The next demand signal may be outside your hotel

When a city becomes the story

We still think about Hanoi in early 2019, when the city was announced as the host of the US–North Korea summit. As we remember it, high-end rooms got scarce quickly once attention turned to the city. The summit itself ran on February 27 and 28.

This isn't a claim that every hotel could have pushed its rates the same way, or that every political headline brings a rush of bookings. It's about timing. An event can change a destination's outlook before the hotel's own reservations show it.

If you only watch your own pickup, you may be waiting for the result of a signal that was already visible somewhere else.

Hotels don't create all their own demand

Occupancy, current rates, booking pace and seasonality are still useful. They describe the business and the patterns it has already seen.

But every hotel sits in a place. A big event, a change in flights, bad weather or trouble somewhere else can shift who travels, when they arrive and where they stay. And the same signal can mean different things for different properties.

A conference near one hotel doesn't matter to every room in the city. A weather forecast isn't a demand forecast. The useful work is tying an outside event to a specific business, not collecting an impressive number of data feeds.

A weather forecast is not a demand forecast.

A forecast should explain itself well enough to be argued with

That's what caught our attention on a call about a prediction engine for physical businesses. The team described forecasts that work with little historical data and name the drivers behind each number.

We'd treat both as things to prove in a pilot, not as reasons to assume the model beats what a property does today. The practical appeal is that a manager can ask a better question: why should next Tuesday look different?

An explanation makes a forecast easier to check. It doesn't prove the named driver caused the change, and it shouldn't hide the uncertainty. Managers need to know when a signal is strong, when it's weak, and when their own local knowledge says otherwise.

Put the prediction where the decision is made

A forecast in a dashboard nobody opens is worth very little. It's more useful inside the booking, POS or industry software where people already set prices, plan shifts or order stock.

For a hotel, the question isn't just whether demand goes up. It's which dates and room types are affected, what you can still do about it, and what that move costs you. For a café, the same forecast might change what gets ordered rather than what gets charged.

The output should help someone decide what to do next. Otherwise it's just a cleverer number on a screen.

Test the decisions, not the headline percentage

We'd start with a fixed period, a clear baseline and a log of what the forecast would have recommended. Check how accurate it was and what it did for the business, and keep modeled opportunity apart from revenue actually earned.

Don't let one unusually good event take credit for a system that has to work through ordinary weeks too. Track the misses and the false alarms, because those decide whether operators keep trusting it.

Better prediction only matters when it helps a team spot something relevant and act while there's still time to choose.