Good forecasts reduces excess inventory that would require markdowns and disposal. There are also cost of space and time/system used to monitor inventory. The savings are significant when you have a huge number of SKUs and/or carry lots of inventory.
One approach I had developed for an global data science competition is as below. It gives pretty good results. It is easily understandable (keep it simple approach), one can compare the accuracy using root mean square error (RMSE) with your current forecast, and most interesting, evolve your own algorithm to give weightage or overrides base on how sales evolves. I achieve an RMSE of about 1.5 :)
Step 1 - First review all the SKUs its date, quantitative or category variables to decide which variables(columns) to ignore, and which outcome(rows) to delete. With a huge number of SKUs, you certainly can't focus on all of them, so you can apply the Pareto rule. If you only have one or a few SKUs, you can skip this step.
Step 2 - Apply regression on each monthly/weekly/daily sales for each SKU as outcomes using the remaining variables as inputs. Remove non-significant factors and run the regression again to get the adjusted coefficients. If you have more than 16 variables, use minitab, SPSS, SAS or r.
Step 3 - Apply regression equation to only the significant factors to come out with outcomes on monthly/weekly/daily basis. Look at your significant factors with time. Does it make sense and did you get new insights in how your sales are affected by different variables.
Step 4 - Use RMSE to compare this new forecast with your older forecast. If the forecast is not better, adjust using your knowledge of the market using weightage, omitting factors, averages or comparison to multiple of averages.
Step 5 - Add in marginal contribution and time series analysis if you want to factor in profit margins and time trends.
Step 6 - Implement changes proposed significant factors identified from the analysis done the the above steps for continuous, self-adjusting capability.
Step 7 - Synchronize sales with operations and capacity planning.
Write to me if you need more help! :)
This blog covers new pull supply chain responsiveness and logistics concepts for hubs with good air and sea-freight connectivity like Singapore. Big data and web analytics are creating new demand opportunities, and help operations meet growing global regulatory standards. Very often, my work also involves helping online retailers improve operations. Discussions spans from raw materials serialization, to manufacturing, marketing and sales. Visualization and analysis techniques are also shared.
No comments:
Post a Comment