Tucked inconspicuously into his 2002 letter to shareholders, Jeff Bezos made a comment that should be instructive to all of us:

And our most sensitive measure of customer satisfaction, contacts per order, saw a 13 percent improvement.

Separately, in the book Invent and Wander: The Collected Writings of Jeff Bezos, Walter Isaacson tells the story of how Bezos pursued his goal of becoming the Everything Store:

His next steps were to branch out to music and videos. Keeping his focus on the customer, he emailed one thousand of them to see what else they would like to be able to buy. The answers helped him understand better the concept of "the long tail," which means being able to offer items that are not everyday bestsellers and, thus, don't command shelf space at most retailers.

The responses from his customers became a guiding light for Bezos:

'I wish you sold windshield wiper blades because I really need windshield wiper blades.' And I thought to myself we can sell anything this way, and then we launched electronics and toys and many other categories over time.

There are two reasons why these two, seemingly disconnected nuggets are so interesting.

First, there was no mention of an NPS or CSAT score when it came to measuring customer satisfaction or feedback at Amazon.

While I'm sure Amazon uses quantitative customer feedback scores within the bowels of its analytics, it's not really how they measure customer experience.

Instead, the emphasis is on an operational proxy for customer experience.

Contacts per order. The fewer touches, the lower the customer effort, and the better the experience (again validating what we've learned from Matt Dixon).

Rather than a 1-3% survey response rate, an operational proxy measure like contacts per order, is measurable across every order and every customer 100% of the time.

In the B2C world, this is often called "O-data" (with the O being operational).

Second, when Bezos needed feedback, he emailed a thousand customers. In it, he asked a very simple and specific question, what did they want to buy more of?

Rather than dumbing down the responses into a meaningless score, he and his team read them. Those responses revealed that Amazon could sell almost anything on its "everything store."

The rest is history.

Sentiment scores don't tell you much about customers, especially in B2B, where sample sizes and response rates are often low.

Operational data does.

Qualitative feedback does.

How are you listening to your customers?

Operationally and qualitatively?