Wayfair: Finding Hidden B2B Buyers, Spotting Designers and Choosing Outreach Targets

Wayfair sells home furnishings both to consumers and to business buyers, but many businesses shop through ordinary accounts, so there is no simple way to tell which customers are buying for a business. Assume you have access to clickstream data, order history, and any other customer data Wayfair holds. Answer the four parts below. ### Clarifying Questions - What counts as a business customer here: anyone buying for an organization (offices, hotels, restaurants, rental properties), or only customers who would qualify for Wayfair's B2B program? - What will the classification drive, such as program outreach, pricing or reporting, and is a false positive or a missed business buyer more costly? - Do we have any confirmed business customers, such as current program members, that can serve as labeled examples? - Which data is off limits for privacy or policy reasons, for example third-party enrichment or customer service transcripts? ### Part 1 — Signals That a Customer Buys for a Business Which attributes from the clickstream, the order history or other data would you examine to decide whether a customer is a business buyer? ```hint Compare two shoppers Picture an office manager furnishing a new floor and a family furnishing a new house. List how the two would differ in each data source you have, then turn every difference into an attribute you could compute. ``` #### What This Part Should Cover - Attributes grouped by data source, each with the reason it separates businesses from households - Cases where an attribute misfires - A way to combine many weak signals into one score or rule ### Part 2 — Recognizing Interior Designers One business segment is interior design, run either as a firm or by an individual designer. Their purchases often look like an ordinary consumer's. What information would you use to recognize that a customer is an interior designer? ```hint Many homes, one account A designer buys for several clients rather than for one home. Think about which traces of that would appear in the data even when each single order looks like a consumer's. ``` #### What This Part Should Cover - Signals that reflect buying for multiple clients over time - Behavior that separates a designer from an enthusiastic homeowner - Ways to confirm the inference rather than rely on it alone ### Part 3 — Picking 1,000 Outreach Targets from 10,000 Customers Wayfair wants to invite customers into its B2B program. You are given a list of 10,000 customers and must narrow it to 1,000 for an outreach campaign. Which attributes or variables would you use to choose them? ```hint Likely is not the same as worth it A customer who is almost certainly a business may still be a poor target. Decide what a successful outreach means for Wayfair, then rank on that. ``` #### What This Part Should Cover - The objective of the outreach and how each variable serves it - A ranking method that combines the chance of being a business, expected value and likelihood to enroll - Exclusions, and how the campaign is set up so the next selection can improve ### Part 4 — Control or Pilot Enrollment Form Wayfair is running two outreach campaigns to enroll business customers. In the Control campaign, the enrollment form requires a mobile number. In the Pilot campaign, the mobile number is optional. The interviewer gives you results data for both campaigns; the figures are not reproduced here. Explain how you would analyze the data, and what pattern of results would lead you to recommend continuing with Control or with Pilot. ```hint Follow both groups past the form Dropping a required field usually helps the form itself. Ask what the mobile number is used for after someone enrolls, and compare the two groups beyond the sign-up step. ``` #### What This Part Should Cover - Funnel metrics per campaign, from outreach through enrollment to later business activity - Checks that the comparison is fair and the difference is real - A decision rule that weighs enrollment volume against the value and reachability of the enrolled accounts ### What a Strong Answer Covers - Attributes that can actually be computed from the stated data, each with a clear reason - Attention to false positives and false negatives, and to which matters more for each use - A path from simple rules to a validated score, using confirmed business accounts as labels - Outreach and experiment decisions tied to downstream business value rather than sign-up counts alone ### Follow-up Questions - How would you validate the B2B score when only a few hundred confirmed business accounts exist? - The outreach enrolls many customers who never place a business order. What would you change in the selection? - Pilot wins on enrollments, but the sales team says it can no longer call new members. How would you resolve that?