Back to home

Pilot Backtest Transparency

Model Performance on Public North Italy Market Data

These results are from backtests on public market data, not deployed hotels. Production results are pending pilot rollout. The referenced data file isnexorev_north_italy_hotel_market_data.json.

Occupancy forecast MAPE
9.8%

Mean absolute percentage error on public-data monthly market backtest

Occupancy forecast RMSE
6.4 pts

Root mean squared error measured in occupancy percentage points

RevPAR lift simulation
+7.6%

Simulated dynamic-pricing lift vs. baseline market ADR, not deployed hotel revenue

Production hotel results
Pending

Requires PMS integration and pilot rollout with real hotel data

Forecast error grows with horizon (occupancy MAPE)

20%15%10%5%0%4.1%7 days6%14 days9.8%30 days13.4%60 days17.2%90 days
Mean absolute percentage error from the public-data backtest. We publish the error curve rather than a single headline number, because a 7-day forecast and a 90-day forecast are not the same product. Longer horizons are directional, not precise.

Forecast vs. actual occupancy — public-data backtest

90%80%70%60%50%JFMAMJJASONDActualModel forecast
Monthly occupancy for a North-Italy market, reconstructed from public ISTAT/ENIT series. Solid line is the recorded outcome; dashed line is what the model produced without seeing it. This is a backtest on public market data, not a deployed-hotel result.

Occupancy vs Forecast

The chart compares actual public-market occupancy aggregates with the forecast used in the pilot backtest. This is not hotel PMS data.

Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

Cyan: actual aggregate occupancy. Purple: forecast occupancy.

ADR Distribution

Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

RevPAR Lift Simulation

Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

Simulated lift is calculated from market aggregates and dynamic-pricing guardrails. It should be treated as a pilot hypothesis until hotel PMS pilots produce real outcomes.

Data Limits

  • Backtest data is market-level and public-data based, not individual hotel reservation data.
  • Production deployment requires PMS integration, channel-manager rate feeds, and property-specific guardrails.
  • Revenue lift is simulated and should not be presented as customer revenue or guaranteed performance.
  • First pilot hotels will be used to replace these public-data benchmarks with measured property-level results.
Read methodology and source list

Understanding the Backtest Methodology

The metrics on this page are derived from backtests on aggregated public market data for North Italy hotel markets. The dataset combines monthly occupancy rates, average daily rates, and tourism arrival figures from ISTAT, Banca d'Italia, and ENIT. The referenced data file (nexorev_north_italy_hotel_market_data.json) is available for download so that anyone can verify the inputs and reproduce the analysis.

Walk-forward validation is used to prevent overfitting. The model is trained on a rolling window of historical months and tested on the immediately following period. At no point does the model see future data during training. This protocol mimics how the model would perform in real-time deployment and produces accuracy estimates that are more conservative than in-sample testing.

What 9.8% MAPE Means in Practice

A Mean Absolute Percentage Error of 9.8% on monthly occupancy forecasts means the model's predictions are, on average, within about 10 percentage points of the actual occupancy rate. For a hotel with 70% actual occupancy, the forecast would typically fall between 63% and 77%. This level of accuracy is useful for pricing and staffing decisions but is not precise enough for same-day operational planning.

The model's performance should be compared against your current forecasting approach. If your hotel does not formally track forecast accuracy today, any systematic forecasting methodology is likely an improvement over ad-hoc estimation. If you already use a structured forecasting process, ask for a side-by-side comparison during the pilot evaluation.

The 6.4 percentage point RMSE captures the typical magnitude of forecast errors in absolute terms. Combined with the MAPE, these metrics indicate that the model performs reasonably well on aggregate market data but will need to be re-evaluated once property-specific PMS data is available. Individual hotel demand patterns can diverge significantly from market averages, particularly for properties with unique positioning, niche segments, or irregular event-driven demand.

RevPAR Lift Simulation: What It Is and What It Is Not

The +7.6% simulated RevPAR lift is not measured revenue improvement from a deployed hotel. It is calculated by applying Nexorev's dynamic pricing rules to historical occupancy and ADR data, then comparing the simulated outcome to a baseline of static market-average pricing. The simulation assumes that rate changes would not significantly alter demand volume, an assumption that may not hold in practice, particularly for large rate increases.

This simulation provides a directional estimate of potential impact, not a guaranteed outcome. Actual RevPAR changes in production will depend on the hotel's competitive positioning, demand elasticity, channel mix, and operational execution of rate recommendations. The first pilot hotels will be used to measure real RevPAR impact and replace these simulation-based estimates with verified production data.

Known Limitations and Applicability to US Markets

The current backtest has several known limitations. First, the data is market-level aggregates, not individual hotel reservation records. Market averages smooth out the variability that individual properties experience. Second, the backtest does not account for competitive response: in practice, competitors adjust their rates in response to your changes, creating feedback loops that the simulation ignores.

Third, the North Italy dataset may not generalize directly to other markets. Demand patterns in US resort destinations, urban business-travel markets, or tropical leisure destinations follow different seasonal and cyclical patterns. The model architecture is designed to adapt to new markets, but accuracy metrics should be verified on each market's data before production deployment. Pilot hotels in new geographies will generate market-specific accuracy benchmarks that will be published alongside the North Italy results.

How to Evaluate These Results for Your Hotel

When evaluating Nexorev's model performance data, consider three factors. First, compare the metrics to your current forecasting accuracy. If you are not tracking forecast accuracy today, any systematic forecasting is likely an improvement over ad-hoc methods. Second, assess whether the data sources and methodology are relevant to your market and property type. Third, ask for a pilot-specific backtest using your historical PMS data, which the founder can produce during the evaluation process.

Hotel owners evaluating revenue management software should request transparency from every vendor they consider. Ask for published MAPE and RMSE figures, the dataset used for validation, whether the results are from backtests or production deployments, and whether the methodology has been independently reviewed. Nexorev publishes all of this information because the founder believes it should be standard practice in the industry.

Book a founder call