• by David Baker , Op-Ed Contributor, September 22, 2014

“Those who think they know more than others annoy those of us who do.” That’s a favorite quote on a T-shirt I’ve had for over a decade, that I roll out around holiday season.  As we gear up for the holidays, using restraint when pressed for time is likely the biggest challenge a marketer will face.

There is a fine line between marketing brilliance and having to walk into your boss’s office and explain a poor customer digital experience that went viral.  Sometimes the problem could be a personalization attempt that went awry.

The concept of contextually relevant brand messages is the basis for how the email industry grew up, from mail merge to dynamic personalization and recommendation engines. Over the last 15 years, I’ve seen personalization deliver amazing lift, and I’d confidently say it’s a best practice for most.  From database-driven personalization of consumer information to collaborative and content-based filtering, the options are virtually endless to deliver that “perfect” message or experience.

But does it always work? That’s tough question to answer directly, as it would be blasphemy to suggest that attempts to use customer data to personalize an experience don’t always help.

But in fact, let’s start with the dark side of when it doesn’t pay off:

-- When your have more customer segments than you have customers.

--When {FNAME} is used in a subject line with the terms “Cheap,”  “Below Market” or “Reduced.”

-- When the account balance is 0.

-- When your daughter gets email promotions for diapers.

Personalization pays off when:

-- The brand’s email and site-side personalization are in sync.

-- Your customer loyalty program drives over half your revenue.

-- You know the name Alex is both male and female.

-- You have more creative production people than email marketing managers.

Making customer recommendations is an entirely different story, since they require some logic based on past behavior.  Doing this in email can be risky and time consuming. Remember the first line in this column!

A few things to think about when it comes to recommendations:

You are not Amazon, Sears, Pandora , Google or Facebook.   You will never have the breadth of personal, performance, or preference data as these leaders, or the real-time requirement to deliver against it.   Real time should mean “near” real time to you.   The more programmatic you become,  the larger the propensity to make mistakes or deliver non-relevant “default” experiences to the 30% of your database you don’t have good data on.  You should think hard about the trade-offs of performance to errors, and what is the risk tolerance your organization can live with.

Don’t get me wrong. If you have the data and the type of brand that prides itself on a connected experience, use it to the best of your ability and try to bend lasers; just don’t try to out smart yourself.   A rule of thumb when in this mode:  If you can’t QA it at the same pace as you create it, you likely shouldn’t take it on a long trip or try to break the speed limit.