Corrective actions that actually close

Every safety programme that has run for more than a year has a list of open corrective actions, and almost every one of those lists is longer than it was last quarter. The instinct is to blame discipline — people are not following through. Usually the cause is structural, and it is visible in the list itself: actions that were never really actions, owners that are departments, dates that were never negotiated with anyone, and a closure step that means nothing more than somebody ticking a box.

Why actions stall

Four causes account for most of it, and none of them is laziness.

• The action was a restatement of the finding. "Housekeeping in bay 3" is an observation. "Install two waste bins at the bay 3 pick face by 14 October — M. Silva" is an action. If the row does not say what will be different, nobody can do it or close it.
• It needed someone who never agreed to it. An action assigned to a person who was not in the room, has no budget for it, and first learns of it from a spreadsheet, is a wish with a name attached.
• The date was arbitrary. A date chosen by the person writing the finding rather than the person doing the work slips once, then stops being treated as a date at all.
• Closure costs nothing. Where closing a row requires no evidence and no second person, the list will always tend toward closed, and the closed state stops carrying information.

One issue, one action

The most common structural defect is a single row that contains three jobs: fix the guard, retrain the operator, revise the procedure. Those have different owners, different durations and different failure modes. Closed as one row, two of them close because the first one did.

Split them. One row per thing that will be done. It makes the list longer, which feels like the wrong direction and is not: a list of eleven real actions is more closable than a list of four compound ones, and the eleven tell you honestly how much work is outstanding.

An owner is a person

"Maintenance", "the site team" and "HSE" are not owners. An owner is one named person who can either do the work or direct it, who knows they own it, and who agreed to the date. Everything else is a queue nobody is standing in.

Dates that mean something

A due date is a commitment between two people, not a field. Three rules make dates hold:

• The owner sets it, or at least agrees it. Imposed dates slip silently; agreed dates get renegotiated out loud, which is far more useful.
• Interim controls have their own date, and it is sooner. If the permanent fix is eight weeks out, what protects people for eight weeks — and when is that in place? That is a separate action with a separate date, and it is usually the more urgent one.
• Slipping a date is an event, not a silence. Changing a due date should require saying why and agreeing a new one. A date that moves without comment is how a list becomes fiction.

Prioritising without inventing a matrix

Elaborate scoring schemes mostly produce arguments about scores. Two questions do nearly all the useful work:

1. If nothing is done, what is the worst realistic outcome? Not the worst imaginable — the worst that would not surprise you.
2. How exposed are people in the meantime? A hazard behind a locked door with one person a month near it is not the hazard in a walkway used two hundred times a shift.

High severity plus high exposure goes first, and gets an interim control today. Everything else is scheduling. If two people disagree after those two questions, the disagreement is about facts, and that is a better conversation than one about whether something is a 12 or a 15.

Complete is not verified

This is the distinction that makes closure mean anything. Complete is the owner saying the work is done. Verified is somebody else confirming it — looking at the thing, not at the row. They happen at different times, and collapsing them into one status is how a tracker fills with closed rows describing work nobody checked.

Verification does not have to be heavy. For most actions it is a person walking past and looking. For the few where the control is the only thing between someone and serious harm, it should be deliberate, recorded, and done by someone other than the person who did the work.

What counts as evidence depends on the action. A photograph for a physical change. A signed record for a training action. An updated document with a revision date for a procedural one. "Done" in a comment box is not evidence of anything.

When the same thing comes back

A finding that recurs after being closed is the most informative row on the list, and the most commonly wasted. It means one of three things, and they need different responses:

• The action addressed the symptom. The spill was cleaned; the leak was not fixed.
• The control depends on someone remembering. It works when observed, which is why it keeps passing inspection and keeps coming back.
• The action was never really completed, and nobody verified.

Tracking where a finding came from is what makes recurrence visible at all. Without the link back to the inspection, incident or report that raised it, a recurrence looks like a new finding, and gets a new action that will also close.

The review that holds it together

One fixed meeting, short, on a cadence that never moves. Weekly or monthly matters far less than never being cancelled. Four questions:

1. What is overdue, and what is the new agreed date?
2. What closed since last time, and was any of it verified?
3. Is anything high-severity still without an interim control?
4. Has anything come back?

Counting overdue actions at every review, out loud, does more for closure rates than any tooling. It is also the number to take to leadership — not how many findings were raised, which measures how hard you looked, but how many were closed and how long it took.