Your Attendance Log Is a Capacity Study

The position that was not there

Two things happen on almost every plant walk I do. A supervisor tells me what the line is staffed at, and the line shows me a different number. Both answers are honest. The roster count, the target on-line count, and the bodies actually standing at stations are three separate figures, and only the last one produced today's output.

The observation that matters is smaller than any of those numbers. On a return walk, a handling station sitting between two automated steps was empty. It had been staffed the week before. Nothing downstream was short, nothing had queued behind it, and nobody had raised it. The supervisor's answer was that someone is normally there, they were just out that day.

That is not an anecdote about one station. That is a controlled test the plant ran at its own expense and then filed under attendance.

Absence is evidence, a pilot is a purchase

A staffing decision is a question about whether a position is load-bearing. Does output, quality, or the downstream queue change when the position is not covered? That question has exactly one clean form of evidence: shifts where the position was not covered.

Plants generate that evidence constantly. Call-outs, vacation weeks, a line pulled short to cover a sanitation overrun, a partial shift where someone got moved. Over a quarter, most stations on most lines accumulate a meaningful sample of unstaffed hours. The data exists on both sides, too: attendance is tracked because it has to be, and output is tracked because it has to be. They are almost never joined.

They are not joined because the two records answer to different owners with different purposes. Attendance belongs to HR and is read as a compliance and coverage problem, where a short shift is a failure to be corrected. Output belongs to operations and is read as a schedule attainment problem. Nobody owns the question that sits between them, which is whether the shifts that ran short actually cost anything. So the evidence accumulates and expires in place.

What happens instead is that someone proposes a headcount change, the room asks how we know, and the answer becomes a pilot. Run the line short for a few weeks under observation, measure, then decide. That pilot is a reasonable instrument and I use staged removals regularly. But it is a purchase. It costs schedule, it costs the attention of a plant team that has other work, and it spends political capital on the first ask, which is usually the ask you most want to land. Paying for it to learn what last quarter's records already show is the expensive route to a decision you could have walked in with.

The second-order effect is worse than the cost. When evidence is scarce and slow, the positions that come out are the ones with a champion, not the ones the data indicts. A supervisor who can articulate why a station is redundant gets it removed. A station nobody thinks about stays staffed for years. You end up optimizing for narrative quality rather than for load.

Read the log first, then test in steps

Start with the join, because it is free. Take one line. Pull the attendance exceptions for a quarter, station by station. For each short-staffed shift, pull that day's output for the line, the downstream hold or rework volume, and any overtime that the following shift absorbed. You are looking for one thing: stations with repeated absences and no detectable downstream signature.

Sort the result into three groups, and be strict about the boundary. Positions with several no-loss absences are candidates for removal now. Positions whose absences correlate with a real loss are confirmed load-bearing, and that is a finding worth as much as the first group, because it ends the argument. Positions with too few absences to read are the ones that actually justify a staged test, and now that test is aimed at a short list rather than at the whole line.

Sequence the removals and let the line prove each one. Take one position out, hold it for a full week including the worst day of the schedule, and check the same three measures you used in the join. Then take the next. Stepwise removal matters because the failure mode of a headcount change is not the first head, it is the third, where you cross from slack into the actual constraint. Taking them together tells you only that something broke, not which one.

Be precise about what each state is worth. A position identified is not a position removed. A position removed for a week under observation is not a position removed. A position out of the staffing plan, with the roster reduced and the line running to schedule for a full cycle, is the only one that has reached the P&L. Keep those three states in separate columns and report them separately, because collapsing them is how a savings number turns into an argument nine months later.

One more habit, which is the cheapest of all of these. When you are standing at a station, ask the person working it how long the task takes with the current crew and how long it takes with one fewer. On the walk I described above, the answer came back immediately and it scaled linearly, two people for a given time, one person for double it. The floor holds a lot of usable numbers, and the constraint on getting them is almost never that the numbers do not exist. It is that nobody asks in the specific form that makes the answer usable.

What a well-run staffing decision reads like

Each station on the line has a recorded count of hours it ran unstaffed in the last four quarters, alongside the output, hold volume, and following-shift overtime on those days. A position with repeated absences and no measurable loss has either come out of the plan or has a written reason it stays, naming the condition it covers. Staged removals run one at a time, one full schedule cycle each, against measures fixed before the test starts. Identified, tested, and removed are tracked as three separate counts, and the number that goes to finance is the third one only.

If you want the first check today, take a single line, pick the four stations your schedulers worry least about, and ask how many shifts each one ran uncovered last quarter and what the line produced on those days. If nobody can answer, that is the finding, and it is the same finding on every line in the building.