How Advanced Segmentation Can Improve Targeting and Metrics
Jeff Teitelbaum, e-Business, Discover
Jared Chase, Client Services Lead, e-Dialog
- Engagement Pyramid (relevancy and specificity)
- * One single customer
- * Segmentation and modeling (Web Analytics, CRM)
- * Demographics
- * General/mass population
- "We are not order takers, we provide insight.".....
- Continuous Improvement Model
- Business Rules inform modeling.
- Modeling influences content.
- Content determines Execution Testing.
- Execution Testing yields Results/Analysis.
- Results/Analysis shapes Business Rules.
- Most responsive will have highest opt-out if you keep targeting the, ("They respond the most - send them more than the ones who are less likely to respond!")
- Discover does "resting" periods to give people a break.
- Customers don't differentiate between marketing and transactional - it's just another email from you amongst all the other emails in their inbox.
- Discover rarely sends an email without testing to see if anything caused a spike in unsub. (survey - weekly?)
- Test results observed by Discover over time:
- Subject line personalization = 5% increase; link and location = 2%; creative = 2%
- Over a 3-year period, Discover is trending up for opens and clicks and trending down on unsubs over a 3-year period. How? LOTS OF TESTING.
- Was asked: "Can we send more?"
- To determine, used a 3-way test over 2-month period:
- (1) Business as usual - no more than one a week, no message if there wasn't a relevant offer for that subscriber
- (2) Hold-out (no email for 2 months)
- (3) One-a-week marketing email - if none identified, force one
- (4) Twice-a-week (aimed for Mon/Thu) - same deal - if computer didn't identify a relevant send, force one
- * Increase frequency did significantly increase unsubscribe rate
- * But no statistical significant change in opens.
- * This validated that the email team's original approach (control) was the right healthy approach for their audience (goal of 1 a week, but only if there was a relevant offer).
- Email is not an impression (like an ad banner). The subscriber should be expected to do something (and is expecting to be asked to do something) if they are engaged enough to open and read/skim your email.
- If there's no clear call-to-action, don't send it.
- Ask about email on every call, even if it looks like they have a valid email address on file.
- Unsubscribe by campaign is wrong. You can learn if one particular email caused a surge, but you also need to look at the larger ecosystem - it may be a combination of factors, or it may just simply be when a particular subscriber has had enough.
- View the slides