The whole technique of insurance rests on a simple idea: the past informs the future, because the same risk, repeated many times, produces a stable frequency and average cost. The module on prototypes and series showed what becomes of that idea when the technology is new. What remains is what one actually does when there is nothing to look at, because that has been the ordinary situation of large energy projects for fifteen years: the first floating wind farm, the first electrolysis unit at that scale, the first battery storage installation at that density. Underwrite or decline are the only two answers, and declining is not a tenable answer when the whole market is moving toward those assets.
The first resource is structural analogy, and it is better than its reputation. A new asset is almost never new all the way through: it is an assembly of elements each of which has a history. An offshore wind turbine is a marine civil engineering foundation, a rotating machine, power electronics and a subsea cable, and each of those four families has decades of losses behind it, in other industries. The useful reasoning is therefore to break down rather than to look for a global comparable, then to treat separately the one item genuinely without precedent, which is usually the interface between two known elements.
That, indeed, is where the losses sit, and it is the most solid lesson of those fifteen years. Failures of new technologies rarely come from the new component itself, which has been tested, calculated and monitored precisely because it was new. They come from the junctions: the cable at the entry to the foundation, the fixing of a conventional item to an unusual structure, the behavior of proven electronics in an environment they had never met. Attention naturally goes to what is visibly novel, and the loss happens on what was thought ordinary, at the exact point where the ordinary met the novel.
The second resource is scenario reasoning, which replaces a distribution one does not have with a few hypotheses one writes down. One no longer looks for an average frequency, one builds three or four plausible sequences, prices them, and looks at what each would do to the program. It is less precise than a statistic and it is more honest, because the imprecision stays visible instead of being buried in a decimal. The method's weakness is known: it says nothing about what was not imagined, and on a new technology the unimagined share is precisely the largest. It is offset by keeping the scenarios open and revising them at each incident in the sector, including those that happen to others.
The third resource is the only one that produces data instead of consuming it: instrumentation. A properly instrumented new asset manufactures its own history, and it does so quickly if anyone listens. Vibration, temperature, current, strain, charge cycles: those measurements almost always exist for operational reasons, and they contain the frequency of precursors well before the first loss occurs. The question is therefore not whether the data exists, it does, but whether anyone looks at it from the risk side rather than the yield side, and whether it crosses the boundary separating the operator from its insurer, which is a question of trust and not of technique.
The symmetrical error must be named, because it is as costly as excessive caution. A new technology is not necessarily more loss-prone than a mature one, and sometimes it is less: it is designed to current standards, closely monitored, operated by attentive teams because nothing is yet routine. What makes it difficult is not its level of risk but the uncertainty about that level, which is an altogether different thing. Treating uncertainty as though it were high risk means confusing what one does not know with what one fears, and leads to turning away assets that would have behaved well.
The last difficulty is one of time, and it is peculiar to series. A series technology deploys faster than experience accumulates: by the time the hundredth machine is installed, the first has only four years of running, and the failure modes that appear at eight years are still unknown while the whole fleet is already in place. History always builds up behind exposure, and that lag is not a flaw of method, it is the structure of the problem itself. It explains why the aggregation question, treated in the module on series, arises so sharply: it is not only that the machines are identical, it is that they are all younger than the defect awaiting them.
An underwriter looks at a 400 megawatt-hour battery storage site using a cell technology deployed for eighteen months, on a single location. There is no loss history on this cell generation. Three items are on the table: an operator report showing zero incidents over the first fourteen months of a 40 megawatt-hour pilot site; a technical file describing container spacing and the detection system; and an offer from the operator to share cell-by-cell temperature measurements. How is an opinion built?
The zero-incident report is the most attractive and least informative of the three documents, and seeing it that way is the first move. Fourteen months on a site ten times smaller says almost nothing: the absence of an event on a short sample is consistent with a low frequency as much as with an average one, and above all cell failure modes appear after several hundred full cycles, which fourteen months of running have not produced. The technical file is more useful, but not for the expected reason: the point is not to judge the cell, about which nobody knows anything, it is to judge what happens when a cell goes, that is, the spacing, the compartmentation and the detection delay, which determine whether the event is one container or the site. That is scenario reasoning applied to what is known, around what is not. That leaves the offer to share measurements, the only one of the three items that produces new data, and the one usually set aside because it fits no box in the submission. On a technology with no history, a cell-by-cell temperature feed is worth more than an empty loss record: it manufactures the missing history at the pace of operation, and it manufactures it before the first loss rather than after.
- 01A new asset is an assembly of elements each with a history: breaking it down beats looking for a global comparable, and isolates the one item genuinely without precedent.
- 02Losses on new technologies sit at the junctions, not in the new component: that one was tested and monitored precisely because it was new.
- 03Scenario reasoning replaces an absent distribution with written hypotheses: less precise than a statistic, more honest, and blind to whatever was not imagined.
- 04An instrumented asset manufactures its own history before the first loss: the data almost always exists, the question is who looks at it from the risk side.
- 05Uncertainty is not high risk: a new technology is sometimes less loss-prone than a mature one, and confusing what one does not know with what one fears turns away assets that would have behaved well.