Adventures in Email Marketing

I love this part of data analysis: finding the hidden pockets in datasets. In a recent post by Denise Cox, “Measure what your subscribers *don’t* do”, she takes on three general areas:
- non clicked links
- non active customers (purchase history, behavioral)
- non responsive customers (clicks & opens)
- non-clicked opens

With segmentation, you can easily hone your list to find “the inactives”- the deadbeats, the non-responders, all of those fun names for our opted in folks who aren’t acting on our emails. There are campaigns to re-invigorate, but let’s focus more on how to find the non-activity, and what non-activity looks like.

The tricky bit in finding black holes is that you can’t describe a negative without first describing the positive, and saying “not that.” That’s the goal, usually, for finding these negative, non-existent behaviors. Too abstract? OK, so to find links that customers haven’t clicked on, you have links, and look at the clicks. For those that are not in the URL click set, they are unclicked. That’s a relatively easy query. The difficult area is to find the last purchase, and create a “has not purchased since” time frame, in buckets. Then move customers slowly into those buckets as they mature. That determines the half-life, so to speak, of the customer. When they purchase again, they go back to “1 day” or a bucket, such as “0-3 months since purchase”. Segmenting your list on customer half-life, as well as other behaviors, allows you to personally message according to their lack of interactions. As we all know personalized, targeted list perform very well.

I can hear some marketers saying that their RFM models serve this same purpose. My general issue with RFM is that it’s not transparent. Opening up a score to the weights and determinants means you can segment and target, with messaging, far more accurately.

The example above was with past purchase data. There’s a lot more data available to you.
- Unsubscription data: If your system is tracking the date and message of each unsubscription request, you can place these users into buckets based on age of unsub, as well as messaging unsub for future service message placements. It is also great analysis for the campaign strategies.
- Web behavior: If you are meshing Coremetrics or other web analysis products with your marketing system, you can determine product clickthroughs or other activity before, or after a certain email system. What email campaigns did not result in a lot of successful orders? Abandoned carts? Product clickthroughs?
- Images off: Did you get a lot of clicks, that were not opens? Is that metric, the anti-click (basically an images-off email) higher for certain layouts or campaigns?
- If your site is more than a retailer, you have vast amounts of data available to you based on user behavior, and all of that positive data has a negative, inactive side. Are those associated with email campaigns and orders from emails? If so, are there dropoffs compared to some campaigns?
- Click and open behavior based on campaigns. Some ESPs can handle this, and some of you have homegrown systems in house that allow you to filter and segment based on opens and clicks. But I’m focusing here on using campaign-specific opens and clicks. Did they open this Christmas campaign last year? Did they stop purchasing a year ago? Have they clicked less than 3 times on any given month? These are all great ways to whittle down a segment into a former-customer-but-lapsed-email-respondent category, for specific messaging.
- Opens that are not clicked. I love this metric, and Denise covers this in her post. Recently at a client we noticed trending where our opens were not resulting in clicks, far more significantly than in other year over year, or quarters. I pretty strongly think it’s subject lines that don’t fulfill the promise of the content, so users are interested, but let down, once they open the message. This can also be a result of poor alt-text describing images (that don’t fulfill promise). Others thought it was a result of other factors- offers that aren’t compelling or creative that isn’t working. Besides analysis and cause, you can act on this data. Selecting out these folks and running creative tests on a random sample could determine what essentially was the cause.