As e-commerce continues to grow rapidly, data on online customer shopping habits and behaviors also continues to expand. Customers are typing in millions of searches an hour, and each results page is a data point that can be fed back to improve the next search’s results. For Wayfair, this is necessary to its success. Exclusively an online retailer, Wayfair does not have a showroom for customers to feel or observe products. Instead, a customer will utilize Wayfair’s search engine to parse through the company’s catalog, often using broad terms such as “living room décor”, and expect appropriate results. Roughly 70% of customers do not look beyond the first page of results, indicating that customers will gauge an online retailer’s catalog on the first results page’s accuracy. It is vital for Wayfair then to have a sophisticated search engine that complements the online furniture shopper, and it will need to continually validate itself as customer habits change, new products are introduced, or furniture trends emerge.
In the short term and stretching beyond to the long term, Wayfair is developing its own machine learning models to improve its search engine, and consequently, its effective product offering to an online customer. The emphasis on software and algorithmic improvements is highlighted in the percentage of employees working on software, which is over 50%. There are two areas in which data is being extracted and fed into a machine learning model: 1) query click data results and 2) customer feedback and reviews. In the first, a query is initially categorized using Natural Language Processing (NLP), and a results page is provided. The click data for the results page is then fed back into the model to determine how well the NLP was able to decipher the customer’s intent and provide accurate results. This feedback improves the NLP, and is especially effective with the broad base of repeated queries. In the second, customer feedback and reviews are analyzed for repeated key terms, as these are often the same terms that customers search for. These key terms are then ascribed to products and are factored in as additional data to the machine learning model that can then better rank products to improve the results page.
It is important for Wayfair management to understand where machine learning may fall short. One area is in what is dubbed, the “Long-Tail of Search”. There are less frequent, more unique queries where machine learning models do not have sufficient data to guide customers to the appropriate products. Another area is with changing trends, such as when a search term takes on a new connotation or meaning based off of social trends. For example, “hip”, “sick”, and “dope” searches would be unique challenges.
As Wayfair’s management continues focusing on utilizing machine learning to improve its search engine, it should also consider other steps to further augment its success. One area of potential improvement is on its website platform itself. Wayfair can apply machine learning is in website layout, where machine learning could be used to determine optimal layouts that improve customer satisfaction and/or result in increased customer traffic to the website. Separately, the organization could also consider improving its supply chain management by utilizing machine learning to cross-integrate online customer behavior and supply chain planning. If, for example, data could be gathered to support the percentage likelihood of a specific product being purchased when a customer searches XYZ, then the supply chain could be anticipatory or predictive in a sense. This could vastly improve shipping times and overall customer satisfaction. Lastly, machine learning could be applied to determine which products Wayfair does not need to carry. There are significant costs to maintaining SKUs, and Wayfair carries ten million products from over ten thousand suppliers. If certain SKUs do not produce sales, have a low click rate, and do not increase traffic to the website, it may be a worthwhile to remove them. This data could be mined and fed through a machine learning model to quickly determine the value of all ten million products to the company.
As an online retailer, Wayfair can easily capture a vast amount of data on its customers upon which it can feed this data into numerous machine learning models for valuable insights. However, if machine learning is self-optimizing, does Wayfair need to continue focusing on improving its machine learning models? Or can it move on to different applications once management believes the model is self-sustaining? (790 words)
 Wayfair Technology Blog, “How We Use Machine Learning and Natural Language Processing to Empower Search,” https://tech.wayfair.com/2018/10/how-we-use-machine-learning-and-natural-language-processing-to-empower-search/, accessed November 2018
 Interview with Wayfair employee, November 12, 2018
 Wayfair Technology Blog, “How We Use Machine Learning and Natural Language Processing to Empower Search,”