Why do Retail Data Science Colleagues Struggle with Bad Data?

Pricing and Merchandising colleagues need high-quality data to ensure their mathematical models are generating precise results. For analytical minds like us, we know the famous Junk-in Junk-out principal. We can spend an inordinate amount of time perfecting our model. Yet, if our source data is wrong or statistically biased, results won’t be effective. We’ve heard this pain from our retail data science colleagues in Pricing Sciences and Revenue Optimization. They are responsible for making the price change for their products. Yet many times, they get red face as their source data is incorrect and is generating junk results.

John Adams – a Pricing Analyst – at Boston Shoes is trying to set a competitive price for his Rockport Black Size 10 shoe. He is getting raw data on how much his competitors are charging from his pricing intelligence tool. He optimizes his price amongst his cohort of competitors – emerging and leading shoe retailers. He applies his advanced pricing elasticity algorithms. He optimizes inventory. Yet, he realizes that he is not getting the revenue growth he expected. Digging deeper, he finds that he has wrong raw data from his competitive analysis tool. The tool is giving prices for size 8.5 vs size 10. His whole model goes haywire as it isn’t doing the right comparison.

We’ve learned that this is a growing problem amongst our pricing colleagues in the apparel, shoes, cosmetics, grocery, industrial and other verticals. Product content is sparse. Images are either poor or absent. Identifiers like GTIN, ASIN or MPN are unavailable. Competitors are not listing their products on marketplaces and ad channels like Amazon, Walmart, eBay, and Google Shopping. They are only selling directly on their site.

Sounds familiar. Here’s our view – 

If you want to win the Formula 1 World Championship, get the best oil for your Ferrari. If you want to win in pricing, get the best quality data.

Here’s how we solve this problem for our retailers, manufacturer and service provider customers–

Go to the source anywhere in the world to get data. Get all the data at an SKU level. Anything and everything. Now match with algorithms. You know data is fuzzy and bad. So, your fancy Machine Learning won’t suffice. Anything fuzzy, get a trained human to verify.

Whether you do this on your own, or use a vendor, follow this principle to get pure data. This will liberate you. You can deliver the top pricing analysis you are trained to generate. With the help of pure data, do Dynamic Pricing and apply your advanced models to it. Excel your profession and rise above bad data.