Forecasting Software

Written by James Young | Aug 22, 2024 6:58:04 PM

Case Study: A store owner wants to optimize their inventory to sell more in upward trending or seasonal products, avoid over stocking down-trending products, and save time from personally making 100s of forecasts.

Step 1: I built a forecasting app that handles all technical parts of the forecasting process

Step 2: Consulting with business owner to build strategy, structure the data, evaluate models, and choose the best forecasts

Step 3: Measure savings an increased revenue by using the best model rather than the base model or field led forecasts. Continue to refine the models and process.

Industry Success

Forecasting is one of my areas of expertise and was one of my strongest focuses while working at NetApp. I forecasted thousands of different SKUs split out by geographic area at multiple distances into the future. My forecasts cut error size 30-50% relative to the field led forecast.

Outcomes of Value from this Case study

While the models here are limited relative to what I have done at NetApp, we already see a ~50% drop in error between the best and worst models
80% and 95% prediction intervals to handle “what-if” scenarios and quantify uncertainty
Time savings to put into more creative and human-centric tasks

Step 1: I built out and end-to-end forecasting app to take user data, back-test different forecast models, and provide tools to pick the best forecasts for planning going forward.

Step 2: Evaluating models on backtest performance to pick the most accurate model. We see that the ARIMA model (pink line) and ETS model (green) are both significantly lower in error (WMAPE) relative to the more basic models. Step 2 Continued: Picking the best model and planning based off of its forecasts. We can use the app to see how different models forecast the forward looking numbers for different products. Step 2 Continued: Finding insights about the biggest movers. Preparing for new trends or seasonal swings can help you capture new opportunities and investigate potential drop-offs.

Step 3: Download forecasts for record keeping, plan based on best-model values. As the actual sales develop, compare them to the forecasts and measure your improvement over base model forecasts to quantify the financial value of more accurate forecasts.

View full post