Skip to content
Ittai Savran

Putting AI to work

An operation that mostly runs itself.

Every square is a unit under management. Lit means it moved without anyone touching it; terracotta is where a person still decides, and stays deciding.

Get a process mapped

Floors 01–02 · 11:00

Running without a personWaiting on a human call

Where it usually lands

Eleven parts of a hospitality or property business. Most operators draw the line far too early in each of them, leaving people on work that stopped needing a person a while ago. What genuinely stays is the policy, and the cases that fall outside it.

01 / 11

  • 01

    Guest communication

    What runs on its own

    Not only directions, door codes and what time breakfast stops. Compensation offered inside limits you set, troubleshooting that used to need a supervisor, upsells put in front of a guest at the moment they are worth something, and an alert while a stay is going wrong rather than a review afterwards.

    Where you still decide

    The limits themselves. How much can be given away without asking, and what happens to the cases that fall outside that.

  • 02

    Operations

    What runs on its own

    Rotas, turnaround order, who goes where when a departure runs late, and a flag the moment the plan stops matching the building. The daily choreography that currently lives in one supervisor's head and a WhatsApp group.

    Where you still decide

    The call on the floor. And what the operation does when the same flag fires three days running.

  • 03

    Maintenance and the building

    What runs on its own

    Faults logged from a guest message or a meter reading, triaged and dispatched with the access codes and the warranty already attached. The same leak flagged as the same leak, not four separate tickets across four months.

    Where you still decide

    Repair or replace, and when a fault that keeps coming back becomes a capital decision rather than another call-out.

  • 04

    Revenue and pricing

    What runs on its own

    Pricing as a weekly decision rather than an annual policy: comp set, draft rates, pushed to every channel, then re-read to confirm what actually went live. Upsells and length-of-stay offers priced per booking rather than per season.

    Where you still decide

    The strategy the exceptions get measured against, and which exceptions are worth making.

  • 05

    Marketing and distribution

    What runs on its own

    Listing copy, channel content, and the same update made across hundreds of listings without a person making it hundreds of times.

    Where you still decide

    What the brand says, and who it is being said to.

  • 06

    Reviews and reputation

    What runs on its own

    Every review answered in your voice within the hour, and the themes read across hundreds of them: which building, which floor, which week the word "noise" started appearing.

    Where you still decide

    What the operation changes because of what the reviews keep saying. The apology is automatic; the fix is a decision.

  • 07

    Procurement and suppliers

    What runs on its own

    Orders sized from the occupancy ahead rather than the month behind, renewal dates watched, and prices compared across suppliers before a contract quietly rolls over.

    Where you still decide

    Who you buy from. A price is a number; a supplier who answers on a Saturday night is a judgement.

  • 08

    Hiring and onboarding

    What runs on its own

    Applicants screened and scheduled, the paperwork of a new hire, and the training that today happens by shadowing whoever is on shift. Written down once and kept current.

    Where you still decide

    The hire itself, and who gets trusted with more.

  • 09

    Compliance and regulation

    What runs on its own

    Guest registration, licence renewal dates, the paperwork a tourist-rental licence demands, and rules that change by city and sometimes by street. Watching what changed, and knowing which of your buildings it touches.

    Where you still decide

    What to do when a rule changes and the answer costs money either way.

  • 10

    Back office and reporting

    What runs on its own

    Invoices issued, reconciliation, suppliers chased, and the weekly numbers somebody currently rebuilds by hand every Monday. Plus the questions nobody asks today, because assembling the answer would take a day.

    Where you still decide

    What falls outside policy, and what the operation does about what the numbers say.

  • 11

    Owners and investors

    What runs on its own

    The monthly statement assembled, checked against the ledger and sent the same day every month, with the questions about it answered from the same numbers.

    Where you still decide

    The call when a month is bad, made by you rather than announced by a report.

The processes, in detail

The same argument, close up. Pick a process and see which steps moved and which stayed with a person.

Guest asks to arrive early

From the inbox to a confirmed time

  1. 01Request arrivesGuest app · Automated
  2. 02Read previous departurePMS · System
  3. 03Check the cleaning rotaOps agent · Automated
  4. 04Room ready by 11:00?Ops agent · Automated
  5. 05Offer 11:30 insteadOps agent · Automated
  6. 06Housekeeping confirmsSupervisor · Human call
  7. 07Times written backPMS · System
  8. 08New door code sentGuest app · Automated
18 min → 40 sec
Time per event
4 → 1
People it passed through
Still a human call
Housekeeping

What it means

An early arrival used to travel from the inbox to reception to housekeeping and back. Now the agent reads the departure, checks the rota and proposes a time the building can actually hold, then stops for the one judgement that needs a person on the floor.

The catch

The supervisor keeps the veto. Remove that and you win an hour and lose the trust of the team that has to deliver it.

Estimates, from work I have done rather than from one audited engagement. The shape is what I would expect to find; the exact hours depend on the building and the team. The repricing figures are the exception and come from something already published.

Before the building

Putting AI to work in the deal

Most of a deal is reading, checking and rewriting. Very little of it needs a person.

Screening and underwriting

From a listing or a teaser to a first model and a go or no-go in hours rather than weeks.

Due diligence

Contracts, licences, planning and technical reports read and flagged by machine, decided by a person.

Investor reporting

The reporting and the data room kept current automatically instead of rebuilt each quarter.

The pipeline as a system

Screenings, visits and passes filed with their reasons, so the GP runs on data rather than on whoever remembers.

I use these systems on my own deals, and where it suits both sides I invest alongside.

Send me a process you run and I'll map it like this on the first call, before anyone buys anything.

Get a process mapped