When you start linking user behaviors across channels, you stop speculating about how your marketing works and actually start seeing it. Even for veteran email services company iPost, in the
business for a decade, its own new Autotarget product rendered some surprises. Chief strategy officer Steve Webster explains how the next stage of email metrics involve combining email response with
other data streams like purchase behavior to create predictive models that clients can use to shape their promotional strategies. Webster says the new data shows that the effect of an email message is
more subtle and long lasting than we suppose. Behavioral Insider: What behavioral data is being gathered with what sort of technology here?
the actual technology was created by us but some techniques are pretty well known. We are continually collecting the email click and view behavior for every mailing for every customer that goes out.
We are also collecting every purchase that takes place in near real time. We see the nightly data feeds of purchases: not just purchases as a direct result of email, but all purchases in all channels.
The lower level of data we can get from them, then the higher the quality of the predictive analysis. That leads to the first level of analysis, which is RFM (Recency, Frequency, Monetary Analytics).
Every night we run a full year's worth of trailing history through an RFM analysis engine for every customer and take into consideration every email view and click as well as every purchase event
that has taken place. For the next version we will move beyond predicting the likelihood of purchase based on past email interactions and purchase behavior and move into what kind of product did they
purchase based on what products they looked at or interacted with in the email or the Web site.Behavioral Insider: Can those be very different?
one of the first and surprising things we saw. There is the presumption that when someone receives an email message they then click on the email go to the Web site and either make a purchase or not
and then they are done interacting with your email.
This turned out to be wrong. We discovered very quickly that the power of an email impression lasts for weeks after the customer has
actually received the message. The particular interaction they will have with you later really depends more on their personal preferences than on your putting a new email in front of them. The email
works like a classic advertisement, like a billboard going by. You have some of their attention but the mere fact that your brand name showed up on the FROM line has a very measurable impact on the
likelihood of their future purchase. And that was a really big surprise to us. We thought the quality of the HTML, the head shots and the creative made all the difference. It turns out that it does --
but not nearly as much as the fact that [the email] made an impression on a customer who actually was interested in receiving an email from you.Behavioral Insider: How do you
manage the classic targeting issue of scale, of parsing a customer base so finely that the segments are not worth pursuing?
The daily analysis slots every customer into
one of 125 different buckets; each RFM cell represents a customer who has a particular recency of purchase, frequency of purchase in the last year, and monetary value of their purchases. But then
another layer of segmentation we perform is to identify the customer interaction with a particular email message that was sent. So what we let the marketer do is target all the people who received
this particular email or group of emails and either did or didn't interact with them -- or all the people who are like the people who got that email and interacted with it but who didn't
get that email but might react well if they did. So it lets you do test mailings of small groups and see how well they do and then do follow-on mailings to the larger group of people in your universe
who share the same RFM cells.Behavioral Insider: What kind of effectiveness metrics are you seeing?
In email you are competing for their attention.
With behavioral analytics you want to be sending email to people at the right frequency of interest. You can bring up the RFM response graph for a set of mailings and grab that slider and drag it to
zero, and you just selected all the customers who have not interacted with the email at all. So you can quickly grab all the people who in the recent past are totally dead on your list, then figure
out what is it about your brand you have not presented to them. Then we will try to get them engaged -- perhaps with one deeply discounted item. You want to get them to buy something. Instead of
working with your intuition of who you think your customer's natural segmentation is, you see from their actual behavior what the proper segmentation is to use to efficiently communicate. Instead
of a bazillion tiny segments, it lets you divide the customer base into to the right 4 or 5 segments you can then manage to market to through your mail channel.Behavioral Insider:
What are the next set of challenges for email marketing that behavioral analysis addresses?
For years in email we worked in this environment where the returns are so good
that the pressure to improve has just not been there the way it has been in the catalog channel for decades. In email you may spend a tenth of a cent to send an email but a single purchase generates
enormous returns. It is typical for customers to get $23 of gross revenue for every dollar they spend. The pressure comes in as more companies and brands get savvy about competing for customers'
attention. They need tools that let them put relevant information in front of the customer. Behavioral Insider: Where is this kind of product and user profiling headed?
There are all kinds of group lensing and merchandise corpus technologies where you build up databases of how products are like each other; you build that both from the
behavior of your own customers and the information you get from the manufacturers and you may also start to get from data companies. It is entirely possible that product interaction behaviors, how
products are like other products, and how views on one product drive the purchase of other the product itself, could become a chunk of data companies may sell to each other.