Running a rental operation on one system instead of five

Enquiries arrive in a shared mailbox that also receives newsletters, invoices and sales spam. Someone reads each message, decides whether it is a real enquiry, and remembers to answer. First-response time drifts and nobody can see how many are waiting.

Availability and promises live in separate heads. A unit can be blocked for maintenance, reserved for someone who has been offered it, or occupied by a tenant whose notice has already been given. Tracked in a spreadsheet, that produces double-bookings and awkward phone calls.

Waitlists go stale, because people join a queue and lose interest months later. Every lease is the same document with different names, produced by hand and chased for a signature. And a rent change or a termination with a miscalculated date is a dispute waiting to happen.

Availability is modelled as two separate axes. A unit has an occupancy state derived from actual rentals and offers, and an operational condition the operator sets by hand, such as blocked or under maintenance. The system refuses a conflicting assignment rather than recording one, and an offer reserves its units the moment it is sent so a colleague cannot promise them twice.

The lease is rendered from an editable bilingual template, emailed as a link, and filed as a signed PDF with the signer's name, timestamp and IP address. Rent changes are a versioned history with effective dates and a notice letter, not an overwrite. Notice periods produce a calculated end date rather than an estimated one.

An offer can be made for a unit that is still occupied, provided the start date falls after the sitting tenant's end date. That is how a handover with no empty period gets arranged, and it is the kind of rule that only exists in a system built for this business.

Encoding those rules into the system changed who can do the work. The eligibility rules decide what may be offered, so a colleague does not need to know the exceptions. The end date is calculated from the notice period rather than worked out. The letters come from templates that already contain the correct Danish wording. What was promised to whom is a reserved unit with an expiry date, and what was said is on the customer's timeline. A new administrator works from the system instead of shadowing the incumbent, and a holiday stops being a risk.

That persistence is also the precondition for the AI being useful at all, and the order matters. The assistant can summarise a thread, extract an enquirer's details, draft a reply in the operator's voice and propose a next action because the correspondence, the units, the prices and the rules are all recorded. Where knowledge stays tacit there is nothing for a model to work from, and no way to check what it produced. The system was not made intelligent by adding AI; the AI became possible because the operation had been written down.

It also means the assistance improves as the process does. A correction an administrator makes to an AI-sourced value is captured as a signal and aggregated into a reviewed, human-approved improvement, rather than a rule quietly rewriting itself.

It handles no money. Every rental's price, deposit and intended payment method are recorded, but no invoice is produced, no payment is tracked, and no deposit is settled. An operator adopting it keeps invoicing in their accounting system and reconciles by hand. The design for how billing would attach exists; the feature does not.

Signing is click-to-accept electronic acceptance, not a national eID such as MitID. That is adequate for these rental agreements, and the code is structured so a signature provider could be substituted, but none is integrated today.

Filling a vacancy from the waitlist is still a human decision. The system ranks the queue, asks whether people are still interested and lapses the ones who do not answer, but it does not work down the queue by itself when a unit frees up.

Renters have no portal. Every customer-facing step is a single-use emailed link, which removes account and password management at the cost of not offering an ongoing self-service view. And the platform runs for one operator today: the data model carries an operator identifier throughout and isolation tests exist, but selling it to multiple operators would make that isolation security-critical and is work not yet done.

The coordination gain is specific. Units are reserved the moment an offer goes out, a blocked or taken unit cannot be assigned, and a notice date becomes visible early enough to line up the next tenant. Those are exactly the points where a spreadsheet-run operation produces double-bookings.