In the past few years, digitally-native vertical brands became one of the primary ways people shopped. Think Everlane, Dollar Shave Club, Harry's, Warby Parker, and Casper. All these companies created their brand and following, online, and manage their own supply chain.
The problem then is, how do they forecast demand for their new products, through only online interactions? Pre-orders work to a certain degree, but how does their online promotional activity affect sales?
To solve this problem, we've created and designed Predictr.
Predictr uses multiple models to optimize and find the best one to fit your sales data. Accounting for trend, seasonality, and one-time promotional events, it gives you best estimates for what your future sales will be.
To build this, we've utilized R forecasting package and full-stack web development to 1) learn and find the most optimal forecasting model and 2) design the front-end tool that e-commerce retailers can use.
Coming with no backgrounds in R, we buckled down and learned the diverse forecasting models in the forecast package. We've learned of so many statistical models along the way (e.g. BATS, smoothing methods, neural networks, etc).
Future: We're looking into how this tool can be used to predict sales for new products as well!