Shadow AI, channel leakage and why ‘bad data’ is no longer an excuse
When I wrote an article in the September 2025 edition of Housing Technology about listening at scale, my argument was that the conversations residents have with us, across every inbound channel, are the richest and most under-used source of insight each housing provider holds. One year on, more organisations in our sector are finding ways to use AI tools to help process that data at pace. This article peels back some of the quick wins and considers whether the outcomes are a step forward or whether we’re just generating new problems.
Housing providers are failing in two ways. Shadow AI is on the march, driven by the widespread availability of high-quality AI tools and the shortage of suitable products that enable usage to be governed and data to be managed safely.
Many surveys suggest that nearly half of workers are using unsanctioned AI products, although I suspect the real number is much higher. Housing faces the same challenges: AI is being used with good intent but often in direct breach of data policies. LLMs might help teams respond to a damp complaint within the Awaab’s Law timescales but is the advice compliant with corporate policy, has data been exposed to a system no one controls, and will that decision and advice appear in any record the organisation holds?
Channel leakage
The second point predates LLMs – we call it ‘channel leakage’. Residents have already decided how they want to reach us and it’s not always via the contact centre. It’s a text to a housing officer’s personal mobile, a message to a community Facebook page or perhaps a conversation on the doorstep that ends in “leave it with me”.
Even contact through sanctioned channels fragments and contributes to the sector-wide problem of ‘bad data’. Phone logs, personal inboxes, spreadsheets and case notes that never join up with a single account of the resident. Some of the signal leaks out of the organisation entirely, the rest gets absorbed into a CRM. Either way, the moment a resident informs us that a problem exists, we need to be listening.
Shadow AI and channel leakage are different sides of the same coin: evidence is escaping audit. While this might be a bigger problem than just bad data alone, it is easier to solve. Left unsolved it perpetuates the bad data problem.
Bad data
Fewer than a third of housing providers say they are confident in the quality of their data. The Housing Ombudsman’s spotlight on knowledge and information management found that poor record keeping is a significant cause of routine requests turning into prolonged complaints.
From incomplete risk data on housing stock to missing contact details for active residents, inaccurate data is rightly being identified as a cause of failure. But the data is poor not only because legacy systems are old, it’s also poor because the truth about residents and homes leaks daily into channels and tools we don’t govern. A housing provider can spend a fortune cleansing what sits in their HMS while the freshest evidence it owns never gets there at all.
None of this will soften the regulator’s position. From the end of this November, the second phase of Awaab’s Law extends the statutory regime beyond damp and mould to additional hazards, with investigations inside ten working days, written summaries to the resident inside three days, and emergencies made safe within 24 hours.
Crucially, the clock starts ticking from when the housing provider ought reasonably to have been aware of the hazard, and the government’s own guidance instructs housing providers to keep “accurate and up-to-date information about their residents and their homes”. Read those together and the position is stark: not knowing, because the report sat in an ungoverned channel or an unread inbox, is not mitigation. Under Awaab’s Law, bad data is not a defence. Increasingly, it is the offence itself.
Work with what you have
So, what can we do? The first principle we work with at Fuzzlab is that ‘the data is the data’. It may be incomplete, out of date or plain wrong, but it’s what you hold and not acting on it isn’t acceptable.
This is where insights can really help. Analysing the data you already manage, conversations, contact history and case notes alongside the housing management system, identifies the signals the structured record has missed and can expose the leakage itself, such as the resident who thanks an officer for a visit that no repair was ever logged against or the complaint that references a conversation your systems have no record of.
Analysis will only take you so far, though, which brings us to our second principle: the resident’s voice is the most up-to-date source of truth a housing provider holds. Stock-condition data starts ageing from the day of the survey whereas a resident describing their home today doesn’t. Therefore, treat every conversation as a fresh reading on the property and the household, and reconcile the existing record against it.
Shadow AI
It follows that the solution to shadow AI is provision, not prohibition. Housing staff reach for ungoverned AI tools because (mentioning no names) the governed ones aren’t good enough, and residents message housing officers’ personal mobiles because the official channels are too slow.
Better communication tools close the leak at the point of contact, capture at source, classify, timestamp, write back to the system of record and make the sanctioned route the easiest route. The inboxes, call recordings, webchat transcripts and WhatsApp threads you already manage then become compliant by design. This is also the same standard we hold our own work to; the omni-channel services we run in production exist precisely to stop the leak at the moment a resident makes contact.
The primary outcome of Awaab’s Law will (hopefully) be safer homes. Its secondary outcome might be to force housing providers to finally work with the data they already manage (rather than waiting for the perfect dataset that never arrives) and face up to the elephant in the room. When the leaks close and the record fills, the numbers will get worse before they get better. The five-fold rise in property-condition complaints since 2019 wasn’t new demand; it was demand finally being recorded. And for some organisations, a complete record will expose an even harder truth: their compliance liabilities will be on a scale that puts their own viability in question.
Uncomfortable as that might be, it’s a necessary step. Only when our sector can see its problem whole can it appreciate the true scale of what it faces and make the case for support and investment. You can’t fix what you can’t evidence and from this November, you can’t defend it either.
Paul Singleton is the founder of Fuzzlab.
