BFCM Challenge 2024, Track II

Goal: Predict hourly email sends for Klaviyo throughout BFCM. Lowest SMAPE wins a prize!

Method: The simplest thing I could think of:

1. Estimate the year-over-year growth rate in email sends based on the average year-over-year growth rate for the days leading up to BFCM
2. Use the previous year's series times the forecasted growth rate
3. Do it all in 40 lines of SQL with DuckDB

Running this solution

You'll need two things to run this "model":

1. You'll need to set up DuckDB
2. You'll need to have the historical data saved to historicals.csv

Once both of those are all set, you can run the model:

duckdb --csv < model.sql > forecast.csv

This should output a CSV file called forecast.csv with the forecasted email sends!