First-of-a-Kind Technology by the Numbers
Chapter 1 of 10 · 11 min read
The price of being first: a proven line against a novel process
A recycler's proven A$16.0m washing line against a novel A$17.5m process: paybacks of 3.39 and 2.45 years, a A$13.8m paper NPV lead, and why that lead is still only a claim.
A new process usually arrives with a better number attached: higher yield, lower energy, less waste. The number is often honest, in the sense that someone measured it somewhere. What it rarely carries is a statement of where, for how long, on what feed and at what size. This chapter sets out the decision that runs through the book, puts the two options side by side exactly as the first board paper did, and explains why a large lead on paper is a claim to be tested rather than a result to be banked. It also defines the two terms the book depends on, first of a kind and nth of a kind, and summarises what the published evidence says about the difference between them.
The recycler and its decision
This is an illustration. A family-owned plastics recycler in Perth’s southern industrial area, with about 55 staff, collects, sorts and bales post-consumer rigid HDPE and PP packaging, the milk bottles, detergent containers and tubs that households put in their recycling bins. For years it sold the bales to processors, some of them overseas. The waste plastic export rules made under the Recycling and Waste Reduction Act 2020 changed that: exports of mixed waste plastics not sorted into a single resin or polymer were stopped from 1 July 2021, and exports of unprocessed single-resin or single-polymer waste plastics from 1 July 2022. The recycler’s bales now go to a smaller domestic market, and its margin on them has narrowed. (Anyone relying on these rules should check the current requirements with the responsible Australian Government department.)
In March 2024 the board decides to add value itself, by building a washing-and-sorting line that turns 18,000 t a year of its own and bought-in bales into washed, sorted flake that processors can extrude into pellets. The design feed is 3.0 t/h for 6,000 operating hours a year, or 1,500 t a month. The site lease runs to February 2036, so the appraisal horizon is 144 months from the decision. All amounts in this book are Australian dollars excluding GST, in real 2026 dollars, discounted at 8% a year real.
The decision owner is the managing director. The operations manager will run whatever is built, the process engineer leads the technical assessment, the finance manager owns the business case, and the board has engaged an independent process engineer as reviewer. Two suppliers have made offers.
- 1 July 2021Mixed waste plastic exports stoppedplastics not sorted into a single resin or polymer
- 1 July 2022Unprocessed single-resin exports stoppedthe recycler's bale market narrows
- March 2024Month 0: the board decisionchoose a washing-and-sorting line for 18,000 t a year
- July 2025Month 16: Option A planned first saleable flakeproven line
- August 2025Month 17: Option B planned first saleable flakenovel process
- February 2036End of the site leasethe 144-month appraisal horizon closes
The two options on paper
Option A is a proven line from an established builder that has supplied 41 similar lines, 23 of them on rigid polyolefin feed like the recycler’s. It uses the conventional sequence: bale breaking, shredding and granulating, a pre-wash, friction and hot washing, sink-float tanks that separate the floating polyolefins from heavier plastics and dirt, conventional water treatment, then drying, near-infrared flake sorting and silos. The builder guarantees 72% yield, measured dry, on feed with no more than 10% contamination.
Option B is a novel process from a small Australian technology company with 14 staff. It replaces the sink-float tanks with a hydrocyclone wash, which separates by density in a spinning flow and recovers finer particles; it recycles its wash water through a closed-loop water treatment system; and it adds a hyperspectral fines sorter that recovers small flakes a conventional line sends to residue. The company claims 84% yield. Its evidence is a 0.30 t/h pilot plant, one tenth of the design rate, which has run for 400 hours.
The first board paper compared the two as follows. Contribution here means flake revenue at A$1,050/t, less the bales’ opportunity cost of A$250/t (the price the recycler could still get by selling them), variable operating cost, residue disposal at A$160/t and fixed operating cost.
| Option A, proven | Option B, novel | |
|---|---|---|
| Capital cost estimate | A$16,000,000 | A$17,500,000 |
| Planned months, decision to first saleable flake | 16 | 17 |
| Yield | 72% guaranteed | 84% claimed |
| Variable cost per t of feed | A$110 | A$115 |
| Fixed cost per year | A$1,600,000 | A$1,700,000 |
| Flake at nameplate | 12,960 t | 15,120 t |
| Residue to disposal | 5,040 t | 2,880 t |
| Contribution at nameplate | A$4,721,600 a year | A$7,145,200 a year |
The difference is almost entirely yield. Twelve extra points on 18,000 t of feed is 2,160 t of flake, worth A$2,268,000 at A$1,050/t. Those tonnes also stop being residue, saving A$345,600 of disposal. Against that, B’s variable cost is A$5/t higher (A$90,000 a year) and its fixed cost A$100,000 higher. The net is the A$2,423,600 in the figure.
A useful number to carry through the book is the value of one yield point. At full feed one point is 180 t of flake, and each tonne gains the flake price and avoids the disposal charge: 180 × (1,050 + 160) = A$217,800 a year. The bales have already been bought or diverted, so the extra flake costs nothing more to make. Bohn and Terwiesch (1999) argued that yield matters most in exactly this situation, where the process rather than demand limits output: a lost point is lost saleable product, not just wasted material.
On these numbers the choice is not close. B pays back its extra capital in about seven and a half months of full output and nearly doubles the net present value. The board paper recommended B.
First of a kind and nth of a kind
The two options differ in kind, not only in yield. This book uses two terms for that difference.
A plant is nth of a kind when its design, or a design close enough that the differences do not matter, has already run at full scale in the same duty: the same kind of feed, the same product, similar size and similar operating hours. Option A is nth of a kind. Its builder has done this 41 times, and the owners of earlier lines have already paid for the lessons: the screens that blinded, the pumps that wore, the controls that needed retuning.
A plant is first of a kind when that is not true for at least one part of it that matters. Novelty does not need to be total. Option B’s bale breaking, shredding, granulating, washing and drying are ordinary equipment. What has never run at full scale in this duty is the combination of the hydrocyclone stage, the closed water loop and the fines sorter, and those three elements decide its yield. A plant can therefore be mostly proven and still be first of a kind, because the evidence for the whole is limited by the evidence for its least proven part. Chapter 2 makes that idea measurable.
The distinction is about evidence, not quality. A first-of-a-kind design may well be better than the proven one; most proven designs were first of a kind once. The point is that the owner of the first full-scale plant is the one who finds out how much better, and pays for the finding out.
What the evidence says in general
The question of whether new technology carries a predictable penalty has been studied for decades, in fields from energy to defence.
The most direct evidence for process plants comes from a series of RAND studies. Merrow, Phillips and Myers (RAND R-2569-DOE, 1981) analysed detailed proprietary data from 44 process plants built by 34 oil, chemical and mineral companies. They found that most of the variation in cost-estimation error was explained by how far a plant’s technology departed from that of earlier plants, how well the project and site were defined when the estimate was made, and how complex the plant was. Most of the variation in performance after start-up was explained by new technology and by whether the plant processed solids. Chapter 3 applies those findings to the recycler; for now it is enough that both options handle wet solids, and only one adds new technology.
A second line of evidence comes from public audits of large technology programmes. In Defense Acquisitions: Assessments of Selected Major Weapon Programs (GAO-06-391, 2006), the US Government Accountability Office reported that, among the programmes it assessed, those that began development with mature technologies averaged 4.8 per cent research and development cost growth, against 34.9 per cent for those that began with immature technologies. These are US defence programmes, and the percentages are not a rate for a recycling plant in Perth. They are cited for the direction and rough size of the effect: committing to cost and schedule before the technology is mature is associated with much larger growth.
Neither source says that new technology should be avoided. Both say that the immaturity of a technology is visible before commitment, and that its cost shows up later if it is ignored.
Why B’s lead is a claim, not a result
Every line in the paper comparison is an input, and the inputs are not equally supported. Option A’s 72% is a contractual guarantee from a builder whose earlier lines run on similar feed. Option B’s 84% is a measurement from a pilot one tenth of the size, on one feed chosen by the supplier, over 400 hours. Both can be true at the same time as the full-scale B plant on the recycler’s own bales does something different.
The paper comparison also assumes, for both options, that the plant reaches nameplate output in its first month, that the capital estimate is the final cost and that the schedule holds. For a proven line those assumptions are optimistic but not far wrong. For a first-of-a-kind plant they are the assumptions most likely to fail, and the research above says why: cost grows with departure from earlier plants, and performance falls short with new technology handling solids.
This is the inside view: a forecast built from the details of the plan in hand, without asking how plans like it have turned out. The inside view and the plan on the table in Reference Class Forecasting by the Numbers explains why such forecasts run optimistic and how outcomes of similar projects correct them. Benefit forecasts suffer from the same bias as cost estimates, and The Hiding Hand and the evidence on optimistic forecasts in Demand and Benefit Forecasts by the Numbers treats the benefit side. The difficulty here is that Option B has no reference class: there are no earlier full-scale B plants to learn from. The rest of this book builds a correction from the parts instead, element by element.
The error: two numbers presented as equals
This is an illustration. The first board paper put A’s guaranteed 72% and B’s claimed 84% in the same row of the same table, in the same typeface, as if they were equally supported. The capital, schedule and cost lines were treated the same way.
The finance manager, reading the draft, asked a simple question: what evidence stands behind each number? The process engineer added a column to answer it. For Option A’s yield the column read: guarantee in the contract, dry basis, feed up to 10% contamination; 41 lines built; 8 reference plants contacted. For Option B’s yield it read: one 0.30 t/h pilot, 400 operating hours, one supplier-selected feed, no full-scale plant. One input rested on eight operating plants of the same kind, the other on a single short pilot.
Nothing in the arithmetic was wrong; the error was in presentation, and it would have carried straight into the board’s decision. The correction was to give every input an evidence class and carry it through the analysis: contractual guarantee, full-scale reference plants in the same duty, full-scale plants in a different duty, pilot at reduced scale, or estimate without test. Every later chapter adds to that column, and Chapter 10 returns to it when the board decides.
- Do: Option A yield 72%contractual guarantee, dry basis, feed up to 10% contamination; 8 reference plants contacted
- Do: Option A capital A$16,000,000firm offer for a builder's 41st line
- Do: Option A schedule, 16 monthsthe builder's record on earlier lines
- Ask: Option B yield 84%one 0.30 t/h pilot, 400 h, one supplier-selected feed; measurement basis not stated
- Ask: Option B capital A$17,500,000estimate for a plant never built at full scale
- Ask: Option B schedule, 17 monthsno full-scale plant to compare with
- Ask: Both options, nameplate from the first monthpaper assumption, not evidence; ramp-up is the subject of Chapter 7
The route through the book
The chapters correct the paper case one input at a time, and each correction has its own method.
Chapter 2 rates the readiness of each critical element of Option B on the nine-level technology readiness scale and shows why the system is only as ready as its least mature element. Chapter 3 uses the pioneer-plant research to locate where cost growth and performance shortfalls should be expected. Chapter 4 places each supplier on its experience curve and asks who pays for the learning. Chapter 5 compares the pilot with the full-scale design, flow by flow. Chapter 6 sizes a qualification test and rebuilds the 84% claim on a measurement basis the recycler can bank. Chapter 7 models ramp-up to nameplate output. Chapter 8 grades cost and schedule allowances by readiness. Chapter 9 values ways of buying the risk down, and Chapter 10 makes the choice and follows it into operation.
As a preview, when every input is replaced by its median estimate rather than its paper value, the net present values become A$13,279,515 for Option A and A$10,257,528 for Option B. The A$13.8 million lead does not just shrink; it reverses, and A leads by about A$3.0 million. The largest single correction is the yield itself, which Qualification testing: run length, feed envelope and yield basis rebuilds from the pilot’s own measurements.
Recovery is a design problem as much as a purchasing one. Turning waste into a resource: why recovery is a process design problem makes the general case for counting quality-adjusted yield rather than gross yield, which is where Option B’s number will turn out to be weakest.
Exercise: a smaller claim
Suppose Option B’s yield were 80% instead of 84%, with everything else on the paper basis unchanged. What is B’s contribution at nameplate, how far is it above Option A’s, and at what yield would B’s contribution equal A’s?
Worked answer. At 80%, B makes 18,000 × 0.80 = 14,400 t of flake and 3,600 t of residue. Revenue is 14,400 × 1,050 = A$15,120,000. Costs are bales 18,000 × 250 = A$4,500,000, variable 18,000 × 115 = A$2,070,000, residue 3,600 × 160 = A$576,000 and fixed A$1,700,000. Contribution is 15,120,000 − 4,500,000 − 2,070,000 − 576,000 − 1,700,000 = A$6,274,000 a year, still A$1,552,400 above A’s A$4,721,600. B’s contribution falls by A$217,800 for each yield point, so it equals A’s when the yield is about 7.13 points lower than 80%: solving 18,000 × y × 1,210 = 4,721,600 + 4,500,000 + 2,070,000 + 2,880,000 + 1,700,000 gives y = 72.87%. On paper, B’s yield can fall a long way before A wins. The question the rest of the book answers is how much of the yield claim survives, and what else B gives up on the way.