Online Personalization 2.0: This Time It's Personal
John Black

After the flame-out of early business models focused on online
personalization – for instance Firefly – it was easy to dismiss
this as another over-hyped casualty of the dot-com boom. But
there was nothing fundamentally wrong with the technology or the
concept. It was primarily ahead of its time. Now personalization
is an integral part of leading e-commerce sites such as iTunes,
Amazon and NetFlix, and a contributing factor to their success.
However, beyond this top tier, use of personalization is not yet
widespread online.

All of the macro trends - expanded product selection,
consumer-generated content and information overload - suggest
that personalization is poised to come back in a big way. What's
needed for mass adoption is a new business model rather than new
technology.


Trends

First, let's look at the trends that build the needs and the
opportunities for personalization

Product selection. The "Long Tail" phenomenon was first coined by
Wired in 2004 to describe how, removed from the constraints of
the physical world, the economics of retailing and the behavior
of consumers have changed radically. Where a traditional retail
store could only dedicate shelf space to high volume products, an
e-commerce site can stock literally millions of products.

Consumer-generated content: Partly in response to distrust of
marketers and professional critics, partly in response to the
ease of personal publishing/blogging, consumers are posting their
views, profiles and opinions online en masse. An estimated 33
million Americans have rated or reviewed products online. Social
networks like MySpace and Facebook have become a cultural
phenomenon. The combined voice of consumers is a powerful force.
Study after study shows that word of mouth by far carries the
most influence on purchase decisions.

Information Overload: Unfortunately, consumer-generated content
is often lacking in relevance. A consumer reading conflicting
reviews of the same product is often left asking: what do people
like me think of this? Was the one-star book review from an
English professor, or from a high school dropout? Further, it is
often difficult to separate objective feedback from
self-promotion.

While online retailing offers consumers unlimited choice, this
choice can be paralyzing. While word of mouth often provides
objective peer opinions, just as often it creates more confusion
without any filter on relevance.


Existing Solutions

New online services and technologies are starting to emerge to
solve these problems. The site Trendwatching.com has coined the
term "Twinsumer" to describe matching consumers up with "their
taste twins; fellow consumers somewhere in the world who think,
react, enjoy and consume the way they do." These solutions
address real and growing consumer concerns:

* Tell me what's right for me
* Help me explore beyond the mainstream, or in the words of
Wired, push people down the long tail

Personalized recommendations are typically driven by statistics,
in the form of "collaborative filtering", or by the user's own
network of contact. In collaborative filtering, "like users" (or
"like items") are matched based on their statistical similarity.
So it Bob and James liked 10 of the same books, the 11th book
that James rated 5-stars would be recommended to Bob. Or if
customers who buy the Godfather Part 1 also buy the Godfather
Part 2,... well you get the idea.

In the social network approach, recommendations are driven by
your friends, or by people you have chosen to bring into your
online circle of trust. This operates more like traditional word
of mouth, but on a much larger scale.

These personalization solutions tend to be tied to either
e-commerce or affiliate marketing business models:

* e-commerce merchants: iTunes, Amazon, NetFlix, eMusic
* online communities: listal.com, nextfavorite.com,
librarything.com. ratingzone.com
* music applications: Pandora.com, last.fm, MusicIP. Yahoo
Music

In most cases, personalized recommendations have focused on
product categories with a) broad selection and b) subjective
tastes. Hence, books, music and movies.


Challenges

With all of the promise of personalization to increase sales and
improve customer loyalty, you'd think its use would be more
widespread. However, every personalization application faces the
dual, and opposing, challenges of critical mass and data quality.
The best recommender technology is worthless without enough data
to populate the recommendations. In categories with a broad
selection, such as books, recommendations are not very effective
beyond the most mainstream titles until the number of
ratings/purchases reach the hundreds of thousands.

So how to get hundreds of thousands of data points from customers
before you can offer effective recommendations? Most e-commerce
sites use observed customer behavior – clicks, searches, carted
items and purchases – to infer product feedback. While this is
the quicker and easier path to critical mass, it sacrifices data
quality. Just because a user clicked on or even bought an item
does not mean they liked it. Often the customer purchased a gift,
did not enjoy the product, or had a one-off need for the product.
I suspect other people have a similar mish-mash of
recommendations at Amazon as I do: from gardening tools to
lullaby CDs to Accounting books.

These data challenges – not technology limitations - have kept
personalized product recommendations out of all but the very
largest, most sophisticated e-commerce sites. And let's not
forgot about traditional brick and mortar retailing, which still
accounts for 90%+ of book and music sales. When was the last time
you got "personalized" service at a big box retailer or chain
store?


A New Approach

There's no good reason why every retailer shouldn't be able to
implement personalization as well or better than Amazon or
iTunes. At least in books, music, movies, video games and
probably consumer electronics and travel. In this new world of
ASPs, Web 2.0, APIs and web-services, the technical barriers have
been all but removed.

Which leaves the data. A new business model that can successfully
aggregate anonymous customer data and product reviews across
multiple retailers could be far larger, and more predictive, than
any database within a single merchant. And literally any
retailer, down to a single-store independent bookseller, could
tap into the benefits by also contributing to this uber-database.
If this sounds farfetched, note that Abacus Direct grew a similar
cooperative database model into a $100 million business in the
offline catalog market.

The benefits are clear for those sites who have successfully
implemented personalized product recommendations: dramatic
improvements in sales, conversion rates and customer loyalty.





---------------------------------------------------------------------
John Black has a long experience with personalization and
predictive modeling. John was the product manager for the first
one-to-one online banner ad targeting product at DoubleClick,
and managed market research and new product development at
Abacus, the leading predictive modeling company in the catalog
market. John is currently the founder and CEO of NextFavorite.com
(http://www.nextfavorite.com), a personalization service provider.