hostify-invest

1 Executive summary

Hostify holds years of nightly rates, occupancy, costs and reviews across thousands of short-term rental listings, and today it earns from that data only through the channel manager subscription. No bank, broker or appraiser has operating data of this depth for short-term rentals. This proposal builds an investment advisory module on top of it. The module answers the questions owners, managers and investors take to a consultant, when they can afford one: is this property worth what I paid, should I sell, convert or renovate, is a loan worth taking, where should I buy next. Each answer is a written memo with the numbers behind it and forecasts that state their own uncertainty. The module also sends owners suggestions they would not have thought to ask for, such as a second bathroom that pays back in under four years.

1.1 What Hostify gains

For existing customers, a premium tier and a reason to stay: an owner whose property plan lives in Hostify has little reason to move to a cheaper channel manager. For investors, lenders and insurers, a paid data product: the market index that Hostify's users build every night as a side effect of running their listings. Later, a real estate transactions business, where properties are sold with a verified operating history and Hostify earns on the sale, the valuation and the financing introduction.

1.2 Why now

Owners have been through three volatile years and want answers. Cities are changing their short-term rental rules one by one. Investors are entering the asset class with nothing to underwrite with. The forecasting and language models the module needs are mature and cheap to run. Every booking adds to the dataset, so whoever builds on it first keeps a lead that is hard to close.

1.3 What we ask of Hostify

An anonymized sample under NDA, read access for the demo, one product owner, a legal review of the memo wording, and a decision on the demo within four weeks of this proposal.

2 A day in the life of an owner

It is a Tuesday in March. Maria owns a two-bedroom apartment in the Oltrarno district of Florence, managed through Hostify. At 8:40 a message arrives in her Hostify inbox: the six two-bathroom apartments closest to hers earn 4,100 EUR a year more than her one-bathroom apartment, a second bathroom in her building type costs 14 to 18 thousand and pays back in just under four years. Would she like the full analysis?

She taps. The memo opens with one sentence: build it before the summer season, a gross return of about 26 percent a year on the money, the result holds unless the build costs more than 22 thousand. Below it, her apartment against the street: revenue, occupancy, review score, the four risk scores, and a note that her cleaning cost is 23 percent above the Florence median with three vendors to call. She asks the chat: "what if I do it in autumn instead?" The engine reruns; the answer is a season of lost uplift, about 2,900 EUR, and a revised payback. She asks whether she should take a small loan for it; the memo shows the loan at her bank's rate against paying cash, and recommends cash because the sum is small and the rate is above the apartment's yield.

She forwards the memo to her husband and to a contractor. The checklist at the end tells her what to confirm: a wet-wall adjacency check, the building's condominium approval, the municipal permit, an insurance update. In April she accepts a quote, blocks three weeks in the calendar, and the platform records the renovation. In September the apartment earns its first month at the new rate; Maria's review score crosses 4.7 in November. In December the quarterly memo shows the uplift came in at 19 percent against a forecast of 18, and her manager uses the case, anonymized, to persuade two other owners in Oltrarno.

The following March, another message: the city council has scheduled a vote on tightening the short-term rental rules in the historic centre; if it passes, her income falls by a third, and two options are already modelled, a mid-term let for visiting academics and a sale before the vote at a price the marketplace has three qualified buyers for. Maria has never built a spreadsheet, never read a market report, and never called a consultant. She has made four good decisions in a year, and the platform that helped her make them is the one she will not leave.

3 Thesis: the data is worth more than the platform

Hostify today sells software that runs short-term rentals. The asset it has accumulated while doing so is larger than the software: years of nightly rates, occupancy, channel mix, cleaning and maintenance spend, guest reviews, cancellation behavior and seasonality across thousands of listings in many markets. No bank, broker or appraiser holds operating data of that depth for short-term rental property. Owners ask the same questions of it: is this property worth what I paid, should I sell it now, would it earn more as a long-term let, and what will it cost me over the next five years. Few of them get a well-founded answer today.

The proposal is an investment planning and forecasting module that turns that data into the kind of advice a property consultant gives: a full financial model per property and per portfolio (IRR, NPV, ROI and the rest), forecasts of every revenue and cost line with honest uncertainty, detection of trends and anomalies the owner cannot see from monthly statements, scenario analysis for hold, sell, convert, renovate or refinance, and agents that watch market, regulatory, political and climate risk. What the owner receives is a decision memo with a recommendation, the numbers behind it, and the risks that would change it.

For Hostify, the point is where the owner acts on the advice. An owner who plans a sale or a refinance inside the platform is likely to carry it out there too. That opens a second business next to the subscription: a marketplace for operating-data-backed property transactions, where every listing comes with a verified track record and a forward forecast, and Hostify earns on the transaction, the valuation and the financing introduction. The module leads from a management tool to a real estate transactions business, and a competitor would need the same years of operating data to copy it.

3.1 The synergy at a glance

Each layer needs the other two. Analytics without the platform's data runs on guessed inputs, and an LLM agent without the analytics has nothing reliable to explain. Together, the data supplies ground truth and peers, the analytics turns it into forecasts and returns with honest uncertainty, and the agents watch the outside world, write the memo in plain language and bring the opportunity to the owner before they ask. When an owner acts on the advice, the result is recorded as a new data point: a renovation with its cost and its effect on revenue, a sale with its price. The next forecast for a similar property starts from more cases.

Terminology: property (a Hostify listing with an owner), portfolio (an owner's or manager's set of properties), model (the per-property financial model), scenario (a strategic option evaluated against the base case), memo (the advisory output), signal (a detected trend, anomaly or external risk).

4 Who decides, and which decisions the module answers

4.1 The decisions, as the module frames them

Every decision is an option set evaluated against a base case over a horizon the user picks (default 5 and 10 years), with the same metric set and the same risk treatment, so options are comparable:

1. Hold as short-term rental, current operation.
2. Hold and improve: pricing strategy change, channel mix change, renovation or amenity investment, management change.
3. Convert: to long-term let, mid-term (corporate) let, or owner use; and the reverse for a long-term property the owner considers moving to short-term.
4. Sell now, or at a chosen later date, with the sale price estimated from operating data and market comparables.
5. Refinance or add leverage, where debt terms are known or assumed.
6. Buy a new property (for investors and expanding owners), with the same model applied to a listing's history or to a market segment.

The module does not take the decision. It states which option has the best risk-adjusted return under the user's constraints, how confident it is, what assumption would reverse the ranking, and what the user should verify before acting. That is the standard a top consultant works to, and it is the standard the memo is held to.

5 Use cases by customer segment

The same engines serve four segments at four price points. The small owner gets a clear answer to a 20,000 EUR question; the fund gets a diligence file it would otherwise pay an advisory firm six figures for. The numbers below are illustrative; the appendix shows worked examples.

5.1 Segment A: owner of one to three properties (self-managed or with a manager)

5.2 Segment B: property manager with 30 to 500 listings

5.3 Segment C: small and mid-size investor (1 to 20 properties, buying)

5.4 Segment D: funds, lenders and institutional buyers

The deep tier, priced as advisory. The same engines, more depth, and the index itself.

5.5 Debt and financing questions (every segment)

Most real decisions involve a bank. The module treats the loan as a first-class input: principal, rate (fixed, floating or indexed), term, amortization, fees, prepayment penalty and the owner's tax treatment of interest. Every option in the scenario analysis can be run with and without leverage, and the memo always shows both.

The point for Hostify: segments A and B are its existing customers and the module is a retention and upsell feature. Segments C and D are new customers who pay for access to what A and B generate every night. The platform's data becomes the product its users were already creating for free.

6 Data foundation

The module is only as good as the ledger beneath it. Hostify already holds most of the operating side; the capital side (purchase price, debt, taxes) and the external side (market, regulation, climate) must be added. The first engineering task is a property ledger: one clean, monthly, per-property time series of every revenue and cost line, with provenance, built from the data below and kept current by the platform.

6.1 What Hostify already has

6.2 What must be added

6.3 Data quality rules

• Every ledger line carries provenance: source system, record id, extraction or input date, and a confidence (recorded, derived, estimated, assumed).
• Missing cost lines are never silently zero. The ledger fills them with a market default, marks them assumed, and the memo lists them under "verify before acting".
• A property enters the model only with at least 12 months of operating history or an explicit "projection from comparables" flag that the memo states on its first page.
• Owner inputs are validated against platform ranges (a cleaning cost ten times the market median triggers a question, not an error).
• Currency: the ledger is in the property's operating currency; the model reports in the owner's currency with the FX path stated.

6.4 The cross-platform index

The asset nobody else has. For every market cell (city, district, property type, bedroom count) with at least 30 active listings from at least 5 accounts, the platform computes monthly: ADR, occupancy, RevPAR, supply count, new-listing rate, exit rate, average review score, cleaning cost per turnover, maintenance cost per night. These indices drive the forecasts, the comparables and the market risk signals, and they are themselves a product (see the business case). Privacy: k-anonymity on the cell, no per-account identification, owner opt-out honored.

7 Financial engine

A full, auditable cash-flow model per property and per portfolio, month by month, over any horizon from one to thirty years. It is the same model a bank or an appraiser would build, built once and kept current by the platform. The AI never computes a number; it reads the model's output and explains it, and every figure in a memo links back to the calculation that produced it.

7.1 Cash-flow structure

The model follows the standard real estate pro forma, so a banker or appraiser recognizes it:

1. Gross potential revenue = available nights × forecast ADR.
2. Vacancy and credit loss = (1 − occupancy) × GPR, plus cancellations and chargebacks.
3. Effective gross income = GPR − vacancy, plus other income (cleaning fees charged, extras, late checkout).
4. Operating expenses: platform and channel fees, management fee, cleaning, linen, supplies, maintenance and repairs, utilities, insurance, property tax, HOA, licenses, accounting, marketing, software. Each line forecast separately (see the forecasting engine).
5. Net operating income = EGI − opex.
6. Capital expenditure: planned renovations, furniture replacement cycle (a reserve of 2 to 4 percent of revenue by default, market-adjusted), major systems.
7. Debt service: interest and principal from the loan schedule.
8. Taxes: income tax per the owner's regime, VAT where applicable, tourist tax pass-through.
9. Net cash flow to owner, and at exit: sale proceeds = exit value − selling costs − debt payoff − capital gains tax.

7.2 Metric set

7.3 Uncertainty

Point estimates alone are how consultants get sued. Every forward metric is reported three ways: base case, a low-high band (P10 to P90) from Monte Carlo over the forecast distributions of the forecasting engine, and the result under named stress scenarios (see the scenario analysis). The Monte Carlo runs 10,000 paths per property in under 10 seconds; correlations between ADR, occupancy and cost inflation are estimated from the platform index, not assumed independent.

7.4 Sensitivity

For each decision the engine produces a tornado analysis: the change in IRR and NPV for a ±10 percent move in each driver (ADR, occupancy, cleaning cost, maintenance, interest rate, exit cap rate, tax rate, renovation cost, hold period), ranked. The memo quotes the top three, which is what a consultant would say in the meeting: "your return depends mostly on occupancy and exit value; costs matter less than you think".

8 Forecasting engine

Every line of the pro forma is forecast separately, with its own model, its own backtest and its own uncertainty band. The engine is a model registry, not one model: for each line and each property it trains or selects several candidates, backtests them on the property's own history and on its market cell, and keeps the one with the best out-of-sample error. Forecast horizon: monthly for 36 months, then annual to the end of the hold period, converging to the market cell's long-run levels.

8.1 What is forecast and how

8.2 Backtesting and model selection

• Every forecast is tested against the past: the model is asked to predict the last two years from the data before them, and only the best performer per line and per market is used; the test is repeated every month.
• A property with less than a year of history is forecast from its market cell with an adjustment for its attributes and reviews; the memo says so.
• The memo states how often the bands were right: "our range has contained the actual cleaning cost 84 percent of the time over the last year".
• Every forecast explains itself: why the rate is expected to fall, which three factors moved it.

8.3 Events and shocks

An events layer overlays the statistical forecast: known future events in the market cell (festivals, conferences, infrastructure openings, regulatory changes with a date, large supply additions such as a hotel opening) add or subtract demand with a magnitude estimated from similar past events in the platform data. The risk agents feed this layer; every overlay is listed in the memo with its source.

9 Trend and anomaly detection

The forecasts say where a property is going; this layer says what is already happening that the owner has not noticed. It runs nightly over the ledger and the platform index and produces signals: typed, scored, explained, and attached to the property until acknowledged. Signals feed the memo's risk section and the manager's portfolio view.

9.1 Trend signals

9.2 Anomaly signals

9.3 Scoring and lifecycle

• Each signal has a type, a monetary impact estimate (annualized EUR), a confidence (from the test's significance and the data volume), and a suggested action.
• Signals are ranked by impact × confidence; the portfolio view shows the top ten across all properties, which is the manager's weekly to-do list.
• A signal is suppressed when the user acknowledges it or marks it expected (an owner who knows the plumbing is being redone), and re-raised if the trend worsens.
• Every detector has a precision target measured on labelled historical cases (platform data has thousands of known vendor changes, renovations and suspensions to label); detectors below 70 percent precision are not shown to users.

9.4 Proactive opportunity signals: where the data, the engines and the agents meet

The owner does not ask "would a second bathroom pay?" because it never occurs to them. The platform knows: the neighbors with two bathrooms earn 18 percent more per night in the same street with the same bedroom count. This is the synergy the proposal sells, and it runs without anyone asking. Each month, for every property, the module runs a gap analysis against its top-quartile peers in the cell, prices every gap with the engines, and sends the owner and the manager the three opportunities worth the most, as a short message with a one-click "show me the full analysis" that opens a memo.

How a proactive suggestion is produced

1. Peer set. Listings in the same cell (district, type, bedrooms) with RevPAR in the top quartile over 12 months; at least 15 peers, else the parent cell.
2. Attribute gaps. The property's attributes versus the peer set: bathrooms, pool, air conditioning, parking, workspace, balcony, elevator, pet policy, instant booking, professional photos, review score, minimum stay, cancellation policy, listing on additional channels.
3. Value of each gap. A hedonic model on the platform data gives each attribute's contribution to ADR and occupancy in that cell, with a band; the financial engine turns it into annual NOI, the capex estimate (from platform renovation data and local cost indices) into an IRR and a payback.
4. Feasibility filter. Structural gaps (a second bathroom in a 45 m² studio) are flagged as "check feasibility"; operational gaps (instant booking, photos, pricing) are marked "this week".
5. The suggestion. The LLM writes three sentences per opportunity from the engine output: what the neighbors have, what it is worth, what it costs, what to do first. The owner replies in chat or opens the memo.
6. Learning loop. When an owner acts, the platform records a before-after case that sharpens the hedonic model for the next owner; the suggestion engine improves with every renovation on the platform.

Examples of proactive messages

Guardrails for proactivity

• At most three suggestions per property per month, ranked by value; nothing under a materiality floor (200 EUR a year or 2 percent of NOI).
• Every suggestion carries its peer count and band; a thin peer set says so.
• Structural suggestions are advice to investigate, never instructions to build; the memo lists the professionals to consult.
• Owners choose the channel (email, Hostify inbox, WhatsApp) and can mute categories; managers can review suggestions before they reach their owners.
• Hostify's own interests (an upsell, a marketplace listing) are labelled as such in the message.

10 Scenario and decision analysis

A scenario is the base model with a defined set of changed inputs and a start date. The engine runs every strategic option through the same model, the same Monte Carlo and the same stress cases, then ranks them. The user can edit any assumption and rerun; the memo records which assumptions were the user's.

10.1 Strategic options and how each is modelled

10.2 Stress cases (always run)

10.3 Ranking and the recommendation rule

• Options are ranked on risk-adjusted NPV at the owner's discount rate, with IRR, cash-on-cash and P10 NPV shown beside it; the P10 column is what separates a brave recommendation from a reckless one.
• The recommendation is the top option only if its P50 NPV beats the base case by more than the model's own error band and its P10 is not materially worse than the base case's P10; otherwise the recommendation is "hold, and here is what would change the answer".
• Constraints are respected before ranking: liquidity needs, minimum DSCR, a date the owner cannot sell before, tax deadlines, an owner who refuses leverage.
• The switch analysis names the assumption and the value at which the ranking flips ("sell beats hold if the exit cap rate rises above 6.1 percent, or if occupancy stays below 58 percent"). That sentence is the heart of the memo.

10.4 Portfolio level

For a manager or investor the same engine runs over the set: aggregate cash flows, portfolio IRR, concentration by city and property type, correlation across cells from the index, and the marginal effect of selling or buying one property on portfolio risk. The output is a ranked list of actions ("sell these two, refurbish these three, convert this one") with the combined effect, which is the manager's quarterly conversation with every owner, prepared.

10.5 Debt and leverage logic

A loan changes three things: it multiplies the return on the owner's own money, it adds a fixed cost that does not fall when bookings fall, and it adds interest-rate risk. The module makes all three visible rather than hiding them in one IRR.

• Does the loan add value? Only when the property's unlevered cash yield after tax exceeds the after-tax cost of debt. The memo states the spread in one line ("your property yields 6.8 percent, the loan costs 4.3 after tax deduction: leverage adds 2.5 points on every borrowed euro") and the rate at which the spread turns negative.
• How much is safe? The engine scans leverage from zero to the lender's maximum and reports, at each level, the levered IRR, the DSCR in the base and stress cases, the break-even occupancy and the probability of a year in which cash flow does not cover debt service. The recommended level is the highest at which that probability stays under the owner's tolerance (default 10 percent) and DSCR stays above 1.25 in the demand-shock case.
• Fixed or floating? Both schedules are run under the consensus rate path and the +300 basis point shock; the comparison includes fees, prepayment terms and the owner's ability to absorb a payment increase.
• Refinance or not? Net present value of the new loan against the old, including penalties and fees, plus the value of any cash-out at the owner's alternative return; break-even months to recover the fees.
• Taxes. Interest deductibility by regime, the effect on after-tax IRR, and the capital gains interaction at sale.
• Lender view. For every property and portfolio the module keeps a current DSCR, LTV and debt yield with history, which is the underwriting file a bank asks for and the monitoring feed it would pay for after closing.

11 Risk intelligence agents

Four agents watch the world outside the ledger. Each runs on a schedule, reads curated sources with web search behind an allowlist, extracts structured facts, scores them for the property's location, and writes risk signals into the same signal store as trend and anomaly detection. A human analyst at Hostify reviews new regulatory and political facts before they change a forecast; market and climate facts flow automatically from datasets. Nothing an agent reads on the web is treated as an instruction; everything is extracted into typed fields with a source link.

11.1 How a risk becomes a number

1. The agent extracts a fact: "City council vote on a 90-night cap scheduled for March 2027; three of five committee members in favor".
2. The analyst confirms it and assigns, with the agent's suggestion, a probability and a magnitude (cap applies to this property; revenue above 90 nights lost; conversion to long-term as the fallback).
3. The scenario engine runs the probability-weighted case; the memo shows the base case, the capped case and the weighted expectation, with the source linked.
4. When the vote happens, the probability becomes 0 or 1 and every memo for the city updates on the next run, with a change log the owner can read.

11.2 Scores the owner sees

Each property carries four scores (0 to 100, with a one-paragraph rationale and the three facts behind it): market, regulatory, political, climate. They appear on the memo's first page and in the marketplace listing (see the business case), where they become a selling point for well-placed properties and a price factor for exposed ones. Scores are recomputed monthly and on any confirmed event.

11.3 Agent guardrails

• Allowlisted sources per agent and country; a fact from an unlisted source is logged but not used.
• Every fact has source, date, extraction confidence and analyst status (new, confirmed, rejected, superseded).
• The agents never write to the ledger or the model; they write signals, which the engine consumes under the analyst's status rules.
• Cost: each agent has a daily budget per market; markets with no active properties are not watched.
• Bias check: the political agent reports facts and institutional ratings, never commentary; the memo quotes ratings and dates, not opinions.

12 Advisory output: the investment memo

The product the owner holds in their hand is a memo, four to six pages, written the way a senior consultant writes for a client who will act on it: the recommendation first, the numbers that support it, the risks that could overturn it, and what to verify. It is generated by an LLM from the engine's outputs and the signals, under a strict template, and every number is a link to its engine run. A chat interface sits on top of the memo for follow-up questions ("what if I sell in 2028 instead?"), and every answer in chat is produced by rerunning the engine, never by the model reasoning about numbers.

12.1 Memo structure

1. Recommendation (half a page): the option, the expected result in one number the owner cares about (NPV gain versus holding, or IRR), the confidence, the one assumption it depends on most, and the next step. Example: "Sell after the 2027 season. Expected proceeds 412,000 EUR, 58,000 EUR more in present value than holding ten years, with 70 percent confidence. The result turns if the district's exit cap rate rises above 6.1 percent before then. Next step: get a formal appraisal in February 2027."
2. The property today: trailing-12-month operating statement versus the market cell, value estimate range, the four risk scores, the top three signals.
3. Forecast: five-year revenue, cost and net cash flow with bands; the three drivers that matter (tornado); the calibration statement.
4. Options compared: one table, every option, IRR, NPV, cash-on-cash, P10 NPV, payback, first-year cash impact; a one-line verdict each.
5. Risks and stress cases: what each stress does to the recommended option and to the base case; the regulatory and climate facts with sources; the switch points.
6. Assumptions and what to verify: every assumed input with its source and confidence, the user's overrides, and a checklist (appraisal, tax advice, insurance quote, contractor quote).
7. Appendix: full pro forma, model versions, data provenance, methodology notes.

12.2 Writing standard

• Lead with the decision; never with methodology.
• Numbers with units and dates; ranges as "380 to 440 thousand EUR", never "significant".
• Plain language for owners, with a "for your advisor" appendix that uses the technical terms.
• Honest about uncertainty: the memo says what the model cannot know (a planning decision, a neighbor's hotel) and how much that matters.
• One voice across the platform; a consultant's, not a marketer's.
• The memo states what it is not: not a formal valuation, not tax or legal advice, and names the professionals to confirm with (see the guardrails).

12.3 Delivery

• In Hostify: a Memo tab per property and per portfolio, regenerated monthly and on any confirmed risk event, with a diff view ("what changed since last quarter").
• Exports: PDF for the owner, xlsx with the full pro forma for their accountant, a data-room pack (verified history plus forecast) for a buyer or lender.
• Chat follow-ups with a visible rerun of the engine and the changed inputs shown.
• For managers: a quarterly owner review pack generated for every property, which is the manager's retention tool and Hostify's reason to charge more.

13 Architecture

Data flows down: sources into one ledger, the ledger into the three analytical engines, their outputs into the scenario engine, and only the ranked, computed result reaches the LLM that writes the memo. A chat question reruns the scenario engine and the memo writer reads the new output; it never reasons about numbers itself.

13.1 Integration with Hostify

The module reads Hostify's data, never writes into bookings or payments. Owners enter their purchase, debt and tax details once in the owner portal. Memos and signals appear as a tab on each listing and as a page on the manager dashboard, and important events (a new memo, a confirmed regulatory change, a high-value opportunity) arrive through the notification channels Hostify already has. The module runs alongside the platform, so it ships without touching the booking engine.

14 Guardrails, explainability and compliance

Advice about buying and selling property attracts regulation, liability and, if it is wrong, lawsuits. The module is built so that it can be defended line by line.

14.1 Numbers and the LLM

• The LLM never calculates. Every figure in a memo or a chat answer is read from an engine run; a post-generation validator checks each number in the text against the run and rejects the draft if any differs or is unreferenced.
• The LLM never invents an input. Assumptions come from the ledger, the rule tables, the owner or stated defaults; the memo lists all of them with their provenance.
• Recommendations follow the ranking rule; the LLM phrases it, it does not choose it. A memo cannot recommend an option the engine did not rank first under the rule.
• Temperature low; structured output; the same run and template always produce the same recommendation, which is tested.

14.2 Explainability

• Every forecast carries its drivers (feature contributions or components) and its calibration history; every signal its evidence; every risk score its three facts.
• Every memo has a "how we got here" appendix: model versions, data window, assumptions, overrides, calibration, and what changed since the last memo.
• The owner can click any number to see the run, and any run to see the inputs.

14.3 What the module is, legally

• The memo states on its first page that it is analytical support based on operating data and public information, not a formal valuation, not investment, tax or legal advice, and that decisions should be confirmed with a licensed appraiser, tax adviser or lawyer, named by role.
• Hostify's legal counsel reviews the wording per jurisdiction; in markets where automated valuation or investment advice is regulated (several EU states, the UK, the US), the module operates as a decision-support tool for the owner's own property and the marketplace's valuation range is labelled an estimate.
• Data rooms shared with buyers or lenders carry the owner's explicit consent, a scope and an expiry.

14.4 Privacy and data use

• The cross-platform index uses k-anonymity (30 listings, 5 accounts per cell) and never exposes a single account's figures; owners can opt out of contributing, in which case their properties still receive forecasts from the index but do not feed it.
• Owner capital data (purchase price, debt, taxes) is visible only to the owner and, if they allow it, their manager; it never enters the index.
• Personal data stays in the EU region for EU accounts; retention follows Hostify's policy; deletion requests propagate to runs and memos.
• Hostify's terms of service must state the analytical use of operating data; the proposal includes the clause.

14.5 Security

• The module reads platform data; it has no way to change bookings or payments.
• Every user sees only their own properties, or the ones they manage; every input change, override and analyst decision is logged.
• The AI components follow the same security standard defined for the voice assistant project: identity-bound access, outside content treated as data, vetted sources only, output filtering, and a review queue for anything off purpose.

14.6 Fairness and bias

• Valuation and forecast models are tested for systematic error by district and property type, so a model does not undervalue one neighborhood because it is thin in the data; cells below the data threshold fall back to a wider parent cell with a wider band, never to a point estimate.
• The political risk agent uses institutional ratings and dated facts, not sentiment; the climate agent uses public hazard datasets with their methodology linked.

15 Demo plan

The demo must show Hostify three things in twenty minutes: that their own data produces a consultant-grade memo, that the forecasts are honest and backtested, and that the risk agents find things a human would miss. It runs on a sample of Hostify's real data (anonymized, 200 to 500 listings across three markets, three years), or, if that is not available before the meeting, on a synthetic dataset generated with the same schema and realistic dynamics, with the synthetic nature stated.

15.1 Demo script

1. One property, one question. Open a two-bedroom apartment in Florence with three years of history. Ask: "Should I sell this?" The memo appears: recommendation, number, confidence, the switch point.
2. Where the number comes from. Click the IRR, see the run, the pro forma, the bands; click ADR, see the forecast drivers and "our bands were right 86 percent of the time".
3. A signal the owner missed. The signals panel shows cleaning cost drifting 23 percent above market and a maintenance acceleration in plumbing, each in euros per year.
4. Options side by side. Hold, renovate (with the uplift from matched listings), convert to long-term, sell in 2027: the table and the tornado.
5. The world outside. The regulatory agent shows a pending night-cap vote with its source; the climate score shows flood exposure and the insurance path; rerun with the cap at 60 percent probability and watch the ranking move.
6. The portfolio. Switch to a manager's 80 listings: the ranked action list, the three to sell, the five to refurbish, the combined effect.
7. The chat. "What if I sell in 2028 instead?" The engine reruns, the answer cites the new run.
8. The vertical. A marketplace listing card generated from the memo: verified history, forecast, risk scores, valuation range, data room.

16 Business case for Hostify and the real estate deals vertical

The module earns three times: as a premium feature that raises subscription revenue and retention, as a data product sold outside the platform, and as the foundation of a transactions business where Hostify takes a share of deals it makes possible. The first pays for the build; the third is the reason to do it.

16.1 Revenue lines

16.2 The vertical

Today a short-term rental changes hands like any flat: a portal listing, a broker, an appraisal that ignores the operating data, a buyer guessing the yield. Hostify can be the venue where these assets trade as what they are, income-producing businesses with a track record. The sequence:

1. Verified listings. Any Hostify property can be listed for sale with its memo, data room and scores, consent-gated by the owner.
2. Qualified buyers. Investors subscribe to the marketplace, set criteria (market, target IRR, risk limits) and receive matches; the module produces the buyer's bid analysis automatically.
3. Continuity of operation. The sale closes inside the platform with the manager, the calendar and the bookings staying in place; the buyer's first memo is ready on day one. This is the feature no portal can copy.
4. Capital partners. Lenders and insurers underwrite from the data room; Hostify earns the introduction and the renewed subscription.
5. Portfolio deals. Managers' entire books become investable; funds buy or finance portfolios with Hostify as the data and operations layer.

16.3 What it does to Hostify's position

• Retention: an owner whose property's value and plan live in Hostify does not leave for a cheaper channel manager.
• Pricing power: the subscription moves from software cost to a share of the owner's return.
• Moat: the index deepens with every booking; a competitor would need the same years of data.
• Narrative: Hostify becomes the operating system and the market for short-term rental property, which is a different company to value than a channel manager.

17 Appendix A. Sample analysis: "Is a 20,000 EUR pool worth it?" (illustrative data)

All figures in this appendix are illustrative, generated with the module's synthetic dataset to show the form and depth of the output; they are not Hostify data. Property: Villa Sunset, three bedrooms, Salento coast in Puglia, four years of history, owner holds no debt, plans to keep it at least five years, discount rate 9 percent.

17.1 Recommendation

Build the pool before the 2027 season. At the P50 it adds 31,400 EUR of present value over a 10-year hold, pays back in 3.4 years and lifts the property's 10-year IRR from 8.1 to 10.6 percent. The result holds unless the summer ADR uplift comes in under 11 percent, half the level the 41 matched listings on the platform achieved, or the owner sells before 2030. Next step: two contractor quotes and a check of the building permit requirement with the municipality; both are listed under "verify before acting".

17.2 The property today

The gap to the market is the pool: 29 of the 41 top-quartile villas in the cell have one; Villa Sunset does not. The signals panel also shows cleaning cost 7 percent above market (worth 600 EUR a year) and a minimum-stay rule costing an estimated 1,400 EUR a year in shoulder season; both are included in the "hold and improve" option below as costless changes.

17.3 Forecast: with and without the pool

Payback is reached in mid-2030, 3.4 years after the build.

The P10 to P90 band on the pool path is 25.6 to 33.1 thousand EUR in 2028, from the 41 matched listings' spread of uplift; the hold path's band is narrower, 20.9 to 24.2. Calibration on this cell over the last two years: 83 percent of actuals inside the band.

17.4 Options compared

The P10 NPV column of the table matters more than the bars in the chart: the pool's downside at the P10 is still positive, the sale options' downsides are not. That is the difference between a recommendation and a gamble.

17.5 What the result depends on

The favourable moves are near-symmetric: +1.8, +1.3, +1.1, +0.6, +0.4 and +0.3 points respectively.

The uplift assumption is the one to challenge. It comes from 41 platform listings in coastal cells that added a pool between 2022 and 2025: median summer ADR uplift 21 percent, interquartile range 14 to 27, shoulder occupancy +6 points. The recommendation survives down to an uplift of 11 percent.

17.6 Risks and stress cases, pool option

17.7 Verify before acting

• Two contractor quotes for an 8 by 4 metre pool with heating; the model assumes 20,000 EUR all-in, the platform range for this region is 16 to 26 thousand.
• Municipal permit requirement and timeline; a spring 2027 build needs an application by November 2026.
• Insurance quote with the pool added; the model assumes +380 EUR a year.
• Owner's intended hold period; a sale before 2030 changes the recommendation.
• Tax treatment of the capex under the owner's regime; the model uses straight-line depreciation over 10 years.

Assumptions marked assumed in the ledger: property tax (market default), accounting fees (market default). Model versions: ADR cell model v14, occupancy v9, maintenance frequency v6, hedonic uplift v3. Calibration on this cell: 83 percent of actuals inside the P10 to P90 band over 24 months.

18 Appendix B. Sample analysis for a fund: "Where to deploy 50 million in Southern and Eastern Europe" (illustrative data)

The institutional tier answers with the index itself. Below, six market cells ranked on forward unlevered IRR, with the downside the fund's investment committee asks about first, and the two scores that decide whether a cell is investable at all.

No cell clears the hurdle at its P10; Sofia has the best downside at 5.6 percent, and Barcelona misses the hurdle even at the P50. The bands and scores behind the bars:

18.1 Allocation the module proposes

18.2 What the diligence file contains per cell

• Index history 2019 to 2026: ADR, occupancy, RevPAR, supply, exits, review quality, cost per turnover and per night, with the data behind each (listing counts, account counts, k-anonymity confirmation).
• Forward five-year paths with bands, backtest and calibration per line.
• Regulatory dossier: every rule in force, every pending change with probability and source, the analyst's notes.
• Climate exposure by peril with the dataset and the insurance path.
• Correlation matrix across the six cells from the platform data, used in the portfolio IRR distribution: Sofia and Athens correlate at 0.31, the two coastal cells at 0.74.
• Stress cases per cell and for the proposed allocation, including the combined downside the committee asks for.
• A pipeline of listings for sale on the marketplace matching the allocation, each with its memo and data room.

Priced per engagement, with the monitoring feed after deployment. The fund gets in one week what an advisory firm assembles in three months from worse data.

19 Appendix C. Sample proactive suggestion: the second bathroom (illustrative data)

This is what the owner receives without asking, in the Hostify inbox, once a month at most, with the full memo one tap away.

Behind the message: the peer set (20 two-bedroom apartments within 400 metres), the hedonic contribution of the second bathroom in this cell (+18 percent ADR, +4 points occupancy, band from the two neighbours' before-after cases and 31 similar cases across the platform), the capex estimate from the platform's renovation records in Florence, the financial engine's IRR and payback, and the feasibility flag (a wet-wall adjacency check the owner must confirm with a contractor). The same pipeline produced the air-conditioning, minimum-stay and cleaning-quote suggestions in proactive opportunity signals; this one ranked first by value.

20 Appendix D. Sample analysis: "Should I borrow 150,000 EUR to buy a second apartment?" (illustrative data)

Owner of Villa Sunset considers a two-bedroom apartment in central Bologna listed at 250,000 EUR, with a bank offer of 150,000 EUR at 5.4 percent fixed for 20 years, 1 percent arrangement fee, no prepayment penalty after year 3. The flat has three years of operating history on the platform under its current owner. Horizon 10 years, discount rate 9 percent, the owner's alternative return on cash 5 percent.

20.1 Recommendation

Take the loan at 5.4 percent, but not above 6.3. Levered IRR on the owner's 100,000 EUR of equity is 12.3 percent against 8.9 percent all cash; the property's after-tax cash yield of 6.8 percent exceeds the after-tax cost of the loan of 4.3 percent, so every borrowed euro earns a 2.5 point spread. Debt service coverage stays above 1.45 in every base-case year and above 1.12 in the demand-shock case; a rate shock does not apply because the rate is fixed. Next steps: lock the fixed rate before the offer expires, confirm interest deductibility under the owner's company regime, and keep six months of debt service in reserve.

The break-even rate is where the levered IRR falls to the all-cash 8.9 percent; above it the bank earns the owner's return. The after-tax spread alone would turn negative only at 8.5 percent, so the IRR test is the one that binds. At the quoted 5.4 percent there are 0.9 points of headroom, which the memo flags as adequate but not generous: a second bank quote is worth asking for.

20.2 Can the property carry the loan?

The one breach year (2027 in the shock case, DSCR 1.12 against a 1.25 covenant) is why the memo recommends a six-month debt service reserve of about 6,100 EUR; with the reserve the lender's covenant is never tested. Break-even occupancy, the level at which net operating income just covers the loan, is 41 percent; the flat's worst year on record was 52.

20.3 Loan options compared

20.4 Verify before acting

• Second bank quote; the break-even rate is 6.3 percent, the headroom at 5.4 is 0.9 points.
• Interest deductibility under the owner's company regime; the model assumes full deduction at 20 percent corporate tax.
• The seller's operating history is from the platform; confirm no owner-use nights were counted as blocked for other reasons.
• Six months of debt service held in reserve from closing.
• Insurance and building fees for the new flat; the model uses the cell's medians, marked assumed.

21 Appendix E. Sample analysis: "Where in Italy should I invest about 500,000 EUR to rent out?" (illustrative data)

An investor from outside Italy, no properties on the platform yet, budget around 500,000 EUR, target return 7 percent, hold 10 years, no strong preference for region or property type, willing to consider a loan. The module runs the question over every Italian market cell in the index, then over the leverage question.

21.1 Recommendation

Buy a house, not an apartment, and buy it in Puglia or Sardinia. The two best uses of 500,000 EUR in cash are a three-bedroom house with a garden in Salento (P50 IRR 9.1 percent, the narrowest downside of the top three) and a two-bedroom apartment near the sea in Costa Smeralda (9.6 percent, wider band, a shorter season). If the investor accepts a 10-year loan of 250,000 EUR at the quoted 4.9 percent, the better choice becomes a four-bedroom house with a pool in north-east Sardinia at about 750,000: levered IRR 12.8 percent on the same 500,000 of equity, DSCR above 1.5 in every base year and 1.2 in the demand shock, and a property type with 31 percent higher nightly rates per square metre than apartments in the same cell. Rome and Florence are excluded on regulatory grounds: licensing freezes in both centres put the P10 under 2 percent. Next steps: a buyer's agent in Olbia and in Lecce, a formal valuation of the shortlisted houses, confirmation of the Italian tax regime for a non-resident owner, and a second loan quote.

Rome and Florence fall below target because of licensing rules. The bands, scores and what the budget buys:

Seasonality decides the risk, not the headline return: the Sardinian cell earns 78 percent of its revenue in 14 weeks, Sicily 54 percent in 20, which is why Sicily's band is narrower and why the memo recommends Puglia or Sardinia only for an investor who can carry a quiet winter.

21.2 The five concrete options

21.3 Why the loan changes the answer

The 250,000 EUR does not buy more of the same; it buys a different asset. In both coastal cells a house with a pool earns 31 percent more per square metre than an apartment, carries a 4.7 average review against 4.5, and loses less occupancy in the shoulder season, because families book houses in May and September. The loan's after-tax cost of 3.9 percent sits well under the house's cash yield of 6.4 percent, so each borrowed euro earns a 2.5 point spread, the same test as Appendix D. The price of that spread is the first-year DSCR of 1.21 in the demand-shock case, which is why the memo pairs the loan with a six-month reserve of about 16,000 EUR and a fixed rate only.

21.4 Risks that would change the ranking

21.5 Verify before acting

• A buyer's agent in Olbia and one in Lecce; shortlisted houses need a formal valuation and a technical survey (pool permits, septic, land registry).
• Non-resident tax regime for rental income in Italy and the cedolare secca option; the model assumes the flat-rate regime at 21 percent.
• Second loan quote; the memo's break-even rate for the levered option is 7.1 percent.
• Management: a local manager on the platform at a 20 percent fee is assumed; a managed house in Sardinia without one does not work for an absentee owner.
• Six months of debt service and two months of operating costs held in reserve from closing.