Mass Media drive online participation
#1056 Created Updated 12/09/2019
Dennis R. DuBe', Pubsource
Apple Electronic Media Lab
In Conjunction with OneNet Colorado
Boulder, April 14, 1995
Consumers behave in statistically predictable patterns while consuming online media. Online activities can be broken into communications and consumptive clusters, and their relative strengths assessed. Consumer behavior patterns provide a basis for modeling online business behavior.
A unique research opportunity occurred when the Apple Electronic Media Lab in Boulder, Colorado was involved with two successful on-line services, one local and one national.
A project of Apple Online Services, the AEML was tasked with studying business models of online services at a time when AOS was preparing to launch eWorld, Apple's international Macintosh-based online service. At the same time, the lab in Boulder was also hosting the national operations of the OneNet Member Network, an affiliation of nearly 500 local Macintosh-based bulletin board systems (BBS) in North and South America, Europe, and Asia.
Consequently, we were able to measure and model the simultaneous behavior of online consumers in two very different environments, which used two opposite approaches to online services: one a main-brain, centrally-organized super service, featuring professionally published content, and a distributed-server local BBS featuring only user-posted content.
The primary research tool was the OneNet National server in Boulder. It served as the hub for a dozen regional servers, which in turn supported district and local servers interconnected into a network of 500 sites. The national-regional hub structure allowed content and mail to flow up and down the tree, providing a virtual conferencing system that replicated all postings to local conferences across the entire net every day.
Performance results from the research server were compared with results from similar systems around the OneNet Network, including the Magic system in Toronto, Ontario, and OneNet Los Altos in Los Altos, California. Performance results were also compared with simultaneous-time use on the eWorld server in Napa, California, although the two environments used fundamentally different metrics for behavior measurement.
The primary researcher was Dennis DuBe', manager of the Apple Electronic Media Lab in Boulder. Analysis software was created by Alan Hoeltje and Dennis DuBe'.
This study was designed to detect predictable behavior patterns among the usage characteristics of online users, as well as identify behavior relationships among various measurable aspects of the behavior of on-line users. At the heart of the study was the desire to connect user behavior patterns with common assumptions about the nature of on-line business, and to create numeric models of on-line services based on real-world measurements. This study sought to measure the performance of models of on-line businesses in stand-alone and distributed configurations.
The Initial User Behavior Study, released 10/26/93, presented preliminary observations about on-line behavior, and summarized some of the questions included in the original study parameters. An interim report in January of 1994 extended and amplified those observations. This report covers that same ground again, with conclusions, and makes a first attempt at constructing the algebra of prediction for on-line services.
Definition of the Business
At the time of this study, there were two basic types of consumer-oriented on-line services in North America, discerned by business models which were driven by communications costs. The national services, exemplified by CompuServe, America Online, eWorld, Prodigy, AppleLink, et al, were high-profile in nature, and oriented toward fulfilling a business model based on network communications costs. With few exceptions, the national services use a variant of the connect-time business model, wherein the service attempts to provide users with a variety of activities, resources, services, and products to encourage increased connect time, which is the basis of billing. In addition, through the power and sophistication of the platforms on which national services are hosted, these services are capable of offering a number of activity-based opportunities to increase revenue from users, i.e. specific charges for downloaded products, on-line fulfillment (such as airline tickets or flowers), connectivity to a variety of outsourced data services, and other activity-based products.
The local service, by contrast, is governed by subscription basis, in which users acquire and use the service for a set calendar period, with connect time limits set for purposes of regulating system resources, rather than controlling network costs. Local bulletin board systems are noted for being low-cost or free, informal, jammed with download software and entertainment conferences, and often featuring the capability to connect with — and communicate with — other users within geographic areas.
The business model for national on-line services is well developed and mature, and undergoing intense refinement and scrutiny within many of the companies competing in this arena.
The business model for local on-line services, on the other hand, is just beginning to emerge from a large array of subsidized or free services.
In addition, activity patterns on the Internet, specifically those associated with Usenet Newsgroups, suggests that similar behaviors exist in this space. Unfortunately, the anonymity of Internet users inhibits good statistical modeling of behavior.
The study supported an attempt to model all major emergent online business models in interactive spreadsheets. The online business is composed of 1.) an access business and 2.) a content business. Currently, all online functionality (navigation, email, graphic display, content organization) is enabled by the access provider's software, while the content is mostly divorced from the revenue streams1.
We were specifically interested in behavior attributes that would contribute to an understanding of subscriber group behavior. A well-understood dynamic in other subscriber-based cable and magazine systems, the online subscriber group behavior's characteristics is the driving factor for all subsequent online revenue.
The study looked for subscriber attraction effects based on the size of the content offering, the size of the existing subscriber base, and the responsiveness of the model to positive marketing.
The study sought the behavior characteristics of subscribers, including frequency and duration of connect, menu and command choices while connected, communications activity, and system usge.
The study also sought to distinguish between "active" and "average" users, understanding the behavior patterns for each identifiable group of subscribers, and any other factor which could be used to predict system performance.
"Word of Mouth" Effect
There was anecdotal evidence that a "word of mouth" marketing effect among online services, possibly due to the exploring behavior of consumers. System operators had reported that new users appeared on their systems, even when they didn't advertise or market. Sysops attributed the subscriber growth to word-of-mouth from current users to friends and associates.
We confirmed this experience with OneNet Boulder, a 20-line First Class system which we switched on in February of 1993, using OneNet content copied from the Los Altos, California system. The content consisted of about 415 "conferences" (discussion boards), which were shared among most of the OneNet Network's 300+ member systems, and which were partially filled with postings. To this core content we added a very thin veneer of new conferences with the word "Boulder" in the titles.
New users were allowed to auto-register, and we set new accounts to expire if they went unused for consecutive 90 days.
No public announcement for OneNet Boulder was made, no telephone listing was secured, and there was no advertising effort. Instead, staff at AEML merely told their friends. After two months of operation, one announcement was made at a local Macintosh user group meeting.
From the beginning, the system maintained a steady effective net daily auto-registration rate. At the end of the 213-day initial period, the system had acquired and retained 818 total users, for a net rate of 3.83 new users per day. We discarded the first six months of operational data, as the system had experienced several interruptions of service, changes in structure, and other instabilities associated with startup and research activities. We began rigorous measurement of both daily registration and daily auto-deletes on August 1, 1993.
|Period||Auto Regis||Auto Deletes||New per Day||Delete per Day||Net New per Day||End of Period|
|Days 1 to 212||3.83||818|
|Days 273 to 453||1,749||1,418||7.8||4.7||3.31||1,266|
During a subsequent six month test period, the system maintained a similar effective registration rate, with total registered users reaching 1,266 on December 31, 1993. Daily net effective registration hovered around 3.1. Total daily registrations averaged 7.8, and daily auto deletes at about 4.7, with a extreme high of 79, and normal limit of 19, and a low of 0. For the period, a total of 1,749 users registered, and a total of 1,418 user had their accounts automatically expire after the default 90 days.
A survey conducted of OneNet subscribers by a graduate students in the University of Colorado School of Journalism and Mass Communications Integrated Marketing Communications Program confirms the word-of-mouth hypothesis. The telephone survey of 105 randomly selected OneNet subscribers showed that 68% heard about the system from a friend, with only seven percent finding out about the service through a commercial channel.
How did you find out about OneNet Boulder? %
Other BBS listing 18%
Digital Matrix (store) 7%
CU Journalism School 7%
Event at Apple Lab 6%
How quickly things change. Three years later, advertising on the Internet and in online services topped $300 million, and is projected to top a billion dollars by the year 2000 (Jupiter Communications, 1997).
Detecting the `word-of-mouth' effect is impossible for a system that is engaging in significant marketing and promotion. The OneNet Boulder study was significant because of the absence of marketing or advertising.
We were unable to measure that effect on other OneNet systems to which we had access, as they were engaged in various types of marketing.
The word-of-mouth effect for new subscribers is evident in eWorld's "Net New Weekly Subs." A surge following the promotion of eworld's launch settled to a steady rate by week 20, after marketing all but stopped. The gentle wave form in the rate was stimulated by a small amount of monthly magazine advertising.
A similar effect was observed with eWorld. eWorld received a marketing boost right after it launched, and eWorld distributed more than 20,000 sign-up disks at the spring MacWorld trade show in San Francisco, resulting in a surge of new users in the first 18 weeks of eWorld's existence. But then officials at eWorld adopted the curious approach of not significantly marketing the service, in the face of a veritable blizzard of sign-up disks being mailed by America Online, Prodigy and CompuServe, their chosen competition.
After the initial surge of registrations from the launch and MacWorld disks, there was no significant marketing program, beyond a limited and short advertising program in two Macintosh-market magazines. In the face of the massing marketing onslaught from the big three online services, the eWorld effort was insignificant. The results of those marketing efforts were identifiable in the eWorld daily and weekly metric.
The flow of new users onto eWorld settled back to a comfortable, steady rate. The "word of mouth" effect was spreading the word of this unique Macintosh-centric online service. No significant marketing, but a steady, low rate of new subscribers.
Unfortunately, there was also a steady rate of decay in the subscriber base. Just as the initial launch and MacWorld promo caused a surge in registrations, a few months down the road it stimulated a surge of churn. The churn caught and surpassed the new registration rate, and eWorld began to falter.
Subscription-based services are annuity businesses, in that the cost of acquiring the initial sale (subscription) is offset by the subscriber maintaining a subscription for a minimum time. Subscription length beyond the minimum represents a system's ability to both grow and be profitable.
The concept of an "average subscription length" is a convenient mathematical artifact useful in building spreadsheets, and in solving algebraic expressions. But in reality, consumers of on-line services probably have multiple "average" behavior sets.
We do not derived the "average" length of subscription, but instead predict the rate at which a "subscriber group" depreciates. In magazine parlance, a subscriber group is commonly all the subscriptions received in one month as the result of a single promotional effort, such as a mailing campaign. Thus, all subscribers gained from one particular effort are measured together, and their performance reflects the effectiveness of the marketing technique. Their individual "churn" rates are muddied by subscription renewals, extended subscriptions, and changing of addresses.
Subscriber groups depreciate sharply at first, and then more gradually over time. The subscriber group may dwindle, but it should never hit zero, as the members of the group expire, cancel, or renew their subscriptions. The "churn" rate for each subscriber group is different for each successive month _ sharp at first, and then flattening out as the early quitters leave only the loyal subscribers behind.
The "churn" rate for a publication's entire subscriber body, consequently, is the average of the churn rates of all the individual "expire groups." Expire groups can have dramatically different churn rates, often a reflection of the quality of the marketing effort or the "value" of the subscription deal. Bargain subscriptions, such as those obtained through Publisher's Clearing House promotions, have notoriously high churn rates.
Since OneNet Boulder had only one significant source of new subscribers _ word of mouth _ the entire subscriber body could be efficiently modeled as a single subscriber group, month by month.
The test period began on August 1, 1993 with users' inactive account "auto expire" feature set to 90 days. Auto deletes during the first 90 days were all for users who registered on the system prior to the start of the test. The final 60 days of the test period were used to measure the rate of auto deletes for all users, some of whom registered after the beginning of the test. The decay rates for both periods were statistically identical when adjusted for extremes, and were a direct expression of elapsed time. The smoothed (Box-Jenkins) rate of decay in the second interval was 4.7 users per day (plus or minus 10%).
There was no apparent change in rate related to size of user base; the total subscription base grew by 50% during the test period, with no measurable change in decay rate. As a consequence, we concluded that the rate of decay was not tied to the size of the user base, but to the rate of word-of-mouth acquisition. In other words, the system was losing subscribers at a rate that was being driven by the rate at which new subscribers came on the system.4
OneNet Boulder Auto Expire Adjustments
|Expire Length||Total users||Expired Users||%|
There was also no apparent change in rate due to a change in the graphic metaphor of the client software. Midway through the test period, and somewhat without warning, OneNet Boulder implemented a new content structure, and a new interface metaphor, over-layed on the existing content. By the end of the test period, more than half of all users had switched to the new publication-based metaphor, with no measurable change in any user behavior performance indicators, or in the rate of registration or decay.
Unfortunately, we are not able to determine if the change in metaphor and structure had a negative or positive effect on the decay rate. One possible interpretation suggests that the rate stayed the same as a result of our changing the metaphor and structure.
An additional test of the hypothesis, however, was provided in January, when we twice adjusted the auto delete period, first to 60 days, and then to 30 days, for the purpose of reducing stress on a failing server. As a result, we were able to get a snapshot measure of the number of users who had not used the system in 90, 60, and 30 days, providing a graphable picture of the decay curve..
The results showed a smooth distribution of expires over the three intervals. The average rate of decay at all three intervals was 4.7 per day, or a cumulative 142.4 per month, as it was for the entire subscriber base over a nine month period.
This is a normal distribution for subscriber group expire rates for the consumer magazine business, and is a similar curve (although not as steep) for the subscription newspaper business.