Reviewing the outturn and scaling the front end

A project that finishes on time and under its funding invites a short close-out: thank the team, file the final account, move on. That wastes the one moment when the business can check its methods against what actually happened. The estimate, the definition score, the contingency and the schedule model were all forecasts. The outturn is the evidence, and a review that reads it carefully is how the next project starts from a better place than this one did.

This chapter closes Northbank’s case. It sets the outturn against the approved plan, separates what the front end found from what it missed, explains why one good result proves less than it seems, updates the definition checklist, and scales the whole method down to a A$380k project. It ends with a close-out that drew the wrong lesson about contingency from a single result.

The outturn against the approved plan

This is an illustration. The upgrade was ready on Monday 17 January 2028, week 63: rooms at temperature and the grader trialled, two weeks before the deadline and three weeks before first intake. Delivery took 41 weeks against 36 planned, a ratio of 41 ÷ 36 = 1.139. Five of the seven weeks of float were used: two by the slab rework of Chapter 7, two by wet weather during the site works, and one by the grader’s commissioning.

On cost, the business drew A$292k of its A$520k contingency, 56.2%. Delivery cost A$6,472k excluding GST against the approved A$6,700k, and the total including the A$248k front end was A$6,720k, against total funding of A$6,948k. That is 4.5% above the Class 3 point estimate of A$6,428k, well inside the Class 3 range of A$5,785k to A$7,714k.

Figure 1Approved plan against outturn

Measure	Approved at Gate 3	Outturn	Difference
Ready week	58 (13 Dec 2027)	63 (17 Jan 2028)	5 weeks later; 2 weeks before the deadline
Delivery duration	36 weeks	41 weeks	ratio 1.139
Delivery cost	A$6,180k base, A$6,700k with contingency	A$6,472k	A$228k inside the approved budget
Total including front end	A$6,428k point, A$6,948k funding	A$6,720k	4.5% above the point; inside the range
Contingency	A$520k	A$292k drawn	56.2% used
Cost of delay	expected A$36k	nil	ready before the deadline

The outturn against each estimate class

The outturn also tests the estimates made earlier. A$6,720k is 24.2% above the Class 5 point of A$5,410k and 12.8% above the Class 4 point of A$5,960k, yet inside both of Northbank’s ranges for those classes (A$3,787k to A$8,115k and A$4,768k to A$7,748k). The ranges did their job; the points, read on their own, would not have. Had the board approved the rush budget of A$5,950k as a fixed sum, the same final cost would have been A$770k over it, before any of the rush path’s own late scope and delay. That is the case Chapter 4 made for quoting every early figure with its class and range, now seen from the far end of the project.

The contingency was drawn for four reasons, and the split matters more than the total.

Figure 2Contingency drawn, by cause

1. Slab rework, Chapter 7A$130k · 45%demolition A$38k and new slab A$92k
2. Time-related costA$70k · 24%5 weeks at A$14k
3. Stormwater detentionA$64k · 22%element B7 left at level 3
4. OtherA$28k · 10%minor items under A$10k each

Reading the schedule result

The delivery ratio of 1.139 was worse than the model’s median of 1.05. Under the front-end path’s assumed distribution (median 1.05, spread 0.08), the chance of a ratio of 1.139 or less is 84.5%: the outcome sat at roughly the model’s 85th percentile. That is a slightly unlucky project, not an alarming one. About one project in six run this way would do worse.

Two of the five lost weeks were not luck at all. The slab was poured to a concept layout against advice that drawings were three weeks away. Without it the ratio would have been 39 ÷ 36 = 1.083, close to the median. The review records both readings: the model was not contradicted by the outcome, and the one controllable overrun was an error the project’s own rules now prevent.

What the front end found and what it missed

The front end found five items of hidden scope worth A$990k at front-end prices, which would have cost A$1,425k found during delivery. It missed one: the council’s requirement for stormwater detention for the larger roof and hardstand, which surfaced when the civil engineer’s drainage design was lodged during delivery and cost A$64k. By value, the front end found 990 of 1,054, or 93.9%, of the hidden scope that came to light.

The miss was not a surprise. At Gate 3 the independent reviewer had held element B7 (food safety, hygiene and environmental requirements) at level 3, because the stormwater question had not been asked, and the business accepted the risk rather than delay approval. The ready-for-delivery check in Chapter 9 recorded it as an open item. The index did its job: it showed where definition was weak, and the cost of that weakness turned up exactly there.

The 93.9% invites a comparison with Chapter 6, whose model assumed that A$248k of front end would find 89.5% of the discoverable problems. The two figures are not measuring quite the same thing: the model’s share counts expected cost avoided, including delay, while the review’s counts the value of scope items, and there may be hidden scope that never came to light at all. They are close enough to say that the model’s assumption about k was not unreasonable, and no more. One project cannot calibrate a curve; it can only warn if the curve is badly wrong, and this one did not.

One outcome cannot prove a method

It is tempting to write that the front end saved A$731k, the difference between the rush path’s expected A$7,451k and the A$6,720k actually spent. The review does not write that. The A$7,451k is an expectation over many possible rush deliveries, nearly half of which miss the harvest; the A$6,720k is one outcome from one delivery. Comparing them mixes two different kinds of number.

The honest comparison was made at the decision: an expected saving of A$889k for A$248k of front end, on stated assumptions. The review’s task is different. It asks whether those assumptions held up. Here they did: the delivery ratio fell inside the assumed distribution, the scope found matched the scope the front end set out to find, and the one miss was in the element already flagged. A single project cannot show that the assumed medians and spreads are right; it can only fail to contradict them. Over several projects the evidence accumulates, which is why Northbank now records each project’s forecasts before approval, as Deciding, tracking and calibrating the forecast in Reference Class Forecasting by the Numbers recommends, so that its P80s can be checked against how often they hold.

The benefits are reviewed separately and later. The business case assumed A$1,400k a year from the 2028 harvest; whether the grader, the cold store and the energy savings delivered their shares will not be known until the stored fruit runs out in late October 2028. That review compares benefits with their forecasts on the same measures, as The post-completion review and the next decision in Demand and Benefit Forecasts by the Numbers describes.

Updating the checklist and the rules

The review made four changes, each traced to something that happened.

• An evidence rule for B7. Level 2 or better now requires written confirmation of the council’s drainage and stormwater requirements and the food safety requirements that apply, not just the team’s view that they are understood.
• No weight change yet. B7’s weight of 40 was questioned. The review noted that one A$64k miss is too little evidence to reweight a 20-element index, and listed the weight for review after two more projects.
• The deadline defined as ready for use in element C2, from Chapter 8, and the early-works rule from Chapter 7, both written into the procedures.
• Change requests before agreement from Chapter 9: no one tells a contractor to allow for a change before it has been assessed.

Figure 3Close-out questions

• Do: Did the delivery ratio fall inside the assumed distribution1.139, at about the 85th percentile
• Do: Was the outturn inside the estimate's class rangeA$6,720k against A$5,785k to A$7,714k
• Do: What did the front end miss, and was it flaggedstormwater, A$64k; B7 held at level 3
• Do: Which overruns were controllableslab rework, two weeks and A$130k
• Do: Which rules change, and on what evidenceB7 evidence rule; early works; change requests
• Do: When will benefits be reviewed against the forecastafter the 2028 season, late October 2028
• Avoid: Setting the next contingency from this project's unspent balanceone result is not a reference class

Scaling the front end to a small project

This is an illustration. Northbank’s next project is a A$380k bin-washer and hygiene upgrade inside the existing packhouse. The same method applies, scaled to the size and nature of the work.

Fewer elements. The team scores 10 of the 20 elements: A1, A2, A4, B2, B4, B7, B8, C2, C3 and C5. Elements that matter for a new building on new ground, such as site and ground investigation, the building and fire design basis and the planning approvals path, do not apply to equipment inside an existing building. The kept elements’ weights add to 530, so each is multiplied by 1,000 ÷ 530 = 1.887. Scores then still run from 100 to 1,000 and the same threshold of 250 applies.

One gate. FEL 2 and FEL 3 are combined into a single stage of 6 weeks costing A$19k, about A$3.2k a week, ending at one approval gate. The work in it is chosen from the elements: a written statement of the benefit and the throughput the washer must achieve, an electrician’s check of the switchboard and supply capacity, a hygiene and wastewater specification agreed with the food safety auditor and the water authority, quotes for the washer against that specification, and a schedule worked back from the line-change window.

A higher share of cost. A$19k is 5.0% of A$380k, against 3.9% for the cold store. That is normal: some front-end work, such as confirming the electrical supply, writing a hygiene specification and defining the deadline, costs much the same on a small project as on a large one. A small front end that is a larger share of cost is not over-planning if each item answers a question that would otherwise be answered on site.

Figure 4Rescaled weights for the bin-washer

Element	Original weight	Rescaled weight
A1 Business objectives and benefit measures	60	113.2
A2 Throughput and storage requirement	70	132.1
A4 Site and seasonal constraints	60	113.2
B2 Electrical supply capacity and connection	70	132.1
B4 Packing line process and layout	60	113.2
B7 Food safety, hygiene and environmental requirements	40	75.5
B8 Equipment list and long-lead items	40	75.5
C2 Schedule with deadline and float	50	94.3
C3 Cost estimate and contingency basis	50	94.3
C5 Commissioning, handover and pre-season trials	30	56.6

The seasonal logic scales too. The bin-washer must also be installed between November and January, so its deadline, defined as ready for use, governs the plan in the same way, and an early start before its design basis is frozen carries the same risk as the slab.

Running the review

Northbank held the close-out review three weeks after readiness, before the harvest absorbed everyone’s attention and while the people involved still remembered why decisions had been made. The project engineer prepared the comparison of forecasts and outcomes; the finance manager, the operations manager and the packhouse manager attended; and the independent reviewer who rescored the index at Gate 3 was invited back to read the draft. The review looked at decisions as well as results: whether each was reasonable on what was known at the time, which is a different question from whether it turned out well.

The error: contingency read as too generous

This is an illustration. The draft close-out, written by the operations team, noted that A$228k of contingency had not been spent and concluded that 8.4% had been “too generous”. It proposed a contingency of 4% of the delivery base for future projects.

The finance manager tested the proposal against the project it came from. Four per cent of A$6,180k is A$247k. The project had actually drawn A$292k, so the proposed rule would have left Northbank A$45k short on the very project used to justify it, and that on a delivery that finished at the model’s 85th percentile rather than in its tail. The slab rework and the stormwater item alone, A$194k, would have used most of a 4% allowance before any weather or time-related cost.

The error was reading one outcome as if it measured the risk. A contingency is set before the work, for the range of outcomes the project might have; an unspent balance shows only which outcome occurred. The correction was to set contingency from the estimate’s class and the definition score, as Chapter 4 did, and to review the guideline only against a record of several projects. From estimate to commitment: keeping the range and setting rules for money held back sets out rules for holding and releasing that money. Northbank’s close-out was rewritten to say that contingency was adequate for an ordinary outcome, and that the unspent A$228k was returned to the business rather than taken as evidence for a smaller allowance.

Exercise: scoring the bin-washer

Score the bin-washer project with the rescaled weights above and these levels: A1 level 1, A2 level 1, A4 level 2, B2 level 2, B4 level 1, B7 level 3, B8 level 1, C2 level 2, C3 level 2, C5 level 2. Use the factors from Chapter 3 (level 1 0.10, level 2 0.30, level 3 0.55). Does it pass the threshold of 250, and where should the remaining front-end work go?

Worked answer. Each contribution is the rescaled weight times its level’s factor: A1 113.2 × 0.10 = 11.3; A2 132.1 × 0.10 = 13.2; A4 113.2 × 0.30 = 34.0; B2 132.1 × 0.30 = 39.6; B4 11.3; B7 75.5 × 0.55 = 41.5; B8 7.5; C2 94.3 × 0.30 = 28.3; C3 28.3; C5 56.6 × 0.30 = 17.0. The score is 232.1 (the rounded contributions add to 232.0), under 250, so the project passes. The largest contributions are B7 at 41.5 and B2 at 39.6. B7 is the element that let the stormwater item through on the cold store, and on a hygiene project it carries the main requirement; raising it to level 2 under the new evidence rule would cut the score to 213.2. That is where the remaining front-end effort belongs, ahead of any further refinement of elements already at level 1.