The Algorithm Says So

Back in the stone ages of email marketing (2004), my team at Nordstrom would assign one of up to ten different email creative treatments to a customer, on a twice-weekly basis.

We used transactional data and customer preferences, back in the day ... we wouldn't have clickstream data integrated with retail and e-commerce purchases until late 2006. Based on testing, we knew that purchase transactions were about 10x more important than customer preferences. In other words, if the customer said she wanted to see merchandise tailored to a 30 year old customer but purchased merchandise tailored to a 55 year old customer, the purchase data was 10x as effective at predicting the future. That's a big deal. Customers spent more when they were sent content aligned with their purchase history. Customers spent less when they were sent content aligned with their personal preferences.

My boss at the time wanted to see merchandise tailored to a 30 year old customer.

My boss at the time purchased merchandise tailored to a 55 year old customer.

And so, one lazy afternoon, my boss asks me why she keeps getting "old person emails"? She's not happy. She expressed her preferences online, and she wants to see product aligned with her preferences.

"Why do I keep seeing old person emails?", she asks.

I said, "because the algorithm says so".

Here's what I want for you to think about today. Which outcome is "best"?

1. Give the customer what the customer wants, making less profit in the process but making the customer happy.
2. Give the customer what the customer is likely to buy, making more profit in the process.

Discuss how you would approach this issue. Describe the cases where you let algorithms make decisions. Describe the instances where you override algorithms, even if it costs you profit. Most important - describe your thought process. How do you balance profit with customer desires and company success?