Phase Completion Rate and the Skipped Phase Problem

Why this matters

A phase completion rate of 70 percent reads like a crew that finishes seven pieces of work out of ten and drops three. That is almost never what happened. The figure collapses three different endings into a two-state number, and the ending it hides, the phase somebody deliberately took off the job, is usually the larger of the two endings that are not completion, and the only one of those two that is good news. Shops end up chasing crews over a figure that is mostly reporting a fault in their own job templates.

What the figure is actually counting

The numerator is phases marked complete. The denominator is phases created. A phase that was deliberately removed from the job sits in the denominator and never reaches the numerator, so it is arithmetically identical to a phase that was abandoned halfway or forgotten entirely.

Two constructions are in common use and they answer different questions:

• Cohort form. Denominator is phases created in the window; numerator is those same phases that are now complete, whenever they finished. This can never exceed 100 percent.
• Throughput form. Denominator is phases created in the window; numerator is phases completed during the window, whatever job they came from. These are two different populations, and on a draining backlog the figure goes above 100 percent.

The tell is simple: if that number has ever printed above 100, you are reading the throughput form, and it is measuring how fast the office opens phases as much as how fast crews close them. Everything below assumes the cohort form, which is the one worth reading.

Three silent exclusions ride along with it, and each one narrows what the figure can honestly be said to describe:

• Jobs with no phases contribute nothing. A one-visit repair that was never broken into phases is invisible here. The rate describes only the subset of work somebody decided was big enough to phase, which is normally the longer and more profitable end of the book.
• Phases created before the window and completed inside it are excluded from the cohort form. On work that runs longer than the window, that is most of what finished.
• Phases still in flight sit in the denominator with no outcome yet. They are opportunities counted against results that have not happened. Read a window old enough for its phases to have resolved; the mechanism is the same one the estimate conversion card owns.

Three outcomes wearing one number

Separate the denominator into four states before you read anything. Three of them are endings and one is not:

Complete and skipped are both fine outcomes. Stalled is the only bad one. The headline rate treats skipped and stalled as the same failure and treats in-flight work as failure too, which is why a shop with clean records and disciplined scope reduction can post a rate in the sixties and be running well.

Worked case: one quarter, 180 phases

A shop runs one four-phase template on its larger jobs: site survey and measure, rough-in, customer walkthrough sign-off, post-work cleanup and haul-off. In the quarter it opened 45 such jobs, so 45 times 4 is 180 phases created. At the quarter's close:

• Complete: 126, which is 126 of 180, exactly 70 percent
• Skipped: 27, which is 15 percent of the 180 created
• Stalled: 9, which is 5 percent of the 180 created
• In flight: 18, which is 10 percent of the 180 created

Those four shares sum to 100 percent of the 180 phases, which is the first check worth doing - if they do not, a state is missing or something is being counted twice.

The headline is 70 percent. Now take both non-results out of the denominator. Pull the 27 skips, because a phase cancelled on purpose was never work the crew failed to do, and pull the 18 still in flight, because they have not had their chance yet. The rate over the 135 phases that were meant to be performed and have resolved is 126 divided by 135, roughly 93 percent. The numerator never moved: of the 23 points between 70 and 93, about 12 are the skips leaving the base and about 11 are the in-flight work. Both figures are shares of phases in the same quarter, so they are comparable; neither is a share of jobs.

Then read the number that actually indicts somebody: 9 stalled phases out of those 135, roughly 7 percent. That is the real miss rate, and it is a different order of problem from the one a 70 percent headline suggests. Nine open items is a morning's work to clear. Fifty-four is a crisis. The headline implied the second and the records held the first.

The skip rate is the better signal

Having pulled the 27 skips out, do not throw them away. Sort them by which phase they were:

The four rows sum to the 27 skips. One row holds 19 of them, roughly 70 percent of the quarter's skipped phases, and that row is not a discipline problem. A phase skipped on 19 of the 45 jobs that carried it was never a phase on those jobs. The crew was folding haul-off into the last working phase and closing out on site, which is what most crews do when the debris fits in the truck they already have there.

The unit of analysis here is instances of one phase type, not jobs and not revenue. Do not read a skip rate off fewer than about 20 instances of a phase type, because at 20 instances a single skip moves the rate 5 points and two skips look like a pattern. All four rows above carry 45 instances, so all four are readable. This floor is stated on phase instances and does not transfer to the job-count floors the margin and ranking cards use; those are a different unit.

What a high skip rate is telling you to change

A phase type that is skipped often is one of three things, and the tell that separates them is where the work went:

• Folded into a neighbour. The work happened, inside another phase. The cleanup row above is this. Fix: take it out of the template for the job types where it folds in, and put its checklist items on the phase that absorbs it. The work does not stop being tracked; it stops being a scheduling unit that somebody has to open and close.
• Genuinely optional. The work happens on some jobs and not others, and the decision is made at survey. The walkthrough row at about 11 percent is probably this. Fix: leave it in the template and stop reading its skips as misses. An optional phase with a stable skip rate is a correctly built template.
• Quietly dropped. The work did not happen and nobody decided it should not. This is the dangerous one, and it hides inside the same count as the other two. The tell is that the reason field is empty, or the skip was recorded after the job closed rather than at the point the decision was made.

That last case is why the skip needs a reason at the moment somebody skips it. Without it, all three collapse back into one number and you are where you started, one level down.

What changes the read

The skip rate is only an estimating signal if the template is the same across the instances you are comparing. If the shop edited the template mid-quarter, phases created before and after the edit are different populations and pooling them produces a skip rate for a template that never existed. Split the read at the edit date.

The other condition that inverts the advice: on work that carries an inspection, a permit sign-off or any third-party gate, a skipped phase can mean the gate was bypassed rather than that the work was folded in. Do not remove a phase from a template on skip rate alone when that phase exists to prove something happened.

Two shops where the number collapses

The clearest way to see what the rate does and does not carry is to break it deliberately.

The shop that opens phases only for work already booked to a date. Nothing is ever skipped, because nothing is opened until it is certain. The rate sits near 100 percent every quarter and moves for no reason connected to the work. The figure is not measuring completion; it is measuring how late the office creates phases, and at the limit it measures nothing at all because there is no variance left in it.

The shop with no skip state, where a dropped phase gets closed as complete. Run the same quarter as above through that convention: the 27 skips move into the numerator, so the rate is 153 of 180, exactly 85 percent, and it reads as the best quarter on record. The 9 stalled phases are still in there and still invisible, the cleanup finding is gone entirely, and the shop now has a flattering number and no signal. This is the more common of the two, and it is worth checking for before you read any of the figures above: if closing and skipping land in the same state, the skip RATE cannot be recovered afterwards, though a list of candidates can be, as below.

Both cases collapse the same way. The rate carries information only when the three endings stay separable, and the endings stay separable only if somebody records the decision at the moment it is made.

If you are in the second shop and want last quarter's skip signal anyway, there is one imperfect proxy: phases closed complete with no logged time against them and no notes. That combination is what a skip looks like in a shop whose only way of closing a phase is to call it complete. It over-counts, because a short phase done by someone who does not log time looks the same, so treat it as a list of candidates to ask about rather than as a count. It is worth one pass to find the phase type that dominates the list, and then the fix is forward-looking anyway.

The two numbers also run on different clocks. The stalled count is a daily read - nine open items with passed dates is a worklist, and it is worthless a week late. The skip rate is a quarterly read, because a template correction needs enough instances to be trusted and because changing a template more than a few times a year means nobody ever learns what the current one is.

References

• See related: Estimate Conversion Rate and the Cohort Problem - why in-flight work in a denominator understates any rate
• See related: Overdue Phases Are Measured Against the Window, Not Today - the stalled count, and the two dates that decide it
• See related: Change Order Approval Rate and Why a High One Is Not Good News - the other place a scoping fault shows up as a performance number
• Trade-standard practice: scope changes recorded at the point of decision, with a reason, rather than reconstructed at close-out