One-to-one marketing strategies drove personalization's initial implementations. Using personalization's analytics
and rules, online enterprises could individually interact with thousands of people per second. To increase online sales, enterprises customized Web pages on the fly to recommend high-interest products based on the individual customer's stated preferences, as well as past browsing and buying behavior. Marketing fervor created the notion that personalization's value resided in treating every visitor as unique.
However, both implementation difficulties and the realization that customers are not always that different have dashed this initial dream of pervasive uniqueness. This Aberdeen InSight report talks through this market evolution and describes the high value role of personalization today -- that of concierge treatment in the digital world.
Initially, too much work
Enterprises that survived the dot.com craze learned an important lesson -- the task of one-to-one marketing via personalization is far more difficult than they imagined.
First, business analysts could not maintain the hundreds or even thousands of business rules necessary to keep up with changing customer preferences. Second, personalization technologies such as collaborative filtering -- designed to remove the hand-coding burden -- consumed significant system resources. Either way, personalization was too much work and enterprises abandoned this initial iteration.
Back to segmentation
Returning to the tried-and-true method of segmentation, enterprises made do with clustering users into like-minded groups. Segmentation allows corporations to acknowledge that all customers are not the same, without going to the expense of divining each customer's unique preferences.
As with personalization, segmentation can be done in two ways: via business rules that enforce meaningful groupings (e.g., customers aged 35 to 44 are a segment) or through data mining, which discerns patterns of behavior in large sets of data.
Each approach delivers significant benefits, but each method imposes different requirements on the enterprise. In contrast to data mining, business rules consume very little processing power. Therefore, they are excellent for quick classification of customers when time is of the essence. For example, this occurs when trolling through millions of customer records to decide which customer gets which catalog or customizing a Web page for a customer in real time. However, customer segments can shift over time as both product capabilities and customer tastes change. Accordingly, business rules may ossify, losing their value.
Data mining can offset the static quality of business rules, highlighting changes that businesses need to make. This explains why enterprises commonly enlist both technologies in tandem -- using data mining to discover a previously undiscovered segment and then defining that segment using a set of business rules.
Although segmentation is once again proving its value, enterprises should not forget the value of personalization. It does have value, as long as enterprises apply the technology sparingly and appropriately.
When to use segmentation or personalization depends on three factors:
- Frequency of the interaction between a company and a customer
- "Uniqueness" of a customer
- Value of that customer
High frequency of interaction
Life insurance companies, for example, rarely interact with their customers; e.g., the customer buys the policy, pays the bill, and the beneficiary receives the death benefit when the customer dies. This lack of interaction has two ramifications: (1) it makes it difficult for insurance companies to directly monitor their customers' changing needs and desires, and (2) there is little value to personalizing enterprise-customer dialogs because the interactions are so few. In this case, segmentation makes more sense -- creating products and collateral targeted at different customer segments, and then leaving it up to the insurance agent to personalize the interaction in a non-digital way.
High 'uniqueness' of a customer
Although a higher frequency of customer interaction may make it easier for a company to gather data for personalization, it does not necessarily follow that personalization is the ideal choice. Some customers may not be that different from others; in other words, these customers actually may belong in a segment. If 10,000 members out of a 50,000-member customer base are quite similar in attitude and buying behavior, applying personalization to each individual is a duplication of effort. Savvy enterprises are recognizing that, in some cases, segmentation is a cost-effective way to offer meaningful customization without going to the expense of "over-customizing."
High customer value
The driver for performing personalization is most often tied to the value of a customer; e.g., current value, lifetime value, or indirect value as an influencer. Since 20% of a company's customers are typically responsible for 80% of its revenue (Pareto's law), limiting personalization to that select group is a good way to leverage the value of personalization while limiting its expense.
Offering concierge treatment
In fact, personalization should be viewed as concierge treatment; i.e., the digital equivalent of a top-notch concierge service reserved for a hotel's most valuable guests. Enterprises that take the concierge route typically classify their customers into a variety of segments, and then apply personalization to the most valuable and dialog-intensive group.
Due to personalization's high value, it needs to be accurate as well as attentive to the customer's evolving needs. A poor recommendation can, at times, be worse than no recommendation at all.
Mutually refining profiles are key
These business requirements demand that enterprises pay close attention to the underlying technology architecture, as well as the resulting data. A suboptimal underpinning and poor data will lead to flawed recommendations.
From an architectural point of view, the system must be able to effectively profile both users and offers (be they content, stories, or products). By capturing attributes of both users and offers, the system can better understand how they interact.
This relationship is symbiotic -- an understanding of the users' vocabulary and viewpoint is required to organize the information correctly for that audience. In turn, what the content is "about" needs to be abstracted so it can be categorized. By monitoring which users read which offers, the system gains a deeper understanding of users' interests and preferences.
The trick here is to balance the sophistication of the two types of profiles. Most vendors favor one or the other, to the detriment of deeply understanding the customer-enterprise dialog. Intricate knowledge of customers -- based on customer-stated preferences, purchase history, demographics, and psychographics -- cannot be further refined if the content is poorly or simplistically categorized. At the same time, superbly categorized content does not help the enterprise further understand its customers if few customer attributes are tracked.
Armed with excellent and deep profiles, enterprises can decide whether segmentation or personalization is appropriate. On the one hand, if some of the profiles are similar, by definition they are a segment. On the other hand, if the high-value group of customers is peppered with different idiosyncrasies, the added effort of personalization makes sense.
In 2000, the personalization sector was sizzling -- one-to-one marketing was generating a lot of buzz, large crowds were attending the fourth Personalization Summit, and NetPerceptions had more than doubled its revenues year-over-year. Today, the sector is in the dumps and the term "personalization" is a dirty word.
It is appropriate that the market has deflated the dot.com-driven hyperbole. At the same time, enterprises should avoid incorrectly ignoring a valuable technology due to political correctness. As the Greeks noted two thousand years ago, "Moderation in all things" -- and that aphorism especially applies to personalization.
Enhancing the customer-enterprise dialog will only gain in importance as the world grapples with information overload. Making the right offer, to the right customer, at the right time will delight the customer that is inundated by noise, as well as increase enterprise sales. When targeted at unique, high-value customers, personalization can be a cost-effective way to align customer and enterprise desires.
Guy Creese is a research director for Internet analytics at Aberdeen Group, a Boston-based IT market research and consulting firm.
To read more articles like this one, visit Aberdeen Group and register for free online access to Aberdeen's IT market research and analysis, and to subscribe to technology-specific e-newsletters.
© 2003 Aberdeen Group Inc.