What AI can and cannot do with a leaseholder inbox

It is a fair question, and the answer is genuinely useful. It only works, though, if the inbox is divided in two first. Some of what arrives is administration. Some of it is a statutory step with a consequence attached. Managing both as a single pile is where agents come unstuck.

Where the volume actually sits

Map a working week properly, rather than estimating it from memory, and the same pattern turns up in businesses that have nothing else in common. The work that consumes the most time is almost never the work that feels most important. It is small, repeated correspondence that nobody logs, because each instance takes four minutes and no single instance is worth recording.

In block management that means the routine acknowledgement, the update on a job already under way, the explanation of what one line on a budget covers, and the request for a copy of a document that was sent out in March. None of it calls for judgement. All of it calls for a reply, and somebody has to write that reply.

This is also a sector with room to move. Government research published by the Department for Science, Innovation and Technology puts average adoption of artificial intelligence (AI) at around 16 per cent across firms with five or more staff, which leaves most of the market yet to try any of it in earnest.

What it genuinely shortens

Three jobs stand out.

The first is the acknowledgement and the status update. A leaseholder emails about a leaking roof reported a fortnight ago. Somebody has to look up where the job sits, write a courteous reply, and log that the reply was sent. The looking up is quick. The writing is not, especially at volume, and it is exactly the sort of drafting a machine handles well.

The second is triage. Correspondence arrives through the portal, by email, on the phone and, increasingly, through the resident group chat that no agent asked to be added to. Sorting that into categories, flagging the urgent, and routing each item to whoever owns it is repetitive pattern-matching. Handing it over frees the morning without touching a single decision.

The third is preparation rather than production. Before a budget goes out, somebody usually reads back through a year of correspondence to work out which items will draw questions. That is a genuinely useful piece of analysis and it almost never gets done, because there is never an afternoon spare. It is also the sort of task where a machine reading 400 emails in a minute earns its place outright.

Where it must not go

The rest of the inbox is a different animal, and the reason is legal rather than technological.

Consultation under section 20 of the Landlord and Tenant Act 1985 is triggered when qualifying works would leave any single leaseholder contributing more than £250, or when a qualifying long term agreement would cost any leaseholder more than £100 in an accounting period. Get the notice wrong and the recoverable contribution is capped at those figures, however large the actual bill.

Section 20B is blunter still. Costs incurred more than 18 months before a demand is served cannot be recovered at all, unless the leaseholder was notified in writing inside that 18 months that the costs had been incurred and would be charged.

Section 21B requires that every service charge demand is accompanied by a summary of the leaseholder's rights and obligations. Omit it and the leaseholder may lawfully withhold payment, and the lease's own provisions on late payment stop having effect for as long as the payment is withheld.

None of those are drafting problems. They are deadline and evidence problems, and the correspondence that carries them has to be right in a way a first draft never is. An agent who lets a system generate a section 20 notice, or a demand, or a written notification under section 20B, has handed a statutory step to something that cannot be held to account for it. The consequence lands on the agent and on the resident management company (RMC), not on the software.

There is a subtler risk too. A well-written machine draft reads as though somebody has already checked it. That is precisely the sort of document that gets approved quickly, which is the opposite of what these ones need.

A test that survives a Monday morning

One question sorts the inbox faster than any policy document. Before a machine goes anywhere near a message, ask what it would cost if it were wrong.

If the answer is an apology and a corrected email, it is administration, and it can be drafted automatically and sent by a person who has read it. If the answer involves a tribunal, an unrecoverable cost or a right the leaseholder can exercise against the RMC, it stays with somebody who understands the lease.

That instinct is not unusual. The same government research found roughly two-thirds of firms already using AI keep a person checking the output before it goes anywhere. Block management has stronger reasons than most to be in that group.

What it takes to get there

The groundwork is duller than the technology and matters more. Correspondence needs to land in one place rather than four. The property management system needs to hold the job status the reply will quote. Templates need to be current, because a machine will reproduce an out-of-date one faithfully and at speed.

Then it is worth starting with a single job, rather than a platform promising to fix the whole inbox at once. Status updates alone are enough to prove whether any of it holds. If it does, the hours come back where they were always going to come back: from the small, repeated work that never appeared on anyone's report.

David Bevan is the founder of HoursBack, which runs AI workflow assessments for UK small businesses.