I’m working on a project with a hotel that wants to use their data for pricing optimization, and I was curious to know what other ML engineers would do in this case.

Let’s say the hotel gives us all the data about:

• their competitors prices
• their favorite market placement within those competitors
• all nearby and internal events, including holidays (and their
  importance ranking from 0-100)
• weather
• historic internal data (sales, revenue) (own hotel for past 3 years)
• historic prices (own hotel and all competitors for past 3 years)
• metasearch data and reviews

All this data has 2 columns like date and insert_date. For example, the price that the hotel sold today (2024-04-25) for day 2024-06-01 would have date = 2024-06-01 and insert_date = 2024-04-25

Conceptually, how would you go about making a pricing optimization system that utilizes the most out of all this available data?

(For the sake of simplicity, let’s assume that all those data sources have 3 years of historic data, and there are no gaps in the data, and all data looks perfect)

This is an area where I see a lot of “gatekeeping” because companies like Airbnb with Smart Pricing don’t really want to share how this is done because this is a sellable product, so it’s hard to find non-general useful information about this.

One of my key struggles with using something like RL or offline RL is that this hotel has a low amount of rooms, and starts selling each room 365 days before. This means that all sales are done in 10 days out of those 365, so I’m struggling to think of how to evaluate what a good price/day is when 355 days are “unsuccessful”.

Thank you in advance, and feel free to brainstorm your ideas here, no idea is a bad idea.