Abstract
Fabric is a Robotics fulfillment company that aims for same-day order fulfillment within cities. The goal of the project is to determine the optimal amount of stock in Robotic facilities in an effort to conserve storage space. For that purpose, it is necessary to create a reliable forecast of the future demand based on previous sales data.
Challenges
- Forecasting on a small positive integer time series.
- Building models based on limited information.
- Dealing with partial information regarding the promotions.
Achievements (according to KPIs)
- Per-SKU SARIMAX models for demand prediction
- Evaluation on 4 intervals
- Stock recommendations according to forecast
Further development
- Provided complete information, using more complex (and nonlinear) demand prediction models: LSTM-based models, VARMAX
- Stock optimization with constraints: linear or stochastic programming
Supervisor Feedback
During the project, Yulia presented a wide range of DS and DE skills As well as presenting the Model to different key figures in the company.
The model that was developed already shows a big improvement on the current stocking logic used by our customer. The reason we still haven’t begun in implementing the MVP recommendations is related to Customer-Success details but our aim is to start using it soon.
It’s important to say that besides focusing on the specific time-series model that was chosen based on the problem and the data at hand, Yulia continued to explore other ways that were used to solve the problem in the past and challenging her own model. This is something that I value very much in a Data Scientist.
Yulia was our first intern in the company. I hope all of our following interns would be as professional and pleasant to work with as her.
– Dima Krochek, Head of Data and Solution Product
