Asset managers do not write lifespan tables. Inspectors do.

A boiler between thirty and one hundred thirty kilowatts has a lifespan of eighteen years. A clay tile roof in southern England lasts sixty to one hundred. A copper water pipe in Type K can reach a century; Type M in acidic water might die at forty. AppKeep’s multi-year plan projects replacement costs across twenty-one house components and seventy-eight variants, and every one of those variants has a number attached. Multiply remaining life against replacement cost, add them up by year, and the plan tells the homeowner which year is going to bite.

For months those numbers carried a flag that said “Sander to validate.”

Sander is the domain expert co-founding AppKeep. Twenty-plus years in property management, runs AAH Oy. When we sat down this week to write his review email, we noticed something: Sander is an asset manager. He is not a building inspector and he is not a material scientist. Asset managers own the framework, the decomposition, the decisions that follow from the data. They do not write lifespan tables. Inspectors and material scientists do.

Asking him to confirm that a copper pipe lasts forty to seventy years was borrowing his credibility for the wrong thing. The numbers have to come from research. His judgment applies one level higher: does the way we decompose a house into variants match how asset managers think? Are we asking him the right questions?

What we did this week

We treated the seventy-eight lifespans as a research problem, not a review problem.

We downloaded the 2012 Rijksvastgoedbedrijf BOEI inspection handbook. That is the document the Dutch state uses to run capital planning on its own buildings, and it cites NEN 2767 for every theoretical lifespan it publishes. Hoofdstuk 4 in each of the three discipline volumes (Bouwkunde, Elektrotechniek, Werktuigbouw) is a structured table: per-component cycle, in years, with an asterisk marking components where the “vangnet” method applies. Vangnet is a professional technique for scoring a component’s condition when no visible defect exists but age says it is near end of life. Roughly one hundred and sixty authoritative figures came out of the handbook, most directly applicable to residential buildings.

Then we cross-referenced a sample of fifteen of our seventy-eight variants against the Dutch figures and five other international sources: the US NAHB 2007 study, InterNACHI, the UK BRE and RICS publications, CIBSE Guide M for mechanical equipment, and UK and Irish trade associations. Every variant got three to five independent data points and a confidence label.

Three adjustments came out of the sample.

Clay tile roofs moved from forty-to-eighty up to sixty-to-one-hundred. The previous floor was below every manufacturer warranty and every UK roofer-trade figure we consulted. Steel panel radiators moved from twenty-five-to-forty down to twenty-to-thirty-five because the UK trade consensus clusters tighter and our previous range trended toward cast-iron territory. Mechanical ventilation with heat recovery moved from fifteen-to-twenty-five down to fifteen-to-twenty because our upper bound had borrowed from commercial air-handling-unit data that does not cleanly represent residential MVHR.

Of the fifteen sampled variants, nine came out high confidence and stay put. Three are medium. Two are low and one is unverified. Air-source heat pumps are tagged for a three-year refresh cadence now, because the technology cohort is still changing faster than field data can catch up. Internal paint carries a note about the definitional split between technical film life (twelve years, per BOEI and NAHB) and practical redecoration (five to ten, the residential replacement cycle).

What stays in the research, not in the review

The interesting finding is one level deeper than the numbers. The Dutch handbook encodes the “cliff at seventy-five percent” that Sander describes in his own framework as an explicit mathematical function: t = L − L × ½^(C−1), where L is theoretical lifespan, t is age, C is condition score one through six. Condition two corresponds to fifty percent of life spent. Condition three to seventy-five percent. Condition four to eighty-seven-and-a-half. Five at ninety-three-point-seven-five. Six at one hundred asymptotically. The derivative of this function grows without bound as age approaches the lifespan. The cliff is the first derivative of the halving function. It is encoded in the professional standard. It is not folklore.

The BOEI also formalises the vangnet principle as a condition-scoring rule: when a component is past half its lifespan with no visible defect, the inspector scores it condition two. Past three quarters, still no visible defect, condition three. This is exactly the signal AppKeep wants to surface to a homeowner. “No visible problem. Statistically near end of life. Schedule a replacement this year and you avoid the crisis.” Dutch professional practice has a name for this principle. We have been reinventing the word.

What we are asking Sander this week

The numbers are mostly settled. The four things Sander is uniquely positioned to answer are all at the framework level:

1. Should engineered wood flooring and solid hardwood be separate variants, or one range? Solid hardwood is a hundred-year product, engineered is a thirty-year product. Does asset-level planning need both?
2. Interior paint: scheduled lifecycle item, or always taste-driven? Which number drives capital reservations?
3. Air-source heat pumps in Finnish, Dutch, and UK residential installations: do the newest-generation inverter systems land closer to fifteen or twenty in practice?
4. MVHR in Finland. UK trade says fifteen-to-twenty. Finnish homes have run MVHR as standard for decades. What does field experience say?

Plus one question about the decomposition itself: we split structural variants by material (brick, render, timber, metal) and equipment variants by product type (combi, system, air-source, ground-source). Does that cut match how asset managers structure capital plans?

What changed in the data

Every one of the fifteen audited variants now carries the date of its last cross-reference, a confidence tag (high, medium, low, unverified), and a short note explaining what the sources disagreed about and why. When the next refresh comes, the audit trail is already there.

What we stopped doing this week: asking the domain expert to validate things that research answers better. What we started doing: writing source citations into the data itself.