← Back to App Docs

DealGuard — Owner’s Guide & User Manual

Welcome! This short guide shows you how to get value from the DealGuard web app—whether you’re just exploring or uploading your own data to power predictions.

What DealGuard does (in plain English)

Guardrails: sanity-checks a potential deal using your numbers (price, down payment, interest rate, NOI) and flags whether it passes common lender constraints (LTV, DSCR) and your own equity limits.
Scenarios: a quick table that shows how the deal looks if rates or down payment change.
Event-aware forecasting: a 60-day table that blends college-event days (WVU), weekends, seasonality and your competitor pricing to estimate Occupancy, Recommended ADR, and RevPAR by date.
Machine Learning (ML): upload a CSV of historical data to train a model; the app uses it to improve predictions.

Quick start (no data required)

Open your site and the DealGuard page.
In Deal Inputs, type your price, NOI, down %, rate, amortization years, and reserves/capex.
In Event & Pricing Context, set:
- Competitor ADR: your local comp-set average.
- WVU flags (e.g., Game Day) when relevant.
- Load/Seasonal/Tourism sliders (defaults are fine to start).
Review Results (LTV, DSCR, MPP) and the Scenario Table (Down% × Rate%).
Scroll to 60-Day Event Heatmap to see daily Occ%, ADR, RevPAR—event days are highlighted to spot pricing opportunities.

Tip: Changing inputs updates tables. On slow networks you might see “Building 60-day forecast…” while dates populate.

Deeper use: Train with your hotel’s data (CSV)

What to prepare

Create a CSV (export from PMS/BI) with at least these columns:

Column	Type	Example	Why it matters
`timestamp`	ISO date or YYYY-MM-DD	2025-10-11T12:00:00	The day we’re predicting.
`occupancy`	0–1 (or %)	0.78	Actual occupancy (target for training).
`competitor_adr`	number	115	Market anchor for pricing.
`wvu_load_factor`	0–1	0.95	Expected campus inflow pressure.
`seasonal_multiplier`	~0.8–1.2	1.08	Seasonality lift.
`tourism_trend`	~0.95–1.10	1.03	Broader travel trend.
`is_game_day`	true/false	true	Big event signal.
`is_move_in`	true/false	false	Semester start signal.
`is_graduation`	true/false	false	Graduation weekend.
`is_conference`	true/false	false	Conferences.
`occ_lag_1h`	0–1	0.72	Recent pickup proxy.
`occ_lag_24h`	0–1	0.66	Daily lag.
`occ_lag_7d`	0–1	0.61	Weekly lag.
`occ_lag_30d`	0–1	0.55	Monthly lag.

Minimal viable dataset: timestamp, occupancy, and a few drivers (e.g., competitor_adr, wvu_load_factor, is_game_day). More varied rows improve learning.

CSV snippet

timestamp,occupancy,competitor_adr,wvu_load_factor,seasonal_multiplier,tourism_trend,is_game_day,is_move_in,is_graduation,is_conference,occ_lag_1h,occ_lag_24h,occ_lag_7d,occ_lag_30d
2025-10-10,0.82,115,0.60,1.02,1.01,false,false,false,false,0.70,0.64,0.58,0.55
2025-10-11,0.98,130,0.95,1.08,1.03,true,false,false,false,0.85,0.78,0.65,0.60
2025-10-12,0.68,112,0.50,1.01,1.00,false,false,false,false,0.66,0.62,0.57,0.54

Upload & train

Scroll to ML Admin — Train RandomForest.
Click Choose File and select your CSV.
After training, you’ll see:
- Rows: how many lines were used.
- MAE (mean absolute error): average miss from true occupancy (smaller is better).
- RMSE (root mean square error): penalizes larger misses (smaller is better).

Interpreting: MAE 0.05 ≈ average error of 5 occupancy points (±5%). Datasets with big event spikes often show higher errors but still yield strong pricing signals.

How predictions use your data

The 60-day heatmap calls the model and heuristic drivers for each day.
Next-Hour Blend combines a heuristic and ML (weights favor event-day heuristics on big weekends, ML on normal days).

Screen-by-screen tour

1) Deal Inputs

Purchase Price, Down %, Rate, Amort Years → loan math. Closing %, CapEx, Reserves → up-front cash.

2) Results (Guardrails)

Equity Required (cash at close), LTV, DSCR, PASS/FAIL chips, and MPP (Max Purchase Price) bounded by equity & DSCR.

3) Scenario Table (Down% × Rate%)

“What if rates go up or I put more down?” Each cell shows DSCR and MPP at that combo. Green cells ⇒ the deal still “works.”

4) Event & Pricing Context

Competitor ADR, WVU flags (Game Day / Move-in / Graduation / Conference), and multipliers (Load / Seasonal / Tourism).

5) ML Admin — Train RandomForest

Upload your CSV and review Rows / MAE / RMSE. Re-train as you add fresh months.

6) Next-Hour Blend (/ml/ensemble)

Enter recent lags and click Compute to get Blended Occupancy, Recommended ADR, and RevPAR for the next hour window.

7) 60-Day Event Heatmap

One row per day: Event column shows WVU weekends (green highlight), weekends (light gray). Occ%, ADR, RevPAR help plan staffing & price.

Tip: On event days, consider higher ADR with a firm floor and longer minimum stays. On shoulder nights, use fenced discounts.

How to use it day-to-day

Weekly: glance at the 60-day table. Tighten/loosen rates around event clusters.
Daily: run Next-Hour Blend with fresh lags for late pickup decisions.
Monthly: append last month to your CSV and re-train.

FAQs

The 60-day table is empty or slow to fill. It loads 60 dates; on slow connections you’ll see “Building 60-day forecast…” briefly. If it stays blank, refresh. If your internet is fine and it persists, ask your admin to check the server.

Recommended ADR seems low/high. Check Competitor ADR and WVU flags for that date. If it’s a game day but the flag is off (or load factor is too low), ADR will be conservative.

What’s a “good” MAE/RMSE? No universal number. Track your metric over time; a declining MAE/RMSE after retraining is a good sign.

How do event days get into the 60-day table? From the WVU event calendar in the app. If dates are missing, ask your admin to add them so they appear and lift demand.

Glossary (hotel & lending terms)

Occupancy: Rooms Sold ÷ Rooms Available (e.g., 80/100 ⇒ 80%).

ADR (Average Daily Rate): Room Revenue ÷ Rooms Sold (e.g., $9,600/80 ⇒ $120).

RevPAR: ADR × Occupancy (e.g., $120 × 80% ⇒ $96).

NOI: Operating Revenue − Operating Expenses (before interest/taxes/depr/amort).

LTV: Loan ÷ Property Value (e.g., cap at 75%).

DSCR: NOI ÷ Annual Debt Service (1.25× means NOI covers payments by 25%).

ADS: Annual (principal+interest) payments.

Amortization: Years to pay down principal (e.g., 25).

CapEx: One-time improvements (renovations, roofs).

OpEx: Day-to-day costs (payroll, utilities).

Reserves: Cash set aside (e.g., 3 months OpEx).

MPP: Max Purchase Price that still satisfies equity & DSCR constraints.

Signals used by the app: Competitor ADR (comp-set anchor), WVU Load Factor (0–1), Seasonal Multiplier, Tourism Trend, event flags (is_game_day, is_move_in, is_graduation, is_conference), and short-term Occ Lags (1h/24h/7d/30d).

Best practices

Keep your comp ADR up to date—it anchors pricing suggestions.
Always toggle event flags for known big weekends.
Retrain monthly with fresh CSV rows to capture recent trends.
Use RevPAR as your north star—selling out at the wrong price leaves money on the table.

Friendly reminder

DealGuard provides decision support, not guarantees. Use your local knowledge (group demand, renovations, weather, citywide events) alongside the app’s signals to set confident, profitable rates.