- Thomas Thurston
Solving the Mystery of the Nile (Blue Nile, that is)
Science is about knowing when you’re wrong. After all, if you don’t know when you’re wrong, how can you improve? As data scientists who predict the rise and fall of businesses, we were wrong for a long time about Blue Nile. It’s been a mystery.
For those who don’t know, Blue Nile (NASDAQ: NILE) is an online jewelry website founded in 1999. Nearly a decade ago our algorithms predicted Blue Nile would be highly disruptive to brick-and-mortar jewelers like Tiffany, similarly to how we predicted Amazon would disrupt Borders, and Netflix would disrupt Blockbuster. We were right about Amazon and Netflix – which may look obvious with the benefit of hindsight, but those predictions were painfully controversial when made in early 2007 (at the time Borders and Amazon were marketing partners and Blockbuster had just posted its highest subscriber growth ever). In contrast, Blue Nile shot up like a rocket at first but then fell into a seven-year malaise.
The question is “why”? Why did Blue Nile fall short while cousins Amazon and Netflix hit the mark? All three used innovative Internet-based business models to undercut long-established retail incumbents with dominant brands. We wondered if the difference was caused by industry-specific details, the recession, mis-management, intellectual property, access to raw materials or even geopolitics. These theories had some merit, but none of them translated into refinements that improved the accuracy of our predictive models while accounting for Blue Nile. The mystery lived on.
Afterwards we sort of forgot about Blue Nile for a few years. It sat in our big pile of anomalies along with other mysteries there aren’t enough hours in the day to solve. Still, in the back of my mind Blue Nile continued to nag me. Then last year, with the advent of some new algorithms, I tossed Blue Nile’s data into the mix and – to my surprise – the mystery was solved.
While I can’t go into every detail, the main issue – at least according to our models – was price.
Price? How obvious (but not really).
Turns out, if you’re shopping for a big, fat, high quality diamond engagement ring, Blue Nile can be 20% – 35% less expensive than alternatives like Tiffany. Blue Nile can do this because, as an online store, it isn’t burdened by the high costs of retail overhead or in-store salespeople. That’s why early algorithms saw advantages for Blue Nile. But slicing the data a different way reveals Blue Nile’s average revenue per transaction is actually higher than Tiffany. A lot higher.
Around 90% of Blue Nile’s sales have a diamond in them (roughly 70% are diamond rings). In contrast, 45% of Tiffany’s 2013 US sales were in the “Fashion” category, generally consisting of non-gemstone items (around 60% is sterling silver jewelry). In other words, people buy a ton of moderately-priced items at Tiffany whereas they mostly go to Blue Nile to save on more expensive diamond purchases. As a result, Tiffany’s average sale is lower than Blue Nile’s.
This isn’t new information for anyone who closely monitors the industry. Analysts have written about it before. However it wasn’t until our last round of algorithms that the full predictive significance of this specific data point came into focus. There are, after all, a lot of variables to consider. Yet by our calculations, this one variable had higher predictive value than just about anything else. Change it enough and you see an entirely different future for Blue Nile. Who knew?
By attacking the high end of the market right off the bat, Blue Nile ended up blunting its disruptive edge while boxing itself into a tight corner. A river of mystery, it seems, boiled down to a single drop.