Executive Summary:

The concept of one-to-one marketing is intuitively appealing, but there is little research that investigates the value of individual-level marketing relative to segment-level or mass marketing. In this paper, the authors investigate the financial benefits of and computational challenges involved in one-to-one marketing. They investigate the impact of customizing promotions on the two most important consumer decisions: the decision to buy from the store and expenditure level. The modeling approach accounts for two sources of consumers’ responsiveness to various marketing mix elements: cross-sectional differences across consumers and temporal differences within consumers based on the purchase cycle.

A series of policy simulations show that for an online retail business, customizing promotions leads to a significant increase in profits relative to current practice of uniform promotions to all customers.

Specifically, they find for this online retail business:

1. Customizing offers based on purchase cycle (Recency or weeks since last purchase) contributes more to profitability than exploiting variations across consumers using previous transactional content (segmenting by purchase category, basket size, demographics, etc.). This is important because the computational burden of implementing the dynamic optimization to account for variations across consumers is far greater than accounting for purchase cycle.
2. A substantial number of customers purchase without a promotion of any kind. Offering any promotion to these customers substantially reduces the profitability of a campaign, and targeting by purchase cycle is key to avoiding this problem.
3. Free shipping tends to be the most profitable promotion for re-acquiring lapsed customers, whereas discounts are the most effective tool for managing active customers. Offering the “wrong” promotion (e.g. free shipping to active customers) substantially reduces the profitability of a campaign.
4. Customizing offers by previous transactional content in addition to purchase cycle increases profitability further, with customizing at the individual level outperforming customizing at the segment. However, gains in profit using individual level targeting when accounting for costs might not exceed the gains relative to cost by segment targeting; outcomes need to be tested.

Review:

This is an incredibly rich study and I highly recommend a personal review for WAA members involved with online commerce. There is a ton of detail on how the different promotions affect response and order size, in addition to how these parameters interact with purchase cycle to variously contribute to profit.

For those not used to discussing purchase cycle as a segmentation variable, I offer this chart on purchase rate (not response rate) from the paper:

What you are looking at is a model constructed from actual test results. The model maps probability of purchase by 4 groups of online customers, by weeks since last purchase. Three of these groups are being offered promotions – Coupons, Free Shipping, and a Reward program. The Baseline group is offered no promotions.

Example: Looking at the Baseline (lowest) curve, with week = 0 being the last purchase date, and remembering these customers receive no promotions: about 3.3% of customers will make their next purchase 1 week later; about 5% of customers will make their next purchase 2 weeks later; about 5.5% of customers will make their next purchase 3 weeks later, and so on.

Please recognize that there is a “Natural” purchase rate, as represented by this “Baseline” group – those offered no promotion. This natural purchase rate peaks at about 4 weeks, and after 4 weeks of no purchases, the likelihood to purchase again begins to fall each week that no purchase is made.

Your business model has a chart that looks similar to this one. The peak may be different, the slope may be different, but the general characteristics will be the same. The Baseline group is often called the “control” group, and is simply a sample of the population that receives no promotion, which allows you to measure the natural purchase rate and revenue generated from these buyers.

The chart above shows what Marketers mean when they talk about “Lift”, as opposed to response. Let’s say the response to a campaign may be 8% from buyers 4 weeks into the cycle. If the natural purchase rate for people receiving a campaign is 4% at that same 4 week point in the purchase cycle, then the campaign is only responsible for generating 4% of behavior – literally 50% of the “response” to the campaign. The Baseline or control group tells you the natural buying rate and revenue generated from natural buyers in each point of the purchase cycle, which starts at last purchase date (week = 0 in chart).

This also means that when you do a financial analysis of your campaigns, you should only be taking credit for the Lift caused by the campaign. Said another way, the full cost of the campaign should be applied against only those sales the campaign is responsible for generating, in the above example, the 4% rather than the 8%. As you might expect, this cost allocation against the true performance of the campaign can dramatically affect profitability.

And this is why segmenting customers by purchase cycle contributes more profitability to a campaign than segmenting by transactional content like category, basket size, demographics, and so forth. The timing of the offer is a more powerful determinant of profitability than the content of the offer.

Why is this important to you?

Because if you believe in the power of interactive to “pull” customers in, if you believe that usability and customer centricity really matter, then it follows you should be thrilled to have a high natural purchase rate. In fact, increases in natural purchase rate can be used to prove that customer centricity drives increased profitability.

Logically, if you accept the above premise, “push” campaigns will encounter higher levels of natural demand as a business becomes more customer-centric. Which means that as your business becomes more customer-centric, you should rely on more and more on purchase cycle targeting to drive higher profitability.

Impact of Different Promotions

An example of how to take action on purchase cycles is represented in the study, where free shipping tends to be the most profitable approach for re-acquiring lapsed customers, and discounts are the most effective tool for managing active customers. Look at the graph above to see how this works.

On the left side of the graph, when weeks since last purchase are low, you can see purchase incidence is higher in the “Coupon” group than the “Freeship” group; the Coupon line is higher than the Freeship group so the delta versus the Natural buying rate is greater for Coupons than for Free Shipping.

If you follow the Coupon line down to the right, you can see it drops below the Freeship line at 6 weeks with no purchase and in the out weeks, closely approaches the purchase incidence of Natural buyers. This is happening while the Freeship group maintains a significant delta to the purchase incidence of Natural buyers. If the purchase cycle analysis for your business looked exactly like this one, what should this data mean to you? Primarily two things:

1. In order to maximize purchase rate, customers who are offered Coupons and are non-responsive after 6 weeks should then be offered Free Shipping.
2. Offering a Coupon after 20 weeks of non-response generates very little lift in purchase rate; virtually all the responders are Natural buyers who would have purchased anyway. This means you are probably generating negative profit after campaign and discount cost on these efforts.

I think it’s worth repeating again that purchase cycle (or more broadly, LifeCycle, to include analysis of any action including visits, log-ins, downloads, etc.) curves will not look exactly like this one for your business, and the optimal timing of switching offers by purchase cycle likely won;t be the same.

However, having seen these same types of curves many times over my 15+ years working with online businesses, I can tell you this kind of work is worthy of your attention and effort – and especially so if your company is actively working on becoming more customer-centric. The more successful you are in pulling customers back to you, the more attention you should pay to purchase cycle segmentation to drive company profitability.

If you believe a fundamental part of your business model is to be “interactive”, time since last interaction – perhaps you’d prefer the term “dis-engagement” – is one of those most powerful segmentation approaches you can use.

Related Readings

Measuring Engagement Series/ contains examples of measuring and acting on days since last action as a segmentation tool for Campaigns, Visitors, and Customers.

14-Jan-10 8:00 AM
More: 
Dynamic Customer Management and the Value of One-to-One Marketing

Popularity: 22% [?]

divstrong/strong/div
span style=font-size: 10pt;span style=font-family: Tahoma;
divstrongExecutive Summary/strong/div
divnbsp;/div
pThe authors note the differences between conventional TV advertising and internet advertising when it comes to ad measurement. One of the main differences is that internet advertising is not bound to a schedule that defines an ad’s insertion into a broadcast. Thus traditional reach amp; frequency exposure models that aim to insert a certain amount of ads across a specific number of vehicles such that the ad is exposed to an audience once per insertion at most may not be the best to apply to online ads. It is argued that one of the main reasons why traditional models do not work online is that web users control their own navigation and can potentially be exposed to the same ad insertion more than once. /p
pHence, they proposed the use of a different exposure model based on Negative Binomial Distribution (NBD) which aims to represent user exposure rates to parts of specific sites where an ad is placed rather than the site as a whole. The model integrates well with web analytics as it relies on clickstream data for calibration. /p
pThe authors went on to conduct an actual study using data collected from panel-based software installed on the computers of 1012 users and applied the NBD model to it. They set out to determine whether the NBD model was effective in determining ad exposure for a given month as well as predictive abilities for a future month by creating 1000 ad schedules for a single ad placed in the top 100 subdomains visited by the sample. They concluded that the model’s predicted outcomes and the actual outcome from the user data were nearly identical. Thus, they were able to determine the optimal ad “schedule” for a particular set of ads for maximum exposure in the most visited subdomains on the internet. /p
br
pstrongReview/strong/p
pFor people responsible for online advertising budgets, this is a model that is worth looking into given its applicability to specific parts of a site rather than the site as a whole. If you utilize tactics such as subdomains and microsites with their own unique addresses, this is a worthy read. Insofar as its application to web analytics, it only makes sense that web analytics is an integral part of how and where online advertising dollars are spent given the sophistication of the tools amp; practitioners out there today. As an example, a model that optimizes reach and exposure can be coupled with web analytics by assigning dollar values to each creative element or determining the average order size/revenue per element to determine per-ad effectiveness and campaigns can be optimized on the fly as necessary. Non-revenue based metrics can also be linked to each element to allow for things like visitor segmentation, determine likelihood of repeat visits, etc. This can then inform future decisions on whether to tweak creative, change its placement on a particular site, or choose an entirely different ad space. /p
pIt is equally important to take a look at your own traffic. Web analytics tools can tell you a lot about referred traffic to your site. The study carried out by the authors in this article based their ad schedules on the 100 most visited subdomains from their sample panel. However, there are some sites out there that don’t necessarily get traffic from the most popular sources. Always keep this in mind – you know your site and who visits! nbsp;/p
pLike other math-based models out there, there are limitations. The authors state that this should only be applied to online marketing campaigns (versus multi-channel campaigns), ad executions that run for the exact same amount of time, and that it is an exposure model for banner ads only. There is clearly room for more research to be carried out to fill the gaps but this is a very good start. It will be interesting to see whether this model can be carried over into tactics like email or mobile. nbsp;/p
/span/span
divspan style=font-size: 10pt;span style=font-family: Tahoma;
pThis article is recommended for web analytics practitioners, agency account managers, or anyone who influences how and where external online ad dollars for marketing campaigns are spent./p
/span/span/div
divnbsp;/div
div**A single copy of the full journal reviewed above is available to members of the Web Analytics Association. To request a copy, email a
pnbsp;/p
brbr26-Mar-09 9:00 AM
More:
Modeling The Audience’s Banner Ad Exposure For Internet Advertising Planning

Popularity: 20% [?]

One of the most exciting things about the internet marketing field is its dynamic nature. Site designers can change pages and even re-launch websites whenever they please. An issue sometimes, however, is what to do with the old content. Visitor and Pageview numbers are very important and these numbers can go down significantly if end-users keep running into blank pages or pages that no longer exist. There are a few different redirects and error pages that a webmaster can use to properly send search engine robots and users to the new, updated pages. Here are some of the pros and cons of all possible redirects and error messages.br
br
Redirects:br
301 MOVED PERMANENTLY – This redirect should be used when the old page will no longer exist on the server or you do not want this information indexed by search engine spiders anymore. This redirect is very easy to implement and is very useful if your website is undergoing a re-launch or reorganization of pages. Usage of a 301 redirect gives your site the best chance to preserve any rankings in major search engines that it may already have. There are many free online tools that you can use to check whether your page is utilizing proper 301 redirects. A good example can be found at a title=redirect checker href=http://www.webconfs.com/redirect-check.php target=_blankhttp://www.webconfs.com/redirect-check.php/a. br
br
The following are examples of different scenarios in which you should use a 301 redirect:br
ul
liPermanently moving or deleting a page/li
liNew Domain Name/li
liAutomatically redirecting requests that omit the www in the url/li
liPage redesign with simplified URL strings /li
/ul
br
302 FOUND – This redirects are to be used when a page’s URL is changed temporarily. There aren’t many instances in which a 302 redirect would be the best redirect to use. This is quite alright because this redirect is not search engine friendly. In fact, there was a time in which a 302 redirect was the basis for a black-hat SEO trick that sent search engine robots to a particular site to increase rankings, but kept human visitors on the same, low-ranking site.br
br
You would use a 302 redirect for the following scenarios:br
ul
liRedirect a URL to a similar page to do site maintenance on the old URL/li
liRunning promotional ads, sites or pages /li
/ul
br
Error Messages:br
404 PAGE NOT FOUND - A requested page can return a variety of error messages. The most search engine friendly error message is a 404 Page not Found. This error message is also a very useful webmaster tool because is helps maintain a positive end-user experience. If a user has tried to reach a non-existent page, for whatever reason, a 404 error page can give that user the information needed to either check that they’ve requested the correct page, or let them know that they are at the right site, just looking for the wrong page. Keeping track of all 404 errors that occur on a site can be a daunting task. There are sites, like a title=404 error fixing tool href=http://www.errorlytics.com target=_blankhttp://www.errorlytics.com/a that help webmasters fix 404 errors while maintaining SEO rankings.br
br
CUSTOM 404 ERROR PAGES – 404 error pages can be automatically generated as very plain pages that serve no other purpose than to inform the viewer that they are looking at a 404 page. These pages look nothing like the rest of your website and it only encourages the user to leave that page, and possibly your entire site, as quickly as possible. This is why many website opt to make custom 404 error pages. These pages often carry the same look and feel of the rest of the site and are entertaining in some way, minimizing confusion and anger from the user. There are great lists out there where you can find examples of 404 pages. A simple, but cool 404 page can be found at a href=http://www.accessionmedia.com/example404 target=_blankhttp://www.accessionmedia.com/example404/a.br
pnbsp;/p
span style=font-family: Arial;
pnbsp;/p
/spanspan style=font-family: Arial;br
/spanspan style=font-family: Arial;
pnbsp;/p
/span
pnbsp;/p
brbr5-Feb-09 10:00 AM
More: 
Maintain SEO Rankings User Experience with the Proper Redirects Error Messages

Popularity: 11% [?]

Executive Summary:

Dwyer introduces a new metric called the Adapted Pagerank or APR, to measure social and informational interaction processes of online communities. He uses the APR metric to not only measure the volume of social capital, but to analyze key influences that drive network growth and decay as well.

The APR function is an adaptation of the well known Pagerank function used by Google, which determines the importance of a page by measuring the page’s value of centrality and prestige. Centrality is defined as the “number of nodes to which a given node is connected” while prestige is a “variant of centrality where a node has many incoming ties but is very selective in initiating ties with others.” The Pagerank function quantifies these factors by calculating the ratio of outbound and inbound links of a page, in relation to the ratios of the linking pages. Using APR as one of the algorithms in his analysis, Dwyer examines the relationships between content and its authors in 10 product-oriented online groups. He uncovers that members with high expertise in the subject matter are the most influential players in growing or deteriorating an online network, regardless of who they are linked to. These experts not only hold the key in shaping informational networks, but also help shape the social network within these environments as well.

Review:

By leveraging the knowledge capital currently available, Dwyer creatively avoids reinventing the wheel while providing marketers a solid approach in understanding how their online communities actually work. However, Dwyer leaves readers wanting more validation of how he adapted the Pagerank function to measure social networks. More specifically, it is not clear how he arrived at the factors set for the outdegree parameter (used to evaluate prestige) and the proportioning factor (evaluating page importance of linking pages). In addition, it is vague how the DAG (directed acyclic graph) models are applied and how it relates to the APR metric. Lastly, the limitations of the study were covered minimally, which leaves room for additional uncertainties.

Methodological considerations aside, the results from analyzing the 10 product-oriented online groups is surprising, particularly because of the way network growth is currently understood. Previous studies in social psychology (e.g. Stanley Milgram’s “six degrees of separation” study) view highly connected members as the primary catalyst for information proliferation. “The law of the few” in Malcolm Gladwell’s international bestseller, The Tipping Point, states that individuals who have a higher number of friends and acquaintances are key drivers of social epidemics. However, Gladwell also does state that human behavior is powerfully affected by the environment in which they interact. All in all, the results indicate that marketers trained in classical market research should reevaluate their online measurement efforts, as it is clear that social dynamics in the online world are different than offline world.

In practical application, Dwyer’s findings validate the use of corporate moderators and industry experts in product-oriented communities. A variety of marketers are already jumping ahead of the curve by integrating industry experts into their online communities and now we have good evidence that this approach is helping marketers drive focus and growth in their communities. Moreover, it would be really valuable for marketers who are interested in understanding whether community-oriented groups exhibit similar behavior. Do Facebook members interact the same way that a community of BMW-enthusiasts? Would they exhibit behavior similar to Milgram-esque models or online product-focused models?

Dwyer’s article is recommended to web analytics practitioners who are interested in delving deeper into the nascent field of social media measurement. Among current methods, using APR sounds promising as it is grounded in quantifiable and conceptually sound. However, it might be helpful to reach out to Dwyer to discuss the tactical details of the actual implementation.

22-Jan-09 2:45 PM
The rest is here:
Measuring the Value of Electronic Word of Mouth and It’s Impact in Consumer Communities

Popularity: 27% [?]

div class=peerJournalArticleDetails
pspan class=peerJournalArticleAuthorsSingh, Nitish, Pereira, Arun, Baack, Daniel W., Baack, Donald., (2008)./span span class=peerJournalArticleTitleCulturally Customizing Websites for U.S. Hispanic Online Consumers/span. span class=peerJournalJournal of Advertising Research, Vol. 48, No. 2/span. span class=peerJournalArticlePages9 pages/span./p
p class=peerJournalReviewerReviewed by Christopher Berry, January 2009/p
/div
h2Executive Summary:/h2
pThe authors note the growing importance and influence of the Hispanic Online Consumer to US businesses, and the relative difficulty in reaching them. Their literature review identifies acculturation differences between US Hispanic Consumers as being a major challenge – and go onto define acculturation as “the process of learning a culture that is different from ones own, which in turn leads to change in values, attitudes, behavior patterns, and language use”. Also noted in the review are previous research on the US Hispanic Consumer acculturation levels and its impact on offline media usage./p
pThey use a survey methodology (n=400) to conclude “U.S. Hispanic customers have clear preferences regarding online marketing content and that acculturation level affects these preferences.” Specifically, that that weakly-acculturated US Hispanics in particular have a “significantly higher levels of preference for both English and Spanish language web pages in comparison to highly acculturated Hispanics.” They also found that only weakly acculturated US Hispanics had significantly higher preferences, attitudes, and even purchase intentions for websites that provided information about Hispanic community involvement./p
pThe authors also tested the use of creative involving grandparents, found that it was important to low acculturation Hispanics, and concluded that a “respect of elders” theme would be helpful for marketing to this group./p
h2Review:/h2
pThis article is directly applicable to web analytics, and raises a wicked web analytics problem./p
pThe decision to translate a significant portion of a website into Spanish involves the consideration of several direct and indirect costs. Aside from the direct cost of translation, there are less well known costs, such as the design consideration of navigation and template sizing due to content differences. Specifically, Spanish can require up to 20% more space in comparison to English content. As a result, if an existing website was not developed with Spanish translation in mind, errors such as crowded or broken navigation might very well result. In sum - the decision to translate into Spanish properly by offering a Spanish experience is not always a light or simple decision./p
pThe weak-acculturation finding has an important consideration for measuring the effectiveness of these pages and the overall value of the effort./p
pMany web analytics packages include “Visitors - Languages” reports, that generally sum up the type of language that the browser has been set up for. For example, ‘es’ is the code for ‘espanol’, or Spanish. Language selections are sometimes prefixed by country language localization. For instance, ‘es-ar’ would be ‘Argentina-Spanish’. The likelihood of weakly-acculturated US Hispanics using the default Spanish ‘es’, is low, making the creation of really meaningful custom segments in web analytics software, and then tracking their behavior, less robust./p
pMoreover, weakly-acculturated US Hispanics might not be likely to actually use the Spanish translated pages at all, but indication of their presence is enough to have a significant impact on conversion, brand perception, and likelihood to return. This is the direct implication of the study – the existence of Spanish translation is what is important to this group – not necessarily the actual use./p
pAs a result of these complicating factors, utilizing solely a pure web analytics approach might not be the most accurate way of gauging the value of translation and effort. Rather, a combined approach incorporating pure web analytics with survey methodology would yield more accurate assessment of translation efforts./p
pThis article is recommended to web analytics practitioners of companies that target US Hispanic consumers. It provides factual evidence that Spanish translated section of a website is important and well worth the effort./p
div class=peerJournalAccess
pA single copy of the full journal reviewed above is available to members of the Web Analytics Association. To request a copy, email
script src=/js/info_lindsay.js/script./p
/div

brbr22-Jan-09 2:00 PM
The rest is here: 
Culturally Customizing Websites for U.S. Hispanic Online Consumers

Popularity: 23% [?]