Peter Leppik over at Vocal Laboratories, the customer service survey firm, had a blog post on survey questions that caught my eye recently.

He relates a story of the pseudonymously-named BigBank, which had the doomed idea to attempt lead generation via live second party post-call surveys.

The idea was that if a bank customer answered "yes" to the question "would you recommend your bank to a friend," the follow-up question was something to the effect of "okay, what's your friend's name and telephone number?" Bad idea. Any regular person can see how this might lead to trouble, but the trouble didn't come from the bank's customers.

"As soon as this question was added to the survey," Leppik writes, "BigBank's 'net promoter' score (the metric generated from the 'recommendation' question) plummeted." Now would could that be?

Leppik:

After some months of investigation, BigBank discovered that the interviewers found the follow-up question unbelievably awkward--and I don't blame them. So awkward, in fact, that they absolutely did not want to ask the question. But they had to ask the question, and if a customer answered 'yes' to the recommendation question, the interviewers would get in trouble if they didn't put something down for a new sales lead.

So when a customer said 'yes' to the recommendation question, the interviewer would simply enter 'no' in the form for the survey--intentionally miscoding the answer to avoid having to ask the awkward follow-up. Since the interviews weren't recorded, the odds of getting caught were small.

Leppik's right, of course, but the biggest lesson I extract from this example is that call recording (and by extension, data mining) of all kinds can save a center months in decyphering service problems. If your survey team asks a terrible question, you'll know early on. And this extends to all kinds of call center interactions; even small centers can afford to record every call now. They ought to.