with less effort. Also check whether they can trace findings to evidence and recognise when a lesson does not apply.

Time saved alone is an incomplete measure. A faster review that misses a consequential issue, or repeatedly flags harmless differences, has not achieved the intended improvement.

Start with one recurring issue

The same approach can support incident triage, handover briefs and follow-up actions. The useful starting point is a workflow where experienced people repeatedly correct the same omissions or explain the same decision.

Choose a recurring issue. Capture the evidence, agree the check, define its scope and test whether it helps the next person. Expand only when that use proves useful.

The aim is to make practical expertise available at the point of work, with enough context for people to exercise their own judgement.

Related perspective: Why System 3 thinking needs an operational layer.