Overdue Phases Are Measured Against the Window, Not Today

Why this matters

An overdue count only means what you think it means when the window ends today. Pull the same figure for last quarter and it stops being a worklist and becomes a historical fact, and nobody is warned that the question changed. Shops act on it either way: they hand a stale list to a dispatcher who works items that closed weeks ago, or they read a flattering historic figure as evidence the backlog is clearing when all it shows is that time passed. The records are fine. The reading is what moves.

The number depends on three dates, and most shops set one

An overdue phase is one that is not finished, not deliberately removed, and whose scheduled end has already passed. Three separate dates decide whether a given phase is in that count, and a shop normally chooses only the third:

• The evaluation date. Not finished as of when? As of the moment you are reading, or as of the moment the window closed? On a window ending today these are the same date, which is exactly why nobody notices the question exists.
• The window start. Overdue by how far back? A phase whose scheduled end passed before the window opened is not in the count at all, however open it still is.
• The window end. Passed means passed before this date, not before today.

Those three produce three different figures off one set of records, and two of them are legitimate answers to two different questions. The third is a number that cannot be compared to anything.

Worth naming up front: the phase with no scheduled end recorded is in none of the three. The count is computed only over phases that carry a date, so it improves the less you record, which is the opt-in denominator problem the SLA compliance card owns.

The same records, read three ways

A shop runs a four-phase template on its larger jobs. Ninety-six phases had a scheduled end falling inside the quarter that closed six weeks ago. Three readings, all of them correct, all of them off the same rows:

Reading one, the historic window with status taken at its close. Twenty-three of those 96 phases were unfinished and unskipped when the quarter ended, which is about 24 percent of the 96 phases due in that quarter. This is a statement about the state of the shop on the last day of the quarter.

Reading two, the historic window with status taken today. Of those same 23, fifteen have been completed since, four were skipped once somebody looked at them, and four are still open. So the identical window, evaluated now, returns 4 of the same 96, about 4 percent. Same rows, same date filter, one different assumption, and the figure falls by a factor of roughly six.

Reading three, the live window. Over the trailing 90 days ending today, 104 phases had a scheduled end on or before today, and 14 are still open. That is about 13 percent of the 104, and it is the only one of the three that is a to-do list.

Fifteen plus four plus four is 23, which is the check that the second reading is a subset of the first rather than a different population.

Why only two of the three can be compared

Readings one and three are comparable to each other: both take a phase's status at the same moment they take its deadline, so both answer "of the phases whose time had run out by the moment of reading, what share was still open then". On that basis the share went from about 24 percent in the prior quarter to about 13 percent over the trailing 90 days, a fall of roughly 10 points. Both sides are shares of phases, both are measured at their own window's close, and neither is a share of jobs.

Do not make that comparison with the raw counts. Twenty-three against 14 is two different denominators - 96 against 104 - and the counts would still have fallen if the shop had simply taken on less phased work.

Reading two cannot be compared to either. Its phases have had six extra weeks to resolve, and nothing in the figure says so. It will always look better than a live reading of the same shop, and it will look better the longer you wait to run it, which means a shop that pulls last quarter's overdue figure in month three of the next quarter is reading its own patience.

The failure mode is not abstract. Hand a dispatcher reading one as a worklist six weeks after the fact and 19 of its 23 rows are already resolved, roughly 83 percent of the list, because 15 were completed and 4 were skipped. The four live items are buried in it, and the second time somebody works that list and finds most of it closed, they stop trusting the list rather than the person who sent it. A worklist is only ever a live reading.

Before crediting the fall from about 24 percent to about 13 percent to better execution, check one thing: whether the number of open phases carrying no scheduled end rose over the same period. Those phases leave the denominator entirely, so moving a dozen of them out of the dated population lowers the share with no change in how the work ran. If undated open phases held roughly steady, the fall is real.

The useful discipline is to say which of the two legitimate readings you want before you pull it. Live for work. Historic-at-close for trend and for the audit below. Historic-evaluated-now for nothing, unless the question genuinely is "what is still open from back then", in which case call it a backlog age report, not an overdue rate.

The live reading is a worklist, and it sorts by age

Fourteen open items is a morning, not a crisis, but only if they are triaged by how late they are rather than worked top to bottom:

• Six are under a week late. Most of these are a close-out that never got recorded. The phase is done and the row is not. Ask, do not schedule.
• Five are one to three weeks late. These are real slips, and at this age the customer usually knows. The action is a date, communicated, not a status change.
• Three are over a month late. These are the ones that are not late, they are lost - a phase waiting on something nobody is chasing, or work that quietly stopped being part of the job and was never removed. Each one gets a decision today: finish it, or skip it with a reason.

Six plus five plus three is 14. The third band is the one the number exists to surface, and it is also the band most likely to fall out of the reading entirely, because of the window start. Those three phases were due 34, 51 and 78 days ago. Run the same live reading over a trailing 30 days instead of 90 and it returns 11 of the 14, dropping precisely the three worst items on the list. A short window is not a conservative choice here. It is a filter that removes the oldest problems first.

The historic reading is a template audit

Reading one is where the value is, and it is not a performance review. Split the quarter's 23 overdue phases by which phase they were:

The rows sum to the 96 due and the 23 open. One row carries 16 of the 23 overdue phases, roughly 70 percent of them, and it is the phase whose completion date the shop does not control. The template was giving the inspection phase the same three-day allowance as the rough-in it followed, so on most jobs it was overdue the moment the inspector's queue ran longer than three days.

That is a template fault and it has a template fix: set that phase's scheduled end from the authority's actual turnaround measured over the last several jobs, not from what the office hopes. If the turnaround is genuinely unpredictable, the better move is to stop giving that phase a scheduled end at all and track it as a hold with a follow-up date, because a date nobody can hit teaches the crew that overdue means nothing.

Three conditions change that call. A phase overdue because it is waiting on the customer needs the date kept, since the date is the evidence in that conversation. A phase overdue because a part is on backorder belongs to procurement and will not be fixed in the template. And if the shop changed the template inside the window, phases created either side of the change are different populations and pooling them produces a finding about a template that never existed.

The phases neither reading can see

Two populations are missing from every figure above, and both are worth counting separately because both grow quietly.

Phases with no scheduled end. These can never be overdue. If overdue is the number anyone is judged on, clearing the date is the cheapest way to improve it, and it does not take bad faith - it takes one person deciding the date was guesswork anyway. Count undated open phases alongside the overdue count, in the same read, so that removing a date moves a number rather than hiding one.

Phases on jobs that were never phased. The whole figure describes only work somebody broke into phases. A job running long as a single undivided block cannot appear here at any window length.

What is in and what is out

Read the second and fourth rows together and the difference between the two historic readings is exactly this: everything that resolved after the window closed. That is the whole gap between 23 and 4 in the worked case, and it is also the reason the second column is the one to trend and the third is the one to leave alone.

References

• See related: SLA Compliance Only Counts Jobs That Carry a Deadline - the opt-in denominator, and what it does to any date-gated rate
• See related: Phase Completion Rate and the Skipped Phase Problem - the stalled phases that feed this count, and how to tell them from deliberate scope reductions
• See related: Change Order Approval Rate and Why a High One Is Not Good News - the other number that reads differently on a window that has not finished resolving
• Trade-standard practice: scheduled dates on work controlled by a third party set from measured turnaround rather than from target