The survey was right about every pipe it saw
It never said which ones it hadn’t. What a contractor needs to know before the next visit, and why the fix starts with the record.
We filmed 60 percent of the risers in a 1948 building and decided to replace all of them. The 40 percent we skipped were the ones replaced in 2009. The problem was not the survey. It was that nothing in the record said where the survey had stopped.
A landlord sent a roofer to investigate a leak. The roofer arrived thinking he was there to repair tiles. Once inside, it was obvious that a plumber was needed, and another appointment had to be arranged.
The case comes from the English Housing Ombudsman’s 2025 report Repairing Trust. In another case, a plasterer was sent to repair the damage before anyone had checked whether the leak that caused it had been fixed. The resident lost a day’s wages waiting in for a visit that could achieve nothing. In the landlord’s internal emails, someone asked why there was no record of the leak and pointed out that they were in effect paying twice for one job.
These are complaint cases, so they cannot tell us how common this is. They show what happens when a visit is booked without establishing what needs to be checked, or whether the previous work is complete enough for the next job to begin. Contractors described arriving without information about the repair or earlier attempts to fix it. One said they were lucky to get a contact number. Another described turning up to plaster a wall that was still wet from the previous job.
The landlords in that report are English social housing providers, many of them with thousands of homes. We have a handful of buildings, and we recognised the pattern anyway, in one of our own this year.
The risers we thought we knew
The building is our most recent acquisition. It is from 1948 and came with no documentation of its plumbing. During due diligence we asked about the plumbing and were told the soil and water risers were original, so we planned on that basis. When we began planning a full riser replacement, we commissioned a CCTV survey covering roughly 60 percent of the risers. We chose a partial survey because we thought it would tell us the general condition of the system. The pipes we saw were in poor enough shape that the conclusion seemed clear: the whole building needed new risers.
We brought in a contractor to produce a feasibility study, work out the method and price the job. During the site visit, we found something the survey had missed. Five of the twelve apartments were served by a separate riser system that had been replaced around 2009. Nobody had told us, no document said so, and the 40 percent of pipes we had not filmed were the ones that would have shown it.
The finding changed the scope of the project considerably. It also changed how we read the survey report. The report accurately described the pipes it covered, but nowhere did it identify which risers had not been inspected. We had decided to replace everything on the strength of a report that covered only part of the system. The feasibility study is still under way as we write this, with a different scope.
What the contractor needed to know
When we asked the contractor to price the replacement, we gave them the survey and our assumption that everything was original. A better brief would have said which risers had been filmed, which had not, what condition the inspected pipes were in, and that the plumbing history before our ownership was unknown. That would have directed attention to the risers nobody had examined. The contractor found them anyway, but only after we had already made the replacement decision.
The same applies to a repair. A report that the heating has failed again means something different if a valve was replaced last week, if replacement was only recommended, or if the last visit ended without access. The latest report cannot tell the contractor which of those situations they are walking into. A useful history records what was reported, what someone found on site, what work was recorded as done, and what remains unconfirmed, each with a date and a reference to the original note. It also needs to say what was not inspected. After the risers, we treat that as the most important line in any inspection we commission.
Two components, not one
Our operations system is built around components. Each building has a register of the things that need looking after, with a type, a status and an installation year where we know it. The risers of the 1948 building entered that register as one component. Inspections and work orders attach to the component they concern, keeping the recorded work together. Each job carries the requirement it was created to meet, frozen when the order was raised.
In that structure, the CCTV survey should be a work order on the riser component. Its record needs to say which risers were inspected, how far the inspection reached and what was found. The component record also needs to describe the full system, so that someone reading the two can see how much the survey covered. Completing that work order would then leave a record of a partial inspection, with its limits visible.
The site visit also showed us that the register needs two components for these risers. Five apartments have one system and seven have another, with different ages and conditions. Each needs its own history from here on. We have met this before: when we inventoried our other building, the ventilation turned out to include both a mechanical system and a natural-draught system. A single entry labelled ventilation could not describe both properly.
We also designed the system so that fault reports and work orders could be linked to the individual apartment, allowing a unit’s history to accumulate. Had the pipe replacement been recorded against those five apartments, and had that record reached us, we could have identified the newer system before commissioning the survey.
We could build the system ourselves because we could define the components in our properties and decide what evidence a completed job needed. The riser project showed us where those definitions still needed work. A single component entry concealed two systems, and the inspection requirement needed to account for the pipes left out of the survey.
The history we couldn’t find
In our case, we could not find the history at all. Someone replaced the risers in five apartments around 2009, and whatever record was made did not follow the building to us. That is a different problem from a history scattered across an old ticketing system, contractors’ invoices, inspection PDFs and people’s memories. The Ombudsman heard from operatives who received jobs through WhatsApp on personal phones, outside any system that could track follow-up work. Commercial property respondents to the 2026 RICS survey also identified data quality and integration with existing systems as barriers to using AI.
Before starting a data-cleaning project, it helps to separate information that exists in an awkward form from information nobody recorded. AI can help recover the former. Finding out what was never recorded may require a visit. For us, it took a contractor standing in a stairwell.
For an operator with years of notes, a reasonable first output from a model is a table connecting each recorded intervention to an apartment or component, with the original text alongside. It might bring an old recommendation back into view or show that several entries describe the same visit. Where an answer is unknown, the table needs to say so. When we inventoried our properties, an installation year we did not know became an item on the next inspection round instead of a number somebody guessed. Before we ordered the riser survey, a history of our building should have shown “plumbing history before acquisition: unknown” against each apartment. That uncertainty belonged in the survey brief.
We would test this on one recurring problem whose history someone on the team already knows well, so they can spot a missing visit or an unrelated issue merged into the record. The work should take place in an AI environment approved for the material, without personal details the task does not need. Putting the history together by hand may be straightforward, and a well-kept maintenance system may already show it. That is what the AI version has to improve on. Time spent gathering files, correcting the output and checking sources belongs in the comparison.
What the record cannot tell you
A record can only carry what someone put into it, and the reader has to resist filling in the gaps. Our survey report is the clearest example we have. It recommended replacing the pipes it had inspected. Read quickly, that becomes “replace the pipes”, and that is how we read it. Any history built from that report, whether by a person or by a model, has to keep the recommendation the exact size it was: these pipes, not all pipes.
The same discipline applies to the small entries that make up an ordinary repair history. A note saying “valve replacement recommended” tells you that someone thought the valve should be replaced. It does not tell you that it was. An invoice showing that a plumber attended tells you that a plumber attended, not what they did. A tenant reporting that the heating has failed again tells you the symptom is back, not why. Each of those entries is useful, as long as nobody reads it as more than it is.
When the evidence that a job was finished is missing, the honest entry is “unconfirmed”. Calling the job unfinished would be a guess in the other direction. Someone may need to call the contractor, ask the tenant, or look for the missing paperwork, and if two accounts disagree, the history should show both rather than pick one.
A model assembling this kind of history will suggest connections: this report and that one probably describe the same fault. That suggestion is a reason to read the two records side by side, and no more than that. Before treating them as one problem, someone checks that the apartment, the location and the equipment actually match. The same goes for any suggestion about the physical cause. Until a visit or a document settles it, it stays a question.
And sometimes the history is already complete and the record is not the problem at all. The work is known and is waiting on money, access or a contractor. Producing a neater account of it changes nothing. Someone has to remove the obstacle.
The colour of the guttering
The Ombudsman’s report includes a landlord that struggled to get guttering fixed at the first visit because contractors could not carry every type on the van. The landlord began asking residents what colour their guttering was during the initial call and sent that detail out with the job. It reported an increase in first-time completions.
In our riser project, the question had been asked. We raised the plumbing during due diligence and were told the risers were original. The answer was an assumption passed on as fact; the previous owner did not know about the 2009 replacement either. What would have helped was not a better question but a record behind the answer, or a question that separates what is documented from what is assumed. The survey brief could have done the same job by asking the contractor to list any risers left out. We would judge an AI trial by whether it supplied information of that kind. The checked history goes out with the next instruction, and the operator records whether it changed what the contractor was asked to do or bring. A claim about fewer return visits needs a count taken after the work is done, not an estimate made during the pilot.
What the next visit leaves behind
Using the history before a visit also changes what is worth asking for afterwards. An instruction to investigate an intermittent fault should say what findings are expected back, including whether the fault could be reproduced. A recommendation for further work needs an owner and a status, so the next person can see what became of it.
The system we built for our own properties, and use as the basis for what we teach at RENAVID, already prevents a control from closing without the required evidence. The riser survey taught us that the requirement also has to cover what an inspection leaves out. For the rest of the riser project, the report has to identify every section inspected and every section left out, and that line goes into the brief for every survey we order from now on.