How to teach property managers to bank corrections that improve an AI agent

Property managers bank a correction by recording why they changed the agent's repair triage, tenant reply or arrears letter, then deciding whether the fix belongs to one lease, one building or the whole portfolio. Portfolio fixes become brief or bound changes; lease and building facts go into sources. Safety triage errors go first, the same day.

What does the agent do in property management?

A property management agent triages maintenance requests and raises routine work orders to approved contractors within a spend limit (depth 3), drafts replies to tenant queries about leases, service charges and access (depth 2), and drafts arrears reminders in line with the landlord's policy. Anything legal, such as notices, and anything above the spend limit stays with a person.

Managers correct it in recognisable ways. It classes a report of a smell of gas as a routine repair. It sends a plumber who is not on the approved list for that building. It tells a tenant the landlord is responsible for internal decoration when this lease puts it on the tenant. It sends a standard arrears reminder to a tenant who has a payment plan agreed with the manager.

The cross-role skills map is in AI agent skills by role.

Lease, building or portfolio?

This is the distinction specific to property. Leases vary: two flats in the same block can have different repair obligations, break clauses and service charge shares. Buildings vary: approved contractors, access arrangements, fire procedures. And some rules hold across the portfolio: how to triage urgency, what counts as an emergency, how arrears reminders escalate.

When the agent gets a lease term wrong, the fix is usually in sources: the agent should read that lease, not a standard template. When it sends the wrong contractor, the building's contractor list is the source to fix. Only portfolio-wide method problems, such as how it classifies urgency, belong in the brief. A manager who banks a lease-specific fact into the brief will break the agent for every other lease.

What are the steps?

1. Agent, depth 3: triages and raises routine work orders within the spend limit, to approved contractors.
2. Agent, depth 2: drafts tenant replies and arrears reminders, citing the lease and policy used.
3. Person (property manager): corrects and records a reason: urgency misclassified, wrong contractor, lease term misread, arrangement missed, method unclear, or one-off.
4. Person (property manager): reports safety misclassifications at once to the head of property and the agent owner.
5. Agent, depth 1: groups other reasons weekly by building and reason.
6. Person (head of property): decides what to bank and where: lease source, building source, or brief.
7. Person (agent owner): makes the change, and Attest replays the agent against past requests and tenant messages, with the escalation and adversarial gates weighted for safety cases.

What must property managers learn?

How to treat safety triage as a red line. Gas, electrical, fire, flooding and structural reports must never be handled as routine. The banked fix is a rule that certain words and situations are always held for a person immediately, with the tenant told what to do in the meantime. Managers should know which words their portfolio's policy treats as emergencies.

How to recognise agreements the agent cannot see. Payment plans, access arrangements and informal permissions are often agreed by phone and recorded loosely. If the agent keeps missing them, the fix is to record them where the agent reads, as a declared or signed fact on the tenancy, rather than to correct reminders forever. The review habits are covered in how property managers review an AI agent's decisions.

How to read a lease well enough to diagnose. A manager does not need to be a lawyer, but they need to tell when the agent read a template instead of the actual lease.

Which controls and evidence matter?

• Emergency words always held. Evidence: held verdicts for safety reports, sampled monthly.
• Contractor lists by building from the system of record. Evidence: provenance on each work order.
• Lease-specific sources. Evidence: the lease cited in each tenant reply.
• Re-certification. Evidence: a dated certificate with escalation and adversarial gate scores.

What should the team measure?

Safety misclassifications, which should be zero after banking. Work orders sent to unapproved contractors. Tenant replies corrected for lease terms, by building. Arrears reminders sent to tenants on agreed plans. And time from a contractor list change to the agent's sources.

Where does it usually go wrong?

Lease facts get banked as rules. A manager corrects a decoration question three times for the same block and asks for the brief to say tenants are responsible for decoration. Now the agent is wrong for every other building.

Informal arrangements stay invisible. The agent cannot respect what it cannot read, and managers keep correcting reminders. Record arrangements properly.

Spend limits get raised to clear the repair queue. Operations managers own that trade-off across teams, as teaching operations managers to bank corrections explains. Portfolios in different markets also bring different tenant protection rules, as AI agent training for property managers in the Netherlands describes.

Where the Academy fits

Pryme Intelligence Academy's Operator track covers human review, teaching the agent, keeping knowledge current and when something goes wrong, which is the full cycle of banking a correction. Module tests are set in the learner's own industry and role, so property managers practise on tenant and repair cases, and labs run in a free Build workspace.