The competence tax has been abolished
Thought leadership by Olivier Jauniaux Founder of NestLink
Somewhere in a lettings office this week, a property manager is reading a six-page document about a deposit deduction of £85. It has a chronology, numbered allegations, and citations to the tenancy agreement by clause and the deposit scheme by guidance note. It is better structured than anything the agency has ever sent out, and it was produced in about four minutes by a twenty-six-year-old on a train, from a phone, for free.
Propertymark has been warning about this since December, reporting that complaints increasingly arrive more polished, more assertive, and more technical, and sometimes built on incorrect assumptions produced by an AI tool. That last clause is the one worth pinning to the wall. Both halves are true at once, which is what makes this awkward to manage.
AI did not give tenants new rights. It gave them the vocabulary, the confidence, and the free time to use the ones they already had. The agent’s problem is not the letter. It is whether the file can answer it.
The number everyone is about to misuse
The figure now circulating is that complaints against letting agents rose 47.4% in 2025, to 4,220 from 2,863, with early resolution improving from 50% to 53%. It is being attributed to Propoly, the letting agency CRM software, and it is being read as a national count of tenant complaints. It is neither.
Those figures are Property Redress’s own published results from its 2025 annual report: 4,220 complaint enquiries received by one redress scheme, a 47% rise on the year and 77% since 2023, with accepted cases up 41% and average handling time down from 39 days to 34. Propoly’s contribution is commentary on top of scheme data, not an independent measurement of anything.
This matters twice over. A scheme’s enquiry volume is not the country’s complaint volume, and anyone building a strategy on 4,220 as a market figure is building on ground as solid as the sinking sands Crosby Beach. Antony Gormley may have something to say on that analogy, but you get the point.
More interestingly: if complaints to that scheme rose 47% while time-to-close fell by five days AND early resolution improved, the story is not that consumers have become more litigious. It is that a larger number of better-organised complaints were disposed of faster. That is not the shape of a crisis.
It is the shape of a market where the input got clearer. And note what the data does not say. It does not identify AI as a cause. Nobody has published a proportion attributable to it because nobody can reliably detect it. The correlation may be real. The causation is currently nothing more than Gen-Z “vibes.”
The competence tax has been abolished
Making a formal complaint used to require four things a tenant frequently did not have: time to research the rules, the literacy to write in the register the recipient respects, knowledge of where redress lives, and the confidence to press SEND. In the past, you could have called it a competence tax. Like most taxes, it was paid disproportionately by the people least equipped to pay it, which is a polite way of saying it filtered complaints by education demographic rather than by merit.
The supporting evidence is not from lettings and should not be dressed up as if it were. A 2026 study in the Nature: Human Behaviour publication analysed more than 1.1 million US financial complaints filed between 2015 and 2024 and estimated that LLM assistance raised the probability of favourable relief by 6.9 percentage points, with controlled experiments suggesting the gain came from presentation rather than from any change to the facts. Different jurisdiction and different sector.
So useless as a UK lettings statistic. However, it is extremely useful as a description of the mechanisms involved, despite the rather more “extroverted” demographic across the pond: AI is not acting as a lawyer here. It is acting as a communications advantage. To agents, this is uncomfortable. The old system rewarded presentation too. It just rewarded the agencies, and the agency had the professional writers. Consumers had… Reddit?
Confident, formal, and wrong
Our property manager, from the beginning of this narrative, is now on page six of the tenants’ deposit argument. It is unbelievably confident and formal. Slightly scary, perhaps, as obscure legal terms and cases are expelled at the rate of a hyperactive leopard in a blind chase with an erratic mouse.
None of which makes the letter necessarily true or correct. A model will produce a beautifully cadenced paragraph asserting a statutory duty that does not exist, cite a clause absent from the agreement it was shown, and escalate a disagreement about a fridge into an alleged breach of the Renters’ Rights Act with the serene confidence of something that has never and can never been sued.
The ICO has already had to warn, in the context of information requests, that AI can misstate legislation, inflate wording and manufacture complexity, and that output should be checked for accuracy and tone before it is sent. Different context, identical failure mode, with the agent now receiving a document that is perhaps a third legitimate, a third arguable and a third invented, delivered in a single tone of unbroken authority.
Sorting those piles is the actual work, and it is extremely hard to delegate to someone five months into the job who has just been told they are in breach of a statute they have never heard of. There are, of course, legal AI platforms such as Legalito, who are training models on genuine legal issues and frameworks, but these are designed to accelerate legal workflows, with actual lawyers. They are different from the free or even the paid open-source tools.
The complaint is a lagging indicator
The obvious next move is already happening agents use AI to summarise the complaint, AI drafts the response, the tenant feeds it back in and asks for a rebuttal. Within three rounds, two language models conduct a correspondence of escalating length and diminishing information, and all about a boiler that is still broken.
The increased volume of complaints is a distraction, and the industry is about to spend a year solving the wrong problem. When both sides can generate a persuasive letter in seconds, persuasiveness is worth nothing! Everything reverts to the evidence, which is the part the sector has systematically underinvested in, because retrievable records have never won a listing.
Inventories, contractor correspondence, dated inspection notes, who agreed what and when: these live in inboxes, on personal phones, in WhatsApp threads, and in the memory of whoever has since left for a competitor. This is the same issue NestLink is working on in property transactions where the most important investment in someone’s life is held together by broken email chains, good will and mostly faith.
All this chaos was survivable when the counterparty could not construct a chronology of issues. The counterparty can now construct one in four minutes. And by the time the six-page letter lands, the useful moment has passed anyway.
The tenant rang at 9pm on a Sunday, got voicemail, rang again on Monday, and was told someone would call back, which only happened on Thursday due to busy schedules and a Post-it note getting blown off a desk. Somewhere in those seven days a repairable problem became a grievance, and the only record of it is a missed call and a vague memory. The complaint is a lagging indicator of a conversation nobody captured.
This is why the more interesting AI in property lettings and management will not be the kind that answers or sorts complaints. It is the kind sitting on the inbound side, before the grievance is actually blown out of proportion. RezX is an example of such a product.
RezX is a new conversational AI that answers tenant voice calls, texts, WhatsApp messages, and voice notes around the clock, triages maintenance issues, troubleshoots from a photograph via image recognition, escalates what it cannot resolve and turns the rest into structured work requests.
Its published case study with Jeffries and Dibbens claims 64% of troubleshooting and work requests became fully automated – with a human in the loop when it could not be. Other related products include VerbaFlo, Latch and BrickWise AI.
The genuinely important feature, however, is not the automation. It is that the interaction is centralised, GDPR-controlled, and timestamped rather than sitting two messages below a selfie of the negotiator’s mum on a Caribbean cruise.
Every out-of-hours call becomes a dated record. Every reported fault becomes a work request with a photograph attached. AI stops being the thing tenants use against agents and becomes the intermediary that both absorbs the 9pm frustration and, almost incidentally, builds the evidence file the agent will need if it escalates anyway.
Solid ground, in other words, laid before anybody starts arguing about what happened. The principle generalises beyond any one vendor: a system that captures the conversation as it occurs beats any system that reconstructs it afterwards, and reconstruction is exactly what an agency is doing at 4pm on a Friday when an adjudicator asks for a timeline.
Your To-Do List this Monday
Do not build an AI-detection capability. It will not work, and it is irrelevant: even a machine-written complaint arrived because something happened. Train your team to separate tone from substance instead. The most exploitable weakness in a lettings business right now is a nervous employee who assumes formal language implies a valid legal claim and concedes accordingly.
Triage the complaints into their component allegations, most of which dissolve on contact with the actual property file and your records, and answer each with a document reference, at least internally.
NestLink’s argument has always been that the information problem and the coordination problem are one problem, and that a controlled environment where the timeline, the documents and the decisions are visible to everyone is worth more than the messages sent inside it.
Put in lettings terms: the value of a system of record is not what it lets you say. It is what it lets you prove. Propertymark arrives at the same place from an entirely different starting point, which is usually a sign the conclusion is right. Keep the final decision with a human, because Propertymark is right that AI cannot assess legal risk, and because an adjudicator will definitely NOT be charmed by a response that reads as though nobody actually looked.
And two sentences for the other side since nobody else will say them. Tenants uploading tenancy agreements, photographs of their homes and the names of other occupants to a general-purpose model are handing over considerably more personal data than they intend. That is a conversation the sector should be honest about rather than gleeful over.
AI has not changed tenancy law, and it has not changed what happened in the property. It has removed the barrier that used to stop a large number of people saying so clearly. The complaint of the near future will be machine-written, immaculately structured and occasionally nonsense. The winning response will depend on something stubbornly analogue: and whether it can still find the proof.
