Case Study 30-2 — An SPI of 1.02 and a Gym Thirty-Four Days Late

The projects, companies, and people in this case study are illustrative composites created for teaching. The arithmetic is real; the names are not.

Setup

Project: Rivermont Elementary School #12 — a replacement elementary school with a classroom wing, an administration block, and a full-size gymnasium. Owner: Rivermont Unified School District. District representative: Harriet Nkemelu, Director of Facilities and Construction. Contractor: Kestrel Construction Group. Delivery: hard-bid design-bid-build, lump sum $22,400,000. Public work — prevailing wage, certified payroll, 100 percent payment and performance bonds. Liquidated damages $2,500 per calendar day; extended general conditions on this job run $2,100 per calendar day; combined exposure $4,600 per calendar day. Kestrel team: Curtis Boone (project manager), Gil Amundsen (general superintendent). Wei Chen (corporate project controls) is the visitor.

Curtis Boone is good at this. He wins work by buying it low, he runs the leanest staff in the company, he knows every foreman's first name and most of their kids', and he manages by force of personality — which works, right up until the month it does not. His bid on this job was built at $22,865,000 and shaved $465,000 on bid day to reach the $22,400,000 he submitted. Remember that number.

Curtis's monthly report to the district is a schedule of values, a payment application, and — because the district's Division 01 specification requires it — an earned-value summary computed from the schedule of values percentages.

Data date: the end of month 9.

What Happened

Page one

Curtis presented it in the district's monthly meeting exactly as it reads. Slightly ahead of plan on production, two percent inefficient on cost, contingency covering it. Harriet Nkemelu thanked him.

Gil Amundsen, who had been staring out the window through the whole item, said, "The gym roof isn't going to be on before the rains."

There was a pause of the kind that ends a meeting, and then the meeting ended.

Nadia Haddad heard about the pause the following Tuesday and sent Wei Chen out for a day.

What was underneath page one

Curtis's own cost codes rolled up into three groups. He had never reported them separately because the specification asked for a project summary and he had given it exactly what it asked for.

Everything foots. Nobody entered a wrong number. And the project line describes a job that does not exist.

Mechanism 1 — the aggregate is the average of a fire and an icebox. Here is the whole of the +$264,000, laid out as it actually formed:

    Critical-path work    −$910,000   ████████████████████ (against)
    Non-critical work   +$1,174,000   ██████████████████████████ (for)
    General conditions          $0
    ------------------------------------------------------------
    Reported net          +$264,000

Gil's crews had float in the sitework and the classroom wing, so when the gym slowed down they moved to work that was available. That is good field management. It is also work that cannot move the completion date, and it earns value at exactly the same rate per dollar as work that can. The instrument added a dollar of paving to a dollar of gym roof and reported the sum.

Mechanism 2 — the schedule of values was front-loaded. Curtis's SOV, submitted at the start of the job and approved by the district, does not match the way the cost actually falls.

Modest front-end loading to cover genuine mobilization cost is normal and defensible. This is $680,000, and it is not mobilization; it is cash flow borrowed from the back of the job.

Now be careful about what it did, because the popular version of this story is imprecise and the precise version is more useful.

It did not move SPI much. A front-loaded line that was both scheduled complete and actually complete by the data date raises EV and PV by the identical amount, and SV does not budge. The amplification only appears where the front-loaded work finished ahead of plan. Of the $430,000 loaded onto sitework and foundations, the baseline had those lines about 85 percent complete at month 9 and they are 100 percent complete, so the front-loading amplified the favorable schedule variance by about $430,000 × 0.15 = $64,500 — of a $1,174,000 total. Strip it out and the honest restated numbers are PV $12,564,500, EV $12,764,000, SPI 1.016. Still above 1.000. Still a lie about the schedule. De-loading the schedule of values barely helps, because the aggregation is the problem, not the loading.

What it did do is destroy the cost forecast. EV is supposed to be the budgeted value of work performed, and a front-loaded SOV is not a budget — it is a billing strategy. Restate EV on an honest cost-proportional basis and the whole picture moves:

Read the three percentages the way §30.6 teaches, and the front-loading becomes visible in one line:

Earned 60.0 percent. Spent 61.2 percent. Actually built 57.0 percent.

Mechanism 3 — level-of-effort dilution. $1,900,000 of general conditions earns exactly its plan every month by definition. That is 8.5 percent of BAC permanently pinned at SPI 1.000, and it pulls the aggregate toward 1.000 from whichever side the aggregate happens to be on. Take it out and look:

The SPI got worse to remove it and the CPI got worse to remove it, which is the same statement made twice: LOE always drags an index toward 1.000. Here it was concealing how far above 1.000 the schedule index really was and how far below 1.000 the cost index really was. On a job whose aggregate SPI is under 1.000, the same mechanism flatters instead. Either way you are not looking at the field.

What the CPM said that the EVM structurally could not

Wei asked Gil for the schedule update. Gil had one, hand-maintained, four weeks old, and correct.

The gym roof structure, the gym overhead MEP, and the gym finishes form a single chain with no float. The chain was at minus 34 calendar days of total float. Substantial completion had moved thirty-four days.

34 CD × ($2,100 extended general conditions + $2,500 liquidated damages) = 34 × $4,600 = $156,400

Of that, $143,548 of extended general conditions will eventually flow into AC and make the EAC worse still, and $93,500 of liquidated damages will never appear in any earned-value report Curtis ever issues, because BAC is the budget for the work and it has no line for being late.

There is no arithmetic operation you can perform on PV, EV, and AC that will produce the number 34. The only instrument that produces it is the forward pass and the backward pass.

But there is one thing you can do with the earned value once you know where to look, and Wei did it that evening. The gym chain's baseline cumulative planned value ran $3,540,000 at month 6, $4,180,000 at month 7, $5,100,000 at month 8, and $5,880,000 at month 9 — so month 9 planned $780,000 across 30 calendar days, $26,000 per calendar day.

Crude conversion: SV ÷ planned daily rate = −$910,000 ÷ $26,000/CD = −35.0 calendar days

Earned schedule on the gym chain: EV of $4,970,000 falls between month 7 and month 8. ES = 7 + ($4,970,000 − $4,180,000) ÷ ($5,100,000 − $4,180,000) = 7 + 0.859 = 7.86 months SV(t) = 7.86 − 9.00 = −1.14 months ≈ −34.7 calendar days, SPI(t) = 0.873

The CPM: −34 calendar days

Three methods, three numbers within a day and a half of each other. The earned value knew. The number was sitting on the critical-path group line the entire time, in Curtis's own spreadsheet, underneath a project total that averaged it away.

Two honest cautions about that result, because it is the kind of thing people over-learn. It worked because the gym chain is a single serial chain with a roughly linear planned burn; on a group with parallel branches or a lumpy curve the conversion will not land anywhere near the CPM. And Wei only knew which rows to isolate because they had read the schedule first. The earned value corroborated the CPM. It could not have found it.

Analysis

Curtis did not do anything wrong on the day of the meeting. He reported the number his specification asked for, computed correctly from the schedule of values the district had approved. That is what makes this the more dangerous of the two case studies in this chapter. Northgate's report was bad news correctly delivered. This one was an arithmetically flawless report that told the owner the opposite of the truth, and everyone in the room believed it, including the person who produced it.

The decisions that made the report useless were all made months earlier, in preconstruction. Front-loading the SOV was a cash-flow decision made at buyout. Reporting a project summary rather than a controlling-path line was a reporting-structure decision made at the first pay application. Running a lean staff meant nobody owned the schedule model, so the CPM was a hand-maintained side document rather than the instrument the report was read against. None of the three is visible in the month-9 numbers, and all three are the reason the month-9 numbers lie.

The front-loading punished Curtis before it punished anybody else. That is the part contractors miss. Borrowing $680,000 of value from the back of the job improved his cash position and cost him the ability to see his own project: it inflated every percent-complete he reported to himself, and it turned a forecast overrun of $446,534 into one of $1,663,679 that he could not detect. Against a bid fee of $784,000, the honest forecast does not merely erase the fee — it erases the fee and about $880,000 more. The contractor who front-loads is the contractor most likely to be surprised in month fourteen.

And the estimate had already said so. Curtis bid $22,865,000 and shaved $465,000 to $22,400,000 on bid day. The month-9 CPI extrapolation — before any restatement — forecasts $22,846,534, within $18,466 of the number his own estimator built before the shave. Nine months of production had merely re-derived the estimate. Shaving a bid does not make a job cheaper; it makes the overrun arrive later and with your signature on it.

The rule this case exists to teach is the one in §30.9, and it does not have an exception: earned value management is a cost-performance instrument. It gives you an early, honest read on efficiency and a defensible, testable forecast of final cost. It is a poor schedule instrument. Use it alongside the CPM, never instead of it. When somebody asks you for one number, tell them there are two, and tell them why.

I have kept a copy of Curtis's month-9 report in my desk for years. Not because it is wrong — every figure on it is right — but because it is the most persuasive wrong document I have ever seen.

Discussion Questions

1. Curtis's report satisfied the district's specification exactly. Draft the two sentences you would add to a Division 01 earned-value requirement to make this failure impossible, and then argue against your own language: what does an owner give up by prescribing report structure this tightly?
2. Stripping the front-loading moved SPI from 1.020 to only 1.016, but moved the forecast at completion by $1,217,145. Explain, in terms of which measure each distortion touches, why one effect is tiny and the other enormous.
3. Gil Amundsen knew. He said so in the meeting, in one sentence, and it changed nothing. Whose failure is that, and what specific mechanism in the reporting process would have converted his sentence into a number on page one?
4. Wei's three schedule estimates — $26,000-per-day conversion, earned schedule, and the CPM — landed at 35.0, 34.7, and 34 calendar days. Under what conditions would you expect them to diverge badly? Name three, and say which of the three you would still defend in a claim under Chapter 33.
5. Compare this case with Case Study 30-1. Northgate reported an SPI of 0.969 while running nine days late; Rivermont reported 1.020 while running thirty-four days late. Both SPIs are equally uninformative about time. Why does the Northgate report nevertheless work and this one not?

Your Turn

You are Wei Chen, and Nadia has asked for a memo to every project manager in the company — not a report on Curtis, who is a good project manager who inherited a reporting structure nobody had examined. Two pages, and it has to survive being read by people who are busy and slightly defensive.

Produce it. It must contain:

(a) A one-paragraph statement of the failure that does not name Curtis or the school. (b) The required report structure, going forward: name the rows every Kestrel earned-value report must carry, in order, and state which single row goes above the project total and why. (c) The rule on level of effort: where it lives, what it may and may not be folded into, and one sentence explaining the mechanism to somebody who has never heard of it. (d) A front-loading test somebody can actually apply to an SOV in ten minutes at buyout. It must produce a number, not an opinion. (e) The one line that must appear on page one of every report, every month, next to the SPI: write it out, exactly as you want it printed. (f) The paragraph that goes to project executives rather than project managers, on why the $680,000 of front-loading cost the contractor $1,217,145 of forecast visibility. Executives approve schedules of values. This is the paragraph that changes what they approve.

Then do the hard part. Write the closing paragraph that would make a project manager who has been front-loading for fifteen years actually read the memo instead of filing it — without accusing anyone of anything, and without pretending the practice is rare.