Markus Breth

Product · AI agent under approval

The agent proposes. A human approves.

Rates, bookings, reviews, cleaning, deadlines — the agent watches the whole operation, proposes changes and lays out what each proposal rests on. Nothing executes before approval. In production since July 2026 at a hotel and rental group in Austria — on their own server, with a local language model.

  • In production since July 2026
  • 243 tests green
  • Local LLM
  • GDPR

The problem

Running a property is an endless run of small decisions.

None of them is hard. All of them together are too many — and the ones left undone cost the most.

  • 01Rates go stale by the hour. Miss a short-notice adjustment and you sell too cheap, or not at all. By hand, nobody keeps up across every unit and every day.
  • 02Deadlines pass in silence. Damage claims, complaints, tax dates: no reminder arrives on its own — until the window has closed.
  • 03AI you cannot watch. Tools show *that* something was changed automatically. Why, based on what and whether it worked stays open — and therefore unverifiable.

What the agent watches

One head for the whole operation.

Eight kinds of action, each governed on its own: execute directly, submit for approval, or never propose at all.

◆ RATES

Rates

Spots gaps in occupancy and proposes an adjustment — with the before/after and the data behind it.

◆ GUEST

Guest messages

Drafts for replies and offers that only go out once approved.

◆ REVIEWS

Reviews

Drafts responses to reviews in the property’s voice — submitted, not published.

◆ CLEANING

Cleaning

Orders cleaning to match arrivals and departures instead of tracking it in a second list.

◆ INVOICE

Invoices

Handles 10 % Austrian VAT and the Vienna city levy — and does not charge twice when the channel already collected it.

◆ DEADLINE

Deadlines

Speaks up before a deadline runs out — not after.

How the agent works

Observe · propose · approve · measure.

Every one of these steps leaves a row in the ledger. The dashboard is nothing but the view onto it.

01

Observe

Bookings, rates, reviews and dates are pulled in and evaluated continuously.

02

Propose

Every proposal comes with its reasoning and the evidence it stands on — not as a bare instruction.

03

Approve

A human sees before and after, and decides. Without approval the change stays in-house and never reaches a booking channel.

04

Measure

Days later the system checks for itself whether the change worked — and writes the result back.

Where it stands

In operation, not in a brochure.

243
tests green
8
kinds of action, each governed on its own
0
external effects without human approval
100%
on their own server, language model local

AI that runs the operation — and lets you watch it do so.

Runs on your own infrastructure and adapts to your property, channels and processes. Live demo on request.