Case Study 1 — Fifty-Four Percent: What Six Weeks of Measurement Found on Northgate

All people, companies, and projects in this book are illustrative composites. Numbers are consistent within the book and are not drawn from any actual project.

Setup

The project. Northgate Outpatient Pavilion. Meridian Health System, owner. Kestrel Construction Group, construction manager at risk with a $47,500,000 guaranteed maximum price. Four stories, 132,000 gross square feet. Contract time 565 calendar days from a March 3, Year 1 notice to proceed, with substantial completion contractually due September 18, Year 2. Liquidated damages $5,500 per calendar day; extended general conditions $5,150 per calendar day; combined exposure $10,650/CD.

Where the job stood. Month twelve. February of Year 2. The structural steel had started 23 days late in August of Year 1 because an anchor-bolt submittal sat in Kestrel's own office for 11 days, and Kestrel had spent $168,000 accelerating to recover 17 of those days and absorbed the other six. The building was closing in on its dried-in milestone of March 28. Interior rough-in was starting on Levels 2 and 3. Fourteen trade contractors were on site.

The disagreement. Wei Chen's month-eleven schedule update, run February 5, put substantial completion at September 21 — three calendar days behind. Margo Deacon, general superintendent, said the number was wrong and could not say why in a form Wei could model.

The people.

• Ray Alvarez — Kestrel senior project manager
• Margo Deacon — Kestrel general superintendent, 31 years
• Wei Chen — Kestrel project controls manager, P6 scheduler
• Dani Okonkwo — field engineer, eighteen months on the job
• Sofia Marchetti — project manager, Cardinal Mechanical
• Devlin Achebe — project manager, Halcyon Electric
• Ruben Ochoa, Ivy Beaudry, Ellis Baird, Nell Ostrowski, Marisol Duarte — trade foremen
• Pri Sethi — owner's representative, Meridian Health System

What Happens

The experiment design

Ray's proposal had one feature that made it worth doing: they would change nothing for six weeks.

Every Monday at 7:00 a.m., each of fourteen trade foremen would state, out loud, in front of the others, what their crew would complete by Friday. Not work on. Complete. Every Friday at 7:00 a.m., they would go down the list and mark each item done or not done — binary, no partial credit — and for every miss the foreman would say why, in their own words, and Dani would write it down and assign it a category.

No countermeasures. No corrective action. No coaching. Just measurement.

Sofia Marchetti asked the question everyone was thinking: "What happens to the guy with the worst number?"

"Nothing," Ray said. "That's not a promise about being nice. It's the design of the experiment. The second anybody gets leaned on for a low score, everybody starts committing to things they've already finished and the data is worthless. If you want to hurt me later with my own numbers, they have to be real numbers first."

Six weeks — February 9 to March 20, Year 2

Note what the weekly numbers do: nothing. Six weeks of flat, mediocre, stable reliability. That stability is itself a finding — it means 54% is not bad luck, it is the system's normal output.

The 86 misses, categorized

The Friday morning that changed the job

March 20, Year 2. Dani put the Pareto on the wall. Ray had expected an argument about subcontractor performance. What he got was Margo doing arithmetic out loud.

Margo: "Twenty-one, fourteen, twelve. Forty-seven of eighty-six. Who owns those three?"

Dani: "Prerequisite work is our coordination. Material is mostly the subs' procurement, but the need-by dates come from our schedule, and we never gave them dates by room. Information is the RFI log. That's mine."

Margo: "So fifty-five percent of the misses are us."

Ray: "Yes."

Sofia Marchetti: "Can I say something that's going to sound rude? My guys have been telling me this for four months. Ruben tells me every Friday: 'I couldn't get in, the tray was in the way, nobody knew.' I couldn't do anything with it because it was a complaint. Now it's twenty-one on a chart."

That is the mechanism. Everyone on the job already knew. What was missing was not information — it was counted information, in a category, with a denominator.

Then Wei Chen did the other piece of arithmetic. Over the six weeks — 30 work days of calendar — the longest path had advanced 26 work days. Net loss: 4 work days, about 6 calendar days, in six weeks. Roughly one calendar day of slip per week.

Wei: "With twenty-six weeks left to substantial completion, that rate produces twenty-six calendar days. Twenty-six times ten thousand six hundred fifty is two hundred seventy-six thousand nine hundred dollars. And Meridian's interim clinic lease expires October 1."

Ray: "So Margo was right and I was wrong, and the number wasn't three days."

Wei: "The number was three days in the model. The model absorbed the rest inside remaining durations. That's not a defect in the model. That's what monthly updating does."

What Kestrel changed

Three things. All boring. All cheap. None of them a new piece of software.

1. Somebody was given the time. Dani Okonkwo was assigned 20 hours a week to constraint removal and taken off two other assignments to create the hours. Not asked to "also do" it. Ray's line to Nadia Haddad, justifying the staffing to the vice president of operations: "Half a field engineer costs us about twenty-seven thousand over the rest of the job. One calendar day costs ten thousand six hundred fifty. This pays for itself if it saves three days."

2. The look-ahead was re-cut by room, not by activity. "Interior rough-in Level 3, 22 WD" became nine zone-level lines with named crews. Prerequisite-work misses fell almost immediately, because a superintendent can walk into a room and verify it is finished, and cannot verify an activity.

3. Material need-by dates were back-calculated to the room and entered on the constraint log with the vendor's name and a phone number. Roughly half the material misses turned out to be items sitting in a warehouse forty minutes away that nobody had asked to be released.

And one thing Kestrel did that cost nothing and mattered more than the other three combined: Kestrel put its own commitments on the same board. Six to nine items a week — RFI answers pushed, inspections booked, areas cleared, material released, access provided — scored in the same column as everyone else's. Kestrel's own PPC was below the job average for the first month, and Ray published it anyway.

The following twelve weeks

Two things to read in that table, and the second is the important one.

First: commitments made went up as well. From 32 a week to 42. That matters enormously. A team improving its PPC while shrinking its commitment count is gaming the metric. A team improving PPC while committing to more work is genuinely getting better. Completed commitments per week went from 17 to 34 — double.

Second: it plateaued at about 80%, and stayed there.

Why it plateaued

Kestrel spent two weeks in June trying to get to 90% and did not. Here is what the reason codes said about the last 20%:

Margo's summary, which is the honest one: "Eighty is where we stop being the problem and start being ordinary. Getting past eighty means fixing people who don't work for us."

Kestrel did not get past 80%, and 80% was enough. The measured slip rate fell from about 1.0 calendar day per week to about 0.3, and the job made its dried-in milestone.

Analysis

What worked, and why.

Measuring before intervening. The six-week no-change baseline is the single best decision in this case. Without it, every subsequent improvement would have been arguable and Ray would have been unable to answer "how do you know?" It also cost nothing but honesty.

The binary rule. Partial credit would have produced a first number around 78% and no urgency. Fifty-four percent got everyone's attention precisely because it was shocking, and it was shocking because it was strict.

Categorizing the reasons. PPC created urgency; the Pareto created direction. The categories are what converted fourteen foremen's accumulated frustration — which had existed for months as complaints — into three numbered problems with owners.

Kestrel scoring itself. This was the credibility purchase. A general contractor that demands public commitments while making none is asking subcontractors to accept a risk it will not take, and foremen detect that arrangement in about three weeks.

What was genuinely hard.

Staffing it. Twenty hours a week of a field engineer is not free and had to be taken from somewhere. The May dip shows exactly what happens when those hours get borrowed for something urgent — which they always will be.

Not weaponizing the data. Ray had, in his hands, twelve documented information-related misses traceable partly to unanswered RFIs from Halvorsen + Pike, on a job with a live delay exposure. The temptation to put that in a letter was real. He did not, because the first time a general contractor uses a PPC log as a claim exhibit, that job's data becomes worthless forever and every foreman in the room learns the correct lesson about writing things down.

The uncomfortable finding. Fifty-five percent of the misses were caused by the general contractor. That is not unusual — it is the normal result. The reason it feels surprising is that unreliability at the general contractor level is invisible without measurement: it shows up as someone else's crew standing around, which reads, to everyone in the trailer, like someone else's problem.

Discussion Questions

1. Ray insisted on changing nothing for six weeks. What would have been lost if Kestrel had started removing constraints in week two — and what would have been gained? Is there a case for skipping the baseline?
2. Kestrel's own PPC was below the job average in the first month and Ray published it. Argue the opposite case: what are the real risks of a general contractor publishing its own poor reliability to fourteen subcontractors on a job with a live delay exposure?
3. The plateau at 80% was caused partly by one subcontractor whose foreman was not permitted to decline a commitment. You are the project manager. Script the conversation you have with that subcontractor's project manager. What do you ask for, and what do you offer?
4. Wei Chen's monthly CPM update said three days; the weekly measurement implied about twenty-six. Both models were run correctly. Explain to a skeptical owner's representative how both numbers can be true, and what you would report to them.
5. Ray declined to use the PPC log as evidence in a potential claim against the architect's RFI turnaround. Was that the right call? Under what circumstances, if any, would you use it — and what would it cost you?

Your Turn

Take the twelve-week PPC table above. Compute, for each week, the number of completed commitments (given) and the number of missed commitments (made minus completed). Plot or tabulate misses per week alongside PPC.

Then answer one question in writing, in under 150 words: in the week of May 18, PPC fell from 81% to 80% — but is that week better or worse than the week of May 11? Defend your answer using both the ratio and the raw counts, and say which one you would put in front of Pri Sethi and why.