Problem-Solving with Data for a Better Business
When working with large datasets, the smallest anomalies can throw a wrench in predictive analysis. For example, if a company manually enters data into its database, a human error like mistyping or improper timestamps in the training data of a machine learning model may give you reduced accuracy results.
At ODSC East’s Accelerate AI Business Summit, professionals from a multitude of fields gave presentations on how they are using artificial intelligence to provide better metrics for their businesses. Kaitlin Andryauskas, Business Intelligence Manager for Wayfair, talked about how her company tackles problems with their data.
[Related article: Integrating Textual and Visual Information Into a Powerful Visual Search Engine]
“As the saying goes, junk in, junk out,” Andryauskas said.
And that’s a true statement, normally — if the results you’re getting aren’t what you expected, you may want to check the incoming data. Wayfair is an online furniture marketplace, and since its formation, it has become vertically integrated to include an international supply chain. Andryauskas opened with a story about Wayfair’s beginnings, detailing how shipments would arrive at the warehouse despite data showing plenty of square footage available.
The first thing Andryauskas did was get to the floor. She explained that transparency with all divisions of labor was necessary for making sure the proper solution was found. They worked together to discover dimensional errors for certain storage items, and realized that the system of manual entry wasn’t working as well as it should. In addition to close collaboration and partnership, Wayfair’s data team also built a reporting portal that allowed workers on the ground to identify shipping issues at any time, rather than waiting until the very end of the month to fix errors.
[Related article: 7 Tips for Visual Search at Scale]
Since implementing these changes, Wayfair’s business intelligence team has increased their shipment arrival date compliance from less than 10 percent to nearly 99 percent accuracy. So if you’re having issues with your predictive measures, it may be helpful to go to the source with transparency and honesty — it may just be the answer to making your business run more efficiently.
Read more data science articles on OpenDataScience.com, including tutorials and guides from beginner to advanced levels! Subscribe to our weekly newsletter here and receive the latest news every Thursday.
