Hey pal,
What you are proposing, the only issue is it isn't really feasible for $200, or anything close. I've completed a near identical project for a company with 282,000 products. It involves taking the time series, piecing into a hold out sample of the last year, building a model on everything else, then selecting the best techniuqe (Holt Winters usually) for the product. You have an additional complication because your sample dataset only includes march, june, sept and december...so missing data. If your budget is closer to $15,000 let's talk because I have direct experience with every aspect of this down to running the models in parallel so it doesn't take 3 days to fit all the models sequentially.
Best,
SstatsC