How Data Science supports us to predict the unpredictable
Just like for any store, forecasting is an important tool to support the business and its profitability. At bol.com we have worked on several major iterations of our forecast. One of the hosts worked on a large leap in forecast performance over 8 years ago. You can check some another leap we have made while listening to our podcast episode from May 2019 - nom more crystal ball for forecasting, Data Science it is.
In this episode, we explore how we have improved on that one.
What the episode covers
One of the leaps we made Is in the operational part of running a Data Science-based forecast in production. This is intriguing because this mechanism at the same time speeds up our innovation speeds because it enables us to do more experiments and train our models more often.
One important input we added to our forecast during the pandemic was the "Covid Severity Index". This improved our forecast performance and enabled us to work with scenarios. The latter also supports all the other retailers that sell on our platform.
Discussed Items on forecasting with Data Science in Covid-19 Times
- Forecasting in general
- Specific forecasts for specific needs
- Aggregation
- Coherence
- Who is responsible for what?
- What can we predict with Data Science?
- How do you forecast sales Data Science?
- Tech used
- Input data
- Experiment by using branches to innovate faster
- Airflow
- Dr. Watson as a gatekeeping tool to validate the output
- Corona impact on forecasting
- Covid Severity
Guests
- Catia Silva – Data Scientist in forecasting. Presenting at conferences
- Eryk Lewinson – Data Scientist in Forecasting