PERSONALIZED CUSTOMER / PRODUCT RESEARCH
I invested $100 in campaign contributions to Barack Obama’s campaign before the 2012 presidential election. The personalized interactions have been among the best I have had with any political organization. Here is an example of something no other politician has done for me. I received a letter from Michelle Obama. First, the letter mentioned the progress the Obama administration has made in immigration reform and climate change—two issues I passionately care about. Next, the letter asked me, “What is the number-one issue you care about?” The letter thanked me for sharing my opinions in the past and encouraged me to keep the communication going. As I have responded with my top issues, the campaign has continued to refine its messages and keep track of my issues, keeping the messages personalized and tracking the president’s actions in response to my concerns.
What we are seeing here is a fine example of consumer research at a personalized level. Customers interact with a number of companies that supply goods and services to us. How often are our top issues tracked by them? I was grumpy about the slow speed of my Internet connection and saw an advertisement from my telecom provider offering me five times the broadband speed for an Internet connection. I called them to ask if I could subscribe to the faster connection. The call center agent told me the service was not available in my area. He casually told me to call back again in the future to check whether or not the broadband service would be available. I asked him whether I should call every five days until the service is available. He was not sure, but said calling once a month was not a bad idea. Each such call would cost this company about $20, and they would not collect any information from their customers to find how many in a given area were calling for higher bandwidth. If I call them five times about this upgraded line, they will have incurred enough call center expenses to forgo any profits for selling the upgrade to me for the first year. They were ignoring simple indicators of customer demand, which they could have used to fine-tune their bandwidth demand and product availability across geographies. The only tracking they required was a list of all customers who had expressed the need for higher bandwidth. As soon as the bandwidth was available, they could connect with customers and offer them an upgraded product. In addition, now they would have demand information for their new upgraded service, which could be used by their network engineering group in deciding where to build the network infrastructure for the upgraded service.
I can recollect a similar story about my favorite airline. I was seeking a seat on a particular flight, and the response I received was identical to the one I received in the example above—“Please call back once a month, and you will hopefully find the seat you are seeking.” An airline can be very sophisticated in their service, especially as they deal with their elite customers. They can easily track what their best customers would like to do and offer them customized packages based on customer needs. All of these are examples of customized product design based on customer requirements. Marketers often cater sales and order processing to the products they have rather than to the needs their customers have. In a micro-segmented, or personalized, marketing situation, they can build a set of customer profiles and offer customers products based on what they need or on what similar customers are buying. As you may have already discovered, this is what Amazon already does. Every week, I receive a list of books that Amazon recommends to me to buy, which is based on my past purchases, new books released, and the purchase behavior of similar customers. Amazon’s recommendation engine mines through past purchases and book classifications to build the recommendations.
Customer usage is the best source for customer and product research. As marketers offer complex products with many features, and often find customers not using many of the features, a product can be rationalized. That is, by analyzing product usage, a product manager may drop product components, features, or accessories that no one is buying. In a study with a telecom provider, I found that 98 percent of their customers purchased 197 product components or features out of nearly 20,000 offered to their customers. As the product managers insisted that they were dealing with a fat tail, I extended the analysis to 99 percent of the customers and found only 500 product components or features in use. Their call centers were training their sales personnel for six weeks, most of the time teaching them how to enter 20,000 product codes in their order-processing systems. While the rest of the product components did not offer any exceptional margins or market advantage, they were available for an occasional buyer. As the company began to introduce a simplified product line through their sales channels and relegated the “once in a blue moon” product components to a specialized sales process, their sales training time was dramatically reduced from six weeks to two days.
Products are often designed to comprehensively cover all customers. With any office software tool, like Microsoft Word, which I am using to write this book, most users employ a very small number of features, and a couple of power users employ a specialized set of features. Could we custom design products based on customer needs and offer additional components as a customer requires those new capabilities? Product usage analysis can help us map product features to customer groups and simplify offerings targeted to those groups.
Product automation provides an enormous opportunity to measure customer experience. Today’s sophisticated consumers take photos digitally and then post them on Facebook, providing an opportunity for face recognition. They listen to songs on Pandora, creating an opportunity to measure what they like or dislike, or how often they skip a song after listening to the part of it that they like the most. They read books electronically online or on our favorite handheld devices, giving publishers an opportunity to understand what they read, how many times they read it, and which parts they look at. They watch television using a two-way set-top box that can record each channel click and correlate it to analyze whether the channel was switched right before, during, or after a commercial break. Even mechanical products such as automobiles are increasing electronic interactions. These customers make all of our ordering transactions electronically, giving third parties the opportunity to analyze their spending habits by month, by season, by ZIP+4, and by tens of thousands of micro-segments. Usage data can be synthesized to study the quality of customer experience, and can be mined for component defects, successes, or extensions. Analysts can identify product changes using this data. For example, in a wireless company, analysts isolated problems in the use of cell phones to a defective device antenna by analyzing call quality and comparing it across devices.
Products can be test-marketed and changed based on feedback. They can also be customized and personalized for every consumer or micro-segment based on consumers’ needs. Analytics plays a major role in customizing, personalizing, and changing products according to customer feedback. Product engineering combines a set of independent components into a product in response to a customer need. Component quality impacts overall product performance. Can product managers use analytics to isolate poorly performing components and replace them with good ones? In addition, can they simplify the overall product by removing components that are rarely used and offer no real value to the customer? A lot of product engineering analytics using customer experience data can lead to the building of simplified products that best meet customer requirements. The solution requires a data-driven mapping of customer needs and product usage to product components. The mapping can be utilized by product marketing to offer product packages,
Figure 4.1 Mapping of Customers and Products
bundles, and customization to specific micro-segments. The mapping can also be used by product engineering to change product components based on customer needs and product usage. Figure 4.1 depicts a data- driven mapping, which can be deduced from the types of observations described in chapter 3 and used for marketing segmentation, product marketing, and product engineering.
The first set of dots represents customers. These customers exhibit certain needs, depicted by the second set of dots. Customers may use a variety of means to communicate their needs, such as using social media to tell friends they are interested in purchasing a product, or by searching for product-specific information on the web. The links between customers and needs can be derived by statistical analysis of observations and can be depicted as strengths in the lines connecting customers to needs. A need is associated with a usage, which represents how the need is fulfilled. Thus need to work at home (a need) may be related to use of high bandwidth (a usage), as these users consume high bandwidth as they share presentations via email and use corporate applications. The usage can be linked to product offerings from a marketer, typically in the form of a grouping of components, which are sold together. The marketer may group these customers based on needs and usage, and may develop specific offerings made up of product components to respond to these customers. In my earlier example, I described my need for a higher bandwidth Internet connection to my telecom provider. In addition, I may be showing higher usage of bandwidth in the daytime on weekdays, and average use outside of business hours. A telecom provider may group such customers into “daytime work at home” (a customer grouping or micro-segment) develop an offering for “higher bandwidth during weekday,” using product components available from the engineering organization. The engineering organization may employ these offerings to create engineering components that offer different bandwidth by time of day to different customers, based on their product subscriptions, by combining components—“bandwidth” and “higher bandwidth network policy” (see figure 4.2). Needless to say, this offering may provide additional revenue to the telecom provider, make use of idle bandwidth during the daytime, when the rest of the neighbors are “daily grinders” and commute to a work location, and lead to a higher loyalty rate among people who work from home. Product marketers can discover many such micro-segments by analyzing the data. They can also offer products based on these segments, focus their campaigns on a targeted set of customers who are exhibiting specific behaviors, and observe the intake for those products in the targeted segments. The links shown in this figure represent observations and data sets and can be derived by mining data observed from customers. Once the model has been identified, it can be used for targeted
Figure 4.2 Daytime Work at Home Micro-segment campaigns to specific customers, or for designing new offering, by combining product components.
Intelligent segmentation and campaign management systems based on these approaches have resulted in significant uptake in revenues and customer satisfaction. The campaigns developed using these techniques were far more focused, and also far more successful, in comparison to broader campaigns. Prior to product automation, data collection was very difficult. However, recent advances in product and touchpoint automation have given rise to observation data, which can be collected and analyzed without significant investment.
The same approach can also be used for product rationalization. Marketers offer a large number of products and components to their customers. Usage observations are key to identifying and isolating sporadically used product components and features. The analysis can also be focused on badly designed components, which need to be either redesigned or removed from the product mix. Product managers may use profitability analysis, along with competitive intelligence, to decide which features are not adding value to product selling or usage, and drop components that offer no leverage to the product mix.
To conduct this analysis and predictive modeling, we need a good understanding of the components used and how they participate in the customer experience. Once a good amount of data is collected, the model can be used to isolate underutilized or badly performing components by isolating the observations from customer experience and tracing them to the component. Complex products such as automobiles, telecommunications networks, and engineering goods benefit from this type of analytics around product engineering.
The first level of analysis is in identifying a product portfolio mix and its success with customers. For example, if a marketer has a large number of products, they can be aligned to customer segments and their usage. We may find a number of products that were purchased and hardly used, leading to their discontinuation in six months, while other products were heavily used and sparingly discontinued.
Once we have identified less-used products, the next analysis question is whether we can isolate the cause of customer disinterest. By analyzing usage patterns, we can differentiate between successful products and unsuccessful ones. Were the unsuccessful ones never launched? Did many users get stuck with the initial security screen? Maybe the identification process was too cumbersome. How many users could use the product to perform basic functions offered by the product? What were the highest frequency functions?
The next level of analysis is to understand component failures. How many times did the product fail to perform? Where were the failures most likely? What led to the failure? What did the user do after the failure? Can we isolate the component, replace it, and repair the product online?
These analysis capabilities can now be combined with product changes to create a sophisticated test-marketing framework. We can make changes to the product, try the modified product on a test market, observe the impact, and, after repeated adjustments, offer the altered product to the marketplace.
Let me illustrate how big data is shaping improved product engineering and operations at content providers—cable companies and telecom providers that are providing regular cable channels, over-the-top content, Internet Protocol television (IPTV), on-demand, and so on. For many decades, the cable infrastructure was essentially a lot of fat pipes connected to a cable company’s content hub where cable employees ran around on roller skates to change contents as requested by the consumers. All this changed with the DOCSIS 3.0 (Data Over Cable Service Interface Specification) standard that started to offer digital content over high-bandwidth digital pipes. In the meantime, telecom providers started to offer IPTV. Also, Netflix, Google, and Apple began offering content on the Web, which could be displayed on the regular television. Interactive television has radically changed the game for the entire content industry. The content is no longer broadcast to a set of homogeneous channels. Consumers have the ability to customize their content, fortunately, under the minute scrutiny of the content provider. Cable operators and telecom providers collect enormous amounts of data about the network, including network transport information coming from the routers and the switches, as well as usage information, which are recorded each time we watch content on a screen. For larger cable and telecom providers, the usage statistics are not only high volume (in billions of transactions a day) but also require low-latency analytics for a number of applications. This data is quite valuable for recommending new content, placing advertisements during the viewing, and designing new programming by content providers.
Netflix rose as a viable competitor to cable and telecom providers. Starting from a DVD mail-order business, Netflix has rapidly grown into an online content provider with on-demand customized content it offers to its subscribers through a monthly subscription program. As customers use Netflix services to watch content, their usage data is meticulously collected, sorted, stored, and used for analytics to provide content recommendations. The Netflix portal offers two major ways to find content to watch. It provides ways to search for a movie, and it also makes recommendations based on past viewing as well as similar viewing by other viewers. According to Netflix’s director of engineering, Xavier Amatriain, “Almost everything we do is a recommendation. I was at eBay last week, and they told me that 90 percent of what people buy there comes from search. We’re the opposite. Recommendation is huge, and our search feature is what people do when we’re not able to show them what to watch.”2 Using big data analytics, Netflix has been successfully winning its customer base from cable and telecom providers. Most Netflix customers use the cable / telecom infrastructure to connect to the Internet, and use Netflix for viewing their content.
To facilitate the development of customized content recommendation, Netflix first decided to crowdsource its recommendation algorithm during 2006-2010. Netflix made anonymized usage data available to anyone interested in competing for the best recommendation engine.3 The competition received widespread attention from researchers worldwide, including research and development organizations, universities, and others. Unfortunately, these crowdsourced algorithms had to be stopped because of privacy concerns.4 However, Netflix has continued to work on their recommendation engine using meta data collected from the movies and usage data collected from their viewers. Netflix employs 40 freelancers to hand-tag television shows and movies. These are product components shown in figure 4.1 above. Netflix has a team of over 800 engineers working at their Silicon Valley headquarters, developing sophisticated algorithms for combining meta data about movies with usage information from their viewers to build recommendations that viewers see on their screens.5
If, through interactive recommendations, a content provider can precisely measure and influence the audience, can this deep insight about the audience be shared with advertisers? Let us now turn our attention to how online advertising is changing in this interactive era.