A week in the life: how the PPM agent fits a UK and a Dutch building manager

Two building managers, two countries, two vocabularies, one week of work. Neither of them wants a new way of working. What follows is a walk through their existing week, with the point marked where the agent does its one job — and the points where it deliberately does nothing.

::: How to read this These are composite pictures drawn from how UK and Dutch estates commonly run planned preventative maintenance, and from what the platform does today. They are not named customers, and nothing here is a case study of a real client. :::

Claire, UK — 41 sites, statutory returns, one inbox

Claire manages a mixed portfolio for a managing agent: two office estates, a leisure site, and a long tail of small retail units. Her week is shaped by returns that must exist — gas, fixed wire, emergency lighting, fire alarm, lifts, water hygiene — and by the fact that half of them arrive as a PDF from a contractor, not as a result in a system.

How the week already runs

Monday morning is a reconciliation. What was due last week, what came back, what came back with observations on it. Contractors email reports. Her assistant files them. Anything with a C1 or C2 on it, or a failed emergency light, becomes a job — if somebody notices it in the document.

That last clause is the problem, and it is rarely a competence problem. It is a volume problem: an engineer's remedial note sits on page nine of a sixteen-page report, and the report arrives in the same inbox as everything else.

Where the agent fits

Nothing about Monday changes. Reports still arrive, still get filed against the property and the asset. What changes is what happens after a return is signed off.

1. A contractor's return is filed and Claire — or whoever attended to it — signs it off with the date it was carried out and a named person. Until that signature exists, the platform treats the return as unsigned, and unsigned is not wrong; it just means nobody has stood behind it yet.
2. The findings recorded on that return sit in the register with a severity.
3. The agent reads them on a rhythm Claire's administrator sets. For each finding at or above the severity threshold, on a signed return, with no task already against it, it opens exactly one investigation task. Unassigned, or handed to somebody on the roster who accepts automatic work.
4. Everything it does not raise, it refuses in writing, with the reason and the boundary it read. A below-threshold observation is not silently dropped — it is visibly refused and still sitting there.

What Claire actually gains

Not automation of her judgement. A shorter distance between page nine of a PDF and a task with her name on it. When her client asks in the quarterly meeting what happened to the three C2s from the summer fixed-wire test, the answer is a list with dates on it, not a search of an inbox.

What it still will not do

It does not close a finding. It does not sign a return. It does not tell a contractor anything. It does not decide that a risk is acceptable. If a return has no carried-out date, nobody can sign it, and the findings behind it stay refused — the platform will not invent the date, and neither should anyone else.

Joost, Netherlands — een VvE, twee kantoren, een meerjarenplan

Joost runs a small Dutch portfolio: one VvE with 68 apartments, two office buildings, a parking deck. His year is shaped differently. Instead of a rolling wall of statutory returns, he has a conditiemeting under NEN 2767, an inspection round framed by BOEI — brandveiligheid, onderhoud, energie, inzicht in wet- en regelgeving — and a long-horizon plan that the owners' meeting has to approve before money moves.

How the year already runs

An inspector walks the buildings. Elements get a condition score, one to six, with the defect, its intensity and its extent behind the number. That survey becomes a plan: what to do, when, and what it costs over ten, twenty, twenty-five years. The ALV approves it, or amends it, or defers it.

Between those cycles, things break, and things get noticed. A flat roof detail flagged during the round. A lift engineer's remark. A fire door that does not latch. These sit somewhere between "in the plan" and "somebody should look at that", and that gap is where Dutch estates lose time exactly as British ones do.

Where the agent fits

Again: not in the plan, and not in the survey. The agent does not score a condition, and a severity in the register is never dressed up as a classified NEN 2767 score — those are two different things, produced by two different acts, one of which requires a qualified person in the building.

It works on the same narrow seam as it does for Claire. An inspection has been signed off with a date and a name. Findings on it carry a severity. Anything at or above the threshold that nobody has triaged gets one investigation task, so the flat roof detail is not waiting on a memory. Everything else is refused in writing, so the fire door nobody raised is visible as "not raised", not as "fine".

What Joost actually gains

A defensible middle. Before the next ALV he can show what was found, what was investigated, by whom, and when — separately from what the plan claims. A plan is a projection; the investigation log is what was done. Keeping them apart is what makes both of them believable.

What it still will not do

It does not touch the meerjarenplan. It does not move money. It does not produce a conditiescore. It cannot decide that a measure should be brought forward — that is a decision for people, and in a VvE it is a decision for the meeting.

The same seven checks in both countries

The agent is a rules engine, not a model. It calls nothing, sends nothing outside the platform, and stops at the first check that refuses:

1. Is the agent switched on for this workspace? It is off until an administrator turns it on.
2. Is the stop switch free? A pull stops everything immediately, and needs no explanation.
3. Is the inspection behind the finding signed off, with a carried-out date and a named person?
4. Does the severity sit inside the boundary the workspace set?
5. Is there already a task against that finding? One finding never collects two.
6. Are the per-run and per-day ceilings still free?
7. Who does the task go to — left unassigned, or handed to a roster member who accepts automatic work?

Every decision and every refusal is written to an append-only ledger with the boundary it read and the moment it ran. Each refusal links to the exact rule that stopped it and what would change the answer.

Words this use case refuses to blur

• Signed off is not verified. A signature records who attended and when, not that the building is safe.
• Raised is not investigated, and complete is not verified. Three separate acts, three separate records.
• Refused is not resolved. A refused finding is still a finding.
• A severity is not a condition score. NEN 2767 scoring is an assessment made by a person in the building.
• Unknown is never nought. A figure nobody recorded is shown as missing.

Where to look next

The mechanics, the boundary and the readings this work moves are set out in the briefing for consulting practices, with a version you can download and hand to a client. The wider discipline is covered in planned preventative maintenance, and the statutory side in property compliance.