The fit of independent variables to the personalization scenario

October 1, 2009

Regression analysis can determine the “fit” of independent variables to a dependent variable.

Not all independent variables are a good fit.

Relying upon a small set of independent variables may produce an incorrect fit -> destroy your chances of doing anything that’s successful in your personalization program.

Thought for the day:-)

Mark

Advertisements

Behavior, algorithms, consumer relevance and the advertiser

February 19, 2009

Can a single algorithm deliver relevance… across seasons, different web properties, between global cultures and among differing offering categories? Personally, I have troubles trying to predict behavior in the people I’ve known for many years. The human heart and mind act in sometimes strangely unpredictable ways. Isn’t that the part of humanity that’s really great? I think so!

Algorithms to deliver relevance need to accurately reflect consumer information – behavior, demo / psycho-graphics, and other.  The difficulty with this model is that only the smallest of slivers of a consumers life revolves around any single brand – though we would all like to believe differently:-)  Also, until you get reams of data surrounding an individual, how do you actually recommend. In the on line world this is particularly debilitating because over half of website traffic bounces from good sites right away and only a small sliver (used the word twice) of traffic actually makes it to a conversion… and an even smaller sliver (gotta stop using that word) comes back and makes it to a second second conversion. So, algorithmic personalization or recommendations really are only capable of helping a small portion of your consumers, after you get to know them.

An actuary can build statistical models that deal with vast population samples, telling the breakdown of what will proportionally happen in certain events. That’s interesting but it also deals with averages across large groups of people. Not necessarily valid to the point of algorithmically driven recommendations or optimization of an individual and their purchase potential or drivers.

I don’t intend to close the door on the subject, I do believe these approaches are helpful when other data is not available or when you have A LOT of information and you want to solve a retention / lifetime value issue… which are both great issues to solve. From my experience though, many brands believe they can use these technologies, in particular recommendation engines, to help solve an acquisition problem. Hmmm.

Love to hear your thoughts!

Mark


Behavior is only one datapoint

September 29, 2008

Much has been written about on-site consumer behavior and its applicability to targeting programs. Despite this, little is really known about types of behaviors that are tracked or trackable, and which ones are truly indicators of or useful as a high-value targeting mechanism. It’s possible that we’ve over-inflated the importance of behavior and overlooked other things that should be considered. And, in doing this, headed towards a focus on a technology solution that relies upon the use of highly personal consumer information to fuel its’ engine… raising the concern of consumer protection groups and the governments of many countries.

Types of behaviors that are tracked or track-able:

  • On-site, single domain behavior – probably the least contentious source of data, easily believed by consumers as a valuable source of data to help marketers connect with consumers. Often called re-marketing, in a recent MediaPost article, Jeff Hirsch, the CEO of Revenue Science and Behavioral Targeting Standards Consortiumfounding member, specifically stated that this a limited version of behavioral targeting.
  • Cross-domain behavior – algorithms consume a large quantity of data, including searches, product views, and domains visited in the attempt to identify a sense of interest on the part of an individual consumer so a marketer can target ad messages. The aggregation of this data can occur via cookie tracking across a network or more recently there’s been attempts to validate the process of intercepting and interpreting all web activity from ISPs (i.e. NebuAd), creepy.

However, in all the dialog about behavior, what other data points are available that marketers can use as targeting methodology? Or, from a more important strategic perspective, what is the objective of a marketing program and which data points can help the marketer accurately identify and target?

My assertion is that other data points are more effective at targeting, achieving marketer objectives. Here’s why I say this:

  • Today, only a minor portion of display ad spending is funded towards behavioral targeting. Display ads are largely brand oriented. Marketers tend to choose domains as their primary targeting mechanism. Behavioral targeting comes too late in the purchase cycle to affect brand preference and attitude, it’s more of a “hail-mary” or a direct response tool.
  • Virtually no on-site targeting is the result of multi-domain data aggregation.

What other data points are available, without any fancy technology:

  • Geography, language, and season – these have been used for generations in the offline environment very successfully, why don’t we use them on line more often? The interesting point is that these data points are largely available to everyone with very limited technology constraints. Between a site visitors IP address and their browser settings, these data points are easy to use. Whether display ad or on-site targeting, these are available today.
  • Returning customer or new prospect – again, these are largely available to marketers today. Why not use them? Certainly, every marketer would like to speak to a new prospect in different terms than they would an existing customer.

In a later post I’ll try to talk about targeting options that require more sophisticated technology. Until then, send me your thoughts!

Mark


Personalization versus Targeting

March 10, 2008

website personalization, dynamic targeting

I’ve read articles and blog replies where others try to distinguish between the terms personalization and targeting, referring to personalization as one-to-one and targeting as one-to-many. As a person who has actually created terms and pioneered strategy in this marketplace, I see the two as more or less synonymous with far fewer distinguishing dimensions than others see.

While trying to create a communications strategy for Kefta’s multi-channel personalization solutions, we determined early on that we wanted to distance ourselves from the failed software based personalization solutions of the late ’90s. They were an expensive, IT driven failure. The key failure was that they were too “heavy” a solution to ever get off of the ground. From a marketing strategy perspective, they were a failure because they relied upon users to self express differences before they could start targeting content, and they simply served a different message, as if that was supposed to be better than the original message.

Targeting became a term that was descriptive yet avoided a reference to the prior perceived failures of what was called personalization in the late ’90s. Using this learning as a guidepost, we landed on “dynamic targeting”. Prior to our use of this term, little was used with reference to targeting and no one in the online marketing space had ever used the combined term “dynamic targeting”.

Today, few people use the term personalization as a description of a type of technology. It’s more frequently used as describing a type of experience, leveraging the term personal. Beyond that, personalization has been a dead term and trying to describe it beyond it’s prior history is futile.

Love to hear your thoughts –

 Mark


E-Commerce News: E-Commerce: Enthroning the E-Shopper

March 6, 2008

In a recent article on E-Commerce news, Enthroning the E-Shopper, Paul Korzeniowski brought up some great points about the value of personalization in the e-commerce world. Specifically, he identifies the youthful state of the approach, its’ ability to increase customer loyalty, and he details several personalization opportunities and functionality. He summarizes the situation very well… “Building brand loyalty has become a struggle for retailers; however, personalization has the potential to help them to enhance customer allegiance and differentiate their products in highly competitive markets. Though in an early stage of evolution at the moment, customized shopping experiences are expected to become more common as the e-commerce market’s ongoing maturation continues.”

Etailers have historically been slow adopters of new technology, preferring to watch for success at competitors before dipping their toe in the water. Examining the evolution of dialog in this industry, just recently simple testing scenarios were all the rage and today we’re hearing this great dialog about targeting consumer needs… this is a huge shift in thought and a wonderful thing to see! In some regards, this is probably an evolution from consumer generated content, more than a shift from testing to personalization, regardless I love it!

One area I would like to elaborate upon is the forms of personalization he details:  from targeting online behaviors to physical goods. The point I would like to make is that these really shouldn’t be viewed as separate efforts, rather, a continuation of a strategy to learn more about prospects and customers and start servicing their needs. That is, if you’re going to manage a program for customization of physical goods, say jeans, why would you not augment that with a program to reinforce those expressed needs when a user comes back to a homepage or receives an email?

Using this example, when a consumer designs a personalized pair of jeans, how much more successful would the campaign be if the ensuing communications on the website, email, and direct mail offered a coordinated communications effort focused on: showing matching accessories and additional fabrics that fit the same genre of jean, then later, as the seasons change, offer different styles and fabric weights that are appropriate for the temperature.

Creating additional sophistication in this campaign, consider the efficiency a retailer could generate if they were to track the responses of this consumer, as they respond to the different communications channels. Over time, one could determine more than just the relevant message, but also, the preferred methods of communication.

It’s the same customer. They have the same needs regardless of product or communications channel.

Yes, there is a lifespan associated with behavioral learning, relevance decays rather quickly. These types of campaigns are most successful if the retailer can generate a response in the first few weeks. However, the reality is that the rate of decay is very specific to environmental variables associated with each use-case. This decay should be a part of the measurement and analysis process so one can optimize their program over time.

Wrapping up my point, focusing on single slice of your relationship with your customers and producing a personalized communication is a great start but must be seen as the first step of a larger program. Learning from consumers that cross your path, storing that data in a central data database where future programs can consume and contribute visitor level data, is the single largest key that a marketer can build into their strategy to increase relevance.

Personalization campaigns must transcend individual contact points – web pages, emails, banners, and even physical goods.

Love to hear your thoughts –

Mark


%d bloggers like this: