Big Data analysis alone will always have a hard time understanding the “why” of what we do. That realm of understanding is MR’s key differentiator.
In the Fourth Quarter 2013 issue of Alert! I wrote about the need for marketing research (MR) to adapt in order to survive. Over the years, our field has struggled to create business relevance and demonstrate return on investment (ROI). Now, Big Data poses a substantial challenge to our profession.
Wired Magazine recently ran a piece entitled The End of Theory: The Data Deluge Makes the Scientific Method Obsolete. You may have seen the Big Data blog posts saying in effect: MR had their chance, and a good one at that, so shouldn’t we get the opportunity to take over and see what we can make of things. So, does a Big Data world still need research design? Or, will unprecedented volumes of data and machine learning algorithms put MR out of business?
Most Big Data folks fail to recognize (or just do not talk about) the problem that is inherent to the type of data they examine in pursuit of answers. We leave digital footprints everywhere and those footprints do a good job of cataloguing the “what” of our activities. At best, that is half the story. Big Data analysis alone will always have a hard time understanding the “why” of what we do. That realm of understanding is MR’s key differentiator. Historically, however, our profession has repeatedly failed to prove its value, leaving business leaders with little reason to continue investing resources. This failure is why we have so many models in MR. Within the customer satisfaction specialty, MR’s failure to consistently demonstrate a compelling ROI is a driving force behind the rise and success of experience measurement/management firms.
My last article highlighted three client cases where a new measurement and analysis technology is achieving cost efficiencies and improving data quality. Utilizing binary-scaled questions, this technology provides for ease of application in any data collection scenario. Simplicity of questions also eliminates bias that could arise from respondents’ educational, language, and/or cultural differences. The resulting data has less measurement error and is comparable, regardless of collection method or respondent profile. With the simple response scales, the adaptation of some data mining techniques, and good research design, we are able to impute missing values for skipped questions at well over 90 percent accuracy. The combination of simple binary questions and imputation allows a client to field short, tactical, behavior-based surveys that yield very high completion rates. We then embed this low-burden, direct customer feedback method in a comprehensive customer experience measurement and management framework.
With this article, I will be detailing a case that illustrates the clear path of superordinate achievement possible when MR and Big Data work together. This path is not only wide enough for the two specialties to traverse together, but calls upon each discipline to utilize its own core capability and strengths. This powerful combination provides business leaders with relevant, action-oriented information and demonstrable ROI. The case utilizes the binary and imputation technology within a holistic customer satisfaction/experience program. The combination of content (from Big Data) and context (from MR) allows us to predict customer behavior and prescriptively determine how to influence customers in order to increase profits most effectively.
Overview of the Model
Figure 1 (page 6) is a simplified version of the comprehensive customer experience model we use both predictively and prescriptively with clients. Certain specifics have been generalized to protect client confidentiality. My discussion starts with the Predictive Framework.
In Figure 1, reading left to right, a cause and effect relationship is illustrated. This model is an extension of Harvard’s Service-Profit Chain (updated as the Value-Profit Chain). The cause-and-effect chain begins with employees. Satisfied employees are more efficient, make fewer errors, innovate, and are more solution-oriented. Ultimately, they produce a higher quality product or service which customers experience. Customers first experience the better quality in an initial transaction interface with a provider organization. The interface, or more correctly each of the entry points, begins with operations and is depicted on the graphic in red. From a measurement standpoint, operational metrics for each of the entry points is collected and monitored by the provider. The measures represent client derived Big Data. The provider uses the specific metrics to guide and manage employees and processes engaged in creating the customer experience within a specific entry or touchpoint. Note the provider-customer interface points for Contact Center, Installation, Repair and Brand/Advertising. Performing well on these metrics initiates a great customer experience.
The green blocks represent customer feedback about the touchpoint(s) they have experienced. The model empirically proves that better customer experiences are the result of better operational performance in the metrics associated with the touchpoint(s). In other words, as operational metrics improve, the customer experience with the associated touchpoint improves. Here, in the collection of customer feedback, is where the value of the binary and imputation methodology (subject of the earlier article) is most directly felt.
The blue area on the far right of the graphic represents the overall customer relationship with the provider. We measure this relationship independent of any touchpoint experience. This independent measurement is important for three reasons:
- A customer’s overall relationship with a provider is the best indicator of the customer’s future behavior.
- Many customers do not have repeated touchpoint experiences with a provider.
- Goodwill (or the opposite) builds over time and touchpoint measurement alone misses this aspect.
As before, the model empirically demonstrates that as the touchpoint customer experience improves, the overall measures improve.
Financial Outcomes and Complaint Behavior, noted in the graphic, are direct expressions of customer behavior. As satisfaction increases, the model predicts better financial outcomes and fewer complaints. The model uses the control variables to adjust for seasonality and geographic differences for this client.
Finally, social media data acts reciprocally in the model. In other words, social media data influences the touchpoint experience of customers, as well as being influenced by the touchpoints. For this client, the data comprising the social media input includes Facebook and Twitter data, as well as other forms of blog comments.
Big Data and MR Coexist in the Model
The inclusion of Big Data in the model (including that which comes from the client’s own operational processes) provides this client two additional primary types of benefits. First, the Big Data provides a number of additional prediction points. Including quantitative Big Data gives the client a greater number and variety of potential variables to manipulate in order to better the customer experience and influence future behavior. It also provides a lens for profiling customers, allowing the client to better match its offers with customers. Second, unstructured (or qualitative) Big Data provides an opportunity for early warning and more tailored service recovery when service levels deteriorate.
Unstructured social media enhances the touchpoint survey feedback. Comments and other content narrative are early warning indicators. For example, this sort of qualitative data is used initially to gauge the severity of a problem or issue with a product or process. After identification, the data helps determine corrective action such as levels of service recovery and/or amends offered to certain customers to mitigate damage to customer relationships.
To assist the company in determining levels of service, pricing, availability of promotion, etc., Big Data from a third party append, provides input for financial profiles of customers and prospects. This particular client follows established business rules that categorize customers with higher lifetime value profiles and treats them as more valuable assets – worthy of investment that translates to better service, pricing, promotion and product. Big Data supports this strategy with key profiling information.
The Power of Collaboration
Big Data provides some big benefits to the framework of customer measurement. But, for all the indirect value coming through the integration of web-sourced Big Data, it does little to provide complete context around the customer. Alone, it is of limited use in predicting the future behavior of the customer. I will not pretend to be the first to point this out as you are likely familiar with some version of this dilemma currently being discussed in the business press.
On the other hand, MR, when well thought out and executed, provides deep understanding and context. It explains why our digital footprints are where they are and why we move at differing speeds and directions. No one forces a customer to click a link, provide information or take a specified route through a Web interaction. In the same vein, no one can make a customer tweet or post to Facebook the “why” of their tweet or post. MR provides the “why” that goes along with Big Data’s “what.” If we were to tap that deeper context we would utilize a conversation – a dialogue, or stated more traditionally, a survey.
Within a customer-provider relationship, there are many more desires for feedback than even a willing customer can accommodate. Look back at Figure 1 and notice there are 6 independent opportunities for feedback. As with all surveys that are designed to actually produce actionable intelligence, each survey has a fair number of questions designed to drill down into the specifics of the touchpoint interaction or the relationship overall. In fact, in the program illustrated by the graphic, there are a total of 102 individual questions that could be asked of a customer of this client. The binary/imputation technique I have already written about makes it possible to utilize all these questions, without overburdening respondents or destroying data quality.
These specific survey questions provide the bridge between the Big Data operational metrics and the customers’ overall satisfaction. Without them, the ability to accurately predict customer behavior is virtually nonexistent, with the best case being the correlation between higher levels of satisfaction and better financial performance. That level of understanding may be adequate when resources are not constrained, but when trade-offs are consistently necessary, there is little comfort in knowing that if you happen to guess correctly things might work out ok.
Using the Model to Improve, Plan and Differentiate
The model in Figure 1 reflects what is now referred to as a causal-linkage model – one of many empirical proofs of the Service-Profit Chain. However, it represents a significant step beyond that thinking in at least two ways:
It is a comprehensive model insofar as it provides complete integration of measures in a single predictive framework.
Related to the point above, the use of Big Data from a variety of sources, representing both quantitative and qualitative data types, is revolutionary. As a roadmap, it provides guidance to survey measurement professionals and data scientists alike who have spent time struggling with smaller versions of the model, with data from legacy systems and rudimentary activities like ETL (Extract Transform Load).
Models like this are deployed for our clients as interactive simulation software tools. Loaded on a computer, a user can change a value at any point in the model and then watch as the software uses the prediction algorithms to change the values in all related dependent variables.
For operational management, customer satisfaction and experience management are difficult, nebulous concepts; who knows what the term “customer focus” means? Rest assured, though, customers want it and operational folks work tirelessly to provide it. Do operational employees need to know what “customer focus” means? Not if the client is using a model like the one described here. Our clients’ operational management uses the model to identify the metrics that are most important to their customers. Managers use their own vocabulary, knowledge, and experience to monitor and optimize these important metrics. The net results they achieve are high levels of customer satisfaction, positive customer experience feedback and significant influence on customers’ future behavior. There is no more direct a way to confidently pair insight, action and outcome. For example, management can link a customer’s report of “satisfied” with time to respond (transactional survey question) with the operational metric that indicates how long it took them to actually respond. Armed with this knowledge, management can “tune” their operational respond time to achieve satisfied customers, while avoiding unnecessary investment in resources needed to respond.
Running a simulation multiple times, with each variation involving a different level of potential improvement, allows operational managers to engage in multiple iterations of “what-if” scenarios. When they have a large variety of potential initiatives, the model is run and reset multiple times to test each one of the ideas. Because there are links to financial outcome, the forecasted outcome for a potential initiative provides the manager with cost-benefit analysis for his/her intended initiative – before dollars are committed to the implementation. Managers who are tasked with improving the customer experience thus use the model to identify efficient means to achieve their objectives and forecast the financial benefit for the achievement of their objectives.
The model offers a ready-made test environment for social media campaign managers. Since the framework combines direct and indirect links with these types of communications, the evaluation of initiatives are more contextually rich and better reflect reality experienced by today’s customers. The same types of interactive scenarios can be run changing the social media data inputs.
Historically, clients struggle with business case development, challenged to provide meaningful forecasts of ROI through changes in customer behavior. Now, the linked financial outcomes included in their experience model are used to create defensible pro forma financial forecasts for business cases. With a market where customer choice is free ranging and switching barriers are low, this specific client has found that a forecast of customer behavior is much more powerful and convincing than vague statements suggesting increases in customer recommendation and correlation to financial success.
Annual business planning is a third application of the model. Management knows that good planning includes new ideas from across the business, but historically they often were forced to cut short their planning because of time constraints. Instead, they opted to go straight to budget development and implementation plans that were focused on controlling costs, doing little planning for the future. Here, the linkage model has played a significant role in expanding the business planning process for clients.
Because of the financial forecasting capability of the model, projections are accomplished quickly thereby allowing time-constrained managers to spend more time in the planning process and less in pursuit of the almighty number for the coming year. Since management is sometimes challenged for business planning ideas, the model can be set up to run prescriptively as well. In these situations, business planning begins with a growth or revenue target. The model is then run backwards as a solver. In other words, the client sets a few parameters and identifies a financial objective. Then the model is run in reverse. Doing so provides one or more scenarios that will achieve the desired targeted financial outcome. Planning can then commence in vetting and fully articulating a plan that will achieve the desired financial target.
When the clients engage in true scenario planning, this framework can be set up to be as wide-ranging as is supported by data. The Internet, social media and Big Data present opportunities to expand the model so that it better represents the dynamics of a real market and population of customers and potential customers. In the example model, unstructured social media data and elements of Big Data are used to better understand and explain customer behavior, provide early warning about problems or, in positive contrast, provide early warning on opportunities for differentiation and success.
Where We Go From Here
I will close with a few hypotheses, suggestions and questions.
- More desire/need for data driven decisions will highlight the inadequacy of the singular views offered by MR or Big Data alone. Realization of this will force the two together.
- Desire for direct feedback from customers and prospects will continue to increase, and with falling barriers to entry – more direct feedback will be available; but at what cost and quality?
- MR surveying will continue to evolve away from questions about attitude and motivation, towards more concrete, behaviorally oriented questions (focusing on provider’s behaviors).
- Prescriptive analysis will mature. Combined with greater customer choice and falling switching barriers, customers will have greater opportunity to create completely customized experiences.
- Companies will be rewarded for enabling individualized customer experiences.
My opinion about the future of measurement applied within business settings is likely neither complete nor completely right, but I believe we all should be excited about our opportunity to contribute to the continuing evolution. In my mind, that evolution involves embracing Big Data and coordinating our work with the new field to achieve more than either can accomplish alone.