Predict, Customize, Convert

As more sites identify and profile users coming through their front door, retailers and publishers gain the ability to optimize the site experience itself for various audiences. Magnify360, which works with Inuit, HSBC and Citrix, uses a combination of behavioral targeting, predictive modeling, and real-time multivariate testing to lift conversion rates. CEO Olivier Chaine walked us through a recent case involving the ContinentalWarranty extended warranty retailer, where the automated system produced some customizations that were as effective as they were counterintuitive.

Behavioral Insider: What problem is Magnify360's technology trying to solve for marketers?

Olivier Chaine:
The mission is to continuously deliver customized experiences to each and every person who comes into their Web site to raise usability and customer experience as well as improving their ROI. Today's system encourages batch mode testing, because there are so many disparate tools -- multivariate testing and lead scoring, etc. The effort to run tests is high and the impact is relatively low. It is a one size fits all Web site -- winner takes all tests that say this page is better than that.

BI: How does this work differently for the user coming in the front door?

As an individual visits the site, before the home or landing page is served, it analyzes against our behavioral data matrix this individual's profiles and tags them with a number of different profiles they may match. They could be visiting at night, from the East coast; we can also detect whether they are an early adopter, visual or analytical, spontaneous shoppers, etc.

BI: Where are you gleaning this data?

It is a predictive model that is monitoring traffic across the entire Internet and creating a predictive algorithm. It says, based on certain data points, this person is likely to be visual because [of] others we have seen who have similar characteristics.

BI: But how do you know the characteristic of the person coming in? What are the identifiers?

The address, browser type, time of day, geo location, keyword they typed in. We have a semantic parser that analyzes whether someone types in the word 'free toys' vs. 'cheap toys' vs. 'good quality toys.' We don't really care about the word 'toys.' We are interested in the adjective that allows us to glean the personality of this person. The goal is [knowing] why are they here now and how do they think, so we can deliver an experience that is matched to them. The optimization engine automatically figures out which pf all the available experiences that have been created by the marketer will have the highest net impact, conversion rate, shipping cart value, lifetime value of that customer.

BI: How many experience typically are created?

We recommend starting with an existing landing page and maybe creating two or three small variations on it. Then every week or so, come up with one to three more variations that target a new group or evolve something that has already been built. Some of our clients have several hundreds of different experiences. But it wasn't a big effort up front. It was an evolution.

The drag and drop interface on the back end enables  marketers to do that by themselves without having to get tech involved. You optimize different pages that are targeted towards specific populations and let the system figure out what will have the highest performance. We are giving each person what they are looking for and what is most likely to convert them.

BI: How did this work with the ContinentalWarranty site?  

They are the leading direct-to-consumer extended auto warranty company. They are a very sales-driven organization. They have two call centers with over 270 people. They need a lot of leads to drive their business on that scale. They had a number of initiatives and had been running PPC for three years, and the last year the conversion rate had plateaued around a very respectable 14%. But they were curious about how high they could make it go.

We took their existing landing page and made a couple of different variations -- not a big redesign  -- just creating small changes to it to target different audiences. And we let the system automate the optimization of it.

Within the first ten days, we had 42% lift and saw the impact immediately in the call center. The noise level went up. And the quality was maintained despite the higher volume. We evolved the creative and the targeting algorithm and the conversion rate kept climbing and reached about 90% lift after four months.

They have also expanded their marketing and gone from 20,000 keywords to almost 75,000 keywords they advertise on, which is bringing a much higher volume of traffic. They have much more control over the profitability of their paid search.

BI: How was the audience segmented -- and what were the corresponding variables in the site design?

For lead generation, they need to ensure customers have the right mileage and the right makes and models. We took the form fields and moved them around so sometime they asked for contact information on the second page and sometimes the fourth.  By stretching out forms with one question per page or forms that were more condensed, we were able to appeal to consumers that were more in the ADD profile or the spontaneous shopper as well as the more methodical shoppers. Changes just in the layout of the form worked for certain audiences and not for others.

By simply creating those variable experiences without changing the design massively, we saw that initial lift. One obvious targeting was, if someone puts in keywords that have a manufacturer brand name, you should put up a picture of that type of a car.  

BI: So you can work at the granularity of individual keywords that land a person on the site, and then follow them with specific content?

Absolutely. But the interesting thing about that kind of targeting is that normally a marketer would say that if the keyword says 'BMW,' serve up a picture of a BMW ad. Because we threw all of these variations into the optimization system and told the system to optimize as best as possible, we found that the behavior patterns dictated different response rates and different experiences that were counterintuitive.

And so for people who are shopping for extended auto warranties with 'BMW auto warranty' keywords during the weekends, at night, from the East Coast at any time of day, definitely the picture of the BMW works best. But from the West Coast, on Firefox during the day, the picture of the Ford Explorer works best.

I was doing a demo one day and the Ford Explorer picture popped up and I told my team something must be broken because it was counterintuitive. They looked at the numbers and found that it actually was performing better by 6%.

We have some hypotheses, including that [these] people were looking for something perceived as low cost -- Ford Explorer. Even though the person owns a BMW. they want to buy from someone who really caters to the Ford Explorer crowd. That is what we are hypothesizing. You wouldn't ordinarily test that as a marketer



Next story loading loading..