Comprehending the complete customer
Which came first – attitude or behavior? This is a question researchers, marketers and product engineers are no longer trying to answer. Instead, the two elements are increasingly being fused to create a composite view of the customer. Behavioral inputs are detailed, accurate and real-time, while attitudinal inputs are measurable KPIs for a business. The insights resulting from integration can potentially influence a wide range of decisions at an individual customer or segment level.
*This post was initially published in Analytics Informs on May 7, 2013*
Which came first – attitude or behavior? This is a question researchers, marketers and product engineers are no longer trying to answer. Instead, the two elements are increasingly being fused to create a composite view of the customer. Behavioral inputs are detailed, accurate and real-time, while attitudinal inputs are measurable KPIs for a business. The insights resulting from integration can potentially influence a wide range of decisions at an individual customer or segment level.
Successful integration of behavioral and attitudinal data inputs requires some investments in people, process and technology. However, the payoffs can be endless and start with busting myths of self-reported perceptions with the reality observed in actual user behavior. This paper examines business cases to apply behavioral and attitudinal data elements together, steps involved in establishing a link between the two, technological and organizational requirements or constraints and last but not the least, the analytical engine to drive this movement.
Why is it important to integrate behavior and attitude?
Understanding how the usage and behavior of a person drives his or her perceptions and attitudes about a product or service has always been the holy grail of primary marketing research and analytics. However, the search for this grail has been made easier and more likely in recent times, thanks to the following enablers:
Technology, which is helping to capture various signals in a cost-effective manner
Analytics, which is becoming core to business decisions, driving the need for more data sources
Consumers, who are not shy to generate and share their digital, behavioral and social footprints
Culture, which encourages experimentation and data discovery, which is also becoming prevalent in organizations
Organizations are continuously trying to improve the understanding of their end customers – to enhance design and better prioritize engineering and marketing efforts. Traditionally, primary surveys have been the chief medium used by organizations to understand customer perception of products. However, with the advent of digital media and its vast ability to understand customer behavior in its granularity, it has augmented the traditional approach by incorporating behavioral data along with the traditional data sources to provide a bigger picture.
Figure 1 indicates that traditional data sources only provides a partial view of the customer’s interaction with an organization and only answers the “who” and the “why” questions. However, integration of behavioral data (usage logs and interactions) gives a holistic view of the consumer, thus answering the “how” and “what” questions as well.
Holistic view of a consumer answers not just the “who” and “why” questions but also the “how” and “what.”
What is stopping everyone from doing this?
Finding a needle in a haystack is akin to deriving insights out of big data, or vice versa. The difficult but not impossible task of behavioral and attitudinal data integration comes with its own share of challenges and roadblocks, including:
1. Big data processing. Behavioral data representing a considerable size of the sample population and encompassing its complete machine usage data is going to run into petabytes of data. Handling, processing and analyzing such voluminous data requires one to look beyond the traditional approaches and architectures available for data management. Solution: Big data platforms such as Hadoop.
2. Asking the right questions. Sometimes a traditional hypotheses-driven approach might not be a suitable solution considering that the problem mandates understanding the scope of the issue and then framing the right questions upfront. Solution: Discovery-driven framework.
3. Business consumption. Academic solutions to a problem are always exotic and exciting. Nonetheless, the key aspect of such an exercise is to make it easily implementable and business ready in terms of consumption and action. The biggest challenge would be convincing business groups of the validity of such an approach and quantifying its impact and findings in a manner that is easily consumable and intuitive. Solution: Setting up a channel for frequent and easily consumable insights.
Making integration a reality
When solving innovative and unchartered analytical and business problems, it becomes imperative to adopt the right problem solving approach. This brings forth the dilemma of using a discovery-driven approach as against the hypotheses-driven approach. Traditionally, lack of structured data has often been a deal breaker in the hypotheses-driven approach, while an abundance of unstructured data is used as a springboard in the discovery-driven approach.
Looking at Figure 2, it is apparent that it is relatively easy to have the required datasets and the skillsets to process data, but it is paramount to also have the right mindset to be able to solve the problem efficiently. And this is where the discovery-driven approach plays a major role.
Datasets, skillsets and mindsets are all requited to solve difficult problems efficiently.
Thus, this paper focuses on a fail-fast, learn-fast discovery-driven initiative to tackle new and rapidly evolving problems. The approach aims to start with the “data story” rather than the business problem.
Case study
The program and its constituents: The case study – an integration program of behavior and attitude – was undertaken for a product-based company that was interested in understanding a customer’s usage behavior and sentiments expressed toward its products in order to better design new features and improve targeting strategies. The program involved end-to-end enabling of a closed-loop system to effectively integrate, analyze and incorporate behavioral and attitudinal elements (people, process & technology) as shown in Figure 3.
Closed-loop system to effectively integrate, analyze and incorporate behavioral and attitudinal elements
The program was brought to fruition as a result of the marriage of two different data sources: machine log data from selected respondents and survey data from the same set of selected respondents.
Machine log data contained behavioral data with spools of information at a user-event level and at the lowest possible granularity. This data constituted the biggest chunk of the analysis (running into terabytes of data) and eventually led to the big data challenge.
Survey data contained attitudinal data where the same set of respondents was surveyed on a host of satisfaction and perception related metrics. Respondents were asked to rate the technology provider and its key products under consideration across a list of dimensions. This was done to gauge their true sentiments of and toward the products and the brand as a whole, and was captured through longitudinal surveys across time.
Challenges: The most critical element of the entire data exercise was to integrate the disparate data sources. This was accomplished by establishing a common identifier for both the respondents and the machine users, thereby establishing a way to bridge their multiple inputs. Availability of a common element is but one of the many key steps in this program. The next and most important step is actually integrating them, which is a huge challenge both in terms of logistics and logic.
From a logistics standpoint, handling a large amount of behavioral machine data requires new data paradigms that are unconventional but efficient. This called for big data constructs (e.g., Hadoop, Hive and R). Data were stored on Hadoop clusters (distributed computing applications), which store and process big data efficiently. R was the statistical program used to interface with the Hadoop cluster to perform statistical analyses on the big data.
In terms of logic, integration poses the challenge of identifying the necessary data to be considered for the analysis. It is imperative to only consider manageable subsets of the entire dataset as against the entire population of telemetry data.
Customer’s product lifecycle journey
Analysis and solution:
The program was planned as a four-phase analysis covering various dimensions of the synergy produced out of integrating behavior and attitude. It was designed to reflect and recognize a customer’s product lifecycle journey, which also has four stages (acquisition, service, engagement & retention and penetration & growth). The focus was not only on integrating and analyzing the data, but also deriving insights from a product lifecycle perspective.
The objectives of each of the four phases of the program include:
Behavioral profiling of various attitudinal segments
Understanding linkage between behavioral and attitudinal data
Behavioral segmentation of the user base, with an attitudinal validation
Event triggers and longitudinal analysis
Phase I: The objective of this phase of the analysis was to quickly understand and appreciate the data at hand. With the linkage between behavioral and attitudinal data previously unexplored, it was considered supremely imperative to get to know the data better and get a pulse on user’s behavior.
Multiple data slices were performed at various levels on the integrated datasets. A set of key questions and hypotheses were addressed at this phase, which helped obtain a better understanding of the consumer’s mindset and also helped break down age-old myths and misconceptions about certain segments of consumers. The questions were designed and prioritized in such a way so as to deliver maximum business impact. In summary, the objective of this phase was to illuminate actions and attitudes that drive customer use and advocacy – connecting the “who” (descriptive profile of the user) to “what” (behavioral data on use and machine profile details) with insights into the “why” (perceptions and opinions about the user’s experience).
Some of the key questions for the phase were:
Are there differences in usage behavior between different categories of product owners?
How does level of engagement with the brand or product affect usage behavior?
What is the impact of the consumers’ lack of awareness of product updates?
How do we remove barriers for migration of users between product versions?
Phase II: The objective of this phase was to establish a linkage between the behaviors and perceptions – in other words, understand the causal relationships that exist between the two disparate sources.
With the axiom that correlation is not necessarily causation in mind, the analysis was carried out with utmost care to establish relationships that were both correlated and causal in nature. Two key questions emerged:
1. What is the relationship between user activity, user profile and attitudes?
Business hypotheses:
Use of more features of a product will increase suite satisfaction and likelihood to recommend.
Certain customer segments and groups post higher activity than others.
Quick finding: Usage is not necessarily the key indicator of satisfaction; however the version used has a strong linkage with satisfaction.
2. How does use and perceptions of competitor products impact product usage and attitudes?
Business hypotheses:
Use of a competitor product is associated with low use or rejection of product in consideration.
Quick finding: Use of competitive products does not erode product-in-consideration’s usage
The next two phases were more focused on leveraging the understanding and insights from the initial two phases and creating actionable customer segments and targeting business rules, which could be readily implemented.
Phase III was dedicated to the creation of behavioral segments formed out of the usage patterns. These behavioral segments were then compared against existing segmentation schemas, to validate and enrich the targeting strategies.
Phase IV was focused on longitudinal analysis of the integrated data, i.e., trying to understand how behavior affected attitude at one point and how the same attitudes affected behavior in a future point in time – thereby trying to understand the virtuous cycle of synergy between usage and perception.
Results and impact:
The program was a powerful and prolific initiative, in that it provided many actionable insights and eye-openers, including:
It was a myth-buster by nature as it brought down many age-old misbeliefs. The program helped create new knowledge about extant segments and busted old myths around non-marketing to a few user segments. This re-look at non-marketed user segments from a behavioral perspective revealed that they were not avoiding the brand, but were actually super-users of a particular product within the brand, which had them wrongly classified as avoiders.
It helped re-affirm some business beliefs by adding some science and data validation to insights. The program’s data shows that version used is a high predictor of activity level. Users on the most recent versions are more active than previous version users, and both user groups far surpass activity levels among remaining oldest version users.
It also provided design inputs and product enhancement ideas from the horse’s mouth. Usage of a certain feature was analyzed, revealing that those who used that feature typically followed usage of another feature or product. This highlighted potential to package features together, thereby heightening engagement.
It delivered on its promise of establishing a bridge between behavior and attitude. Users with high satisfaction and likelihood to recommend show higher activity than less satisfied users. But the usage of those who consider the product to be easier to use is not significantly different from those who have lower perception toward ease of use.
It also helped perform competitive analysis of products from a usage standpoint. Among cross-users of both suites, competitive products turned out to be complementary to the product under consideration. Combined users of both products were more active than those of individual products.
What can be learned from this exercise?
To implement a similar initiative in any organization, one needs to be cognizant of the following factors:
Define business objectives in advance. It is paramount to define the business objective of the exercise upfront and also have a clear idea of the potential business impact of each objective.
Estimate infrastructure needs, especially for big data. The exercise dealt with nearly two billion rows of data. To be able to do that, getting infrastructure right is absolutely essential. Infrastructure here denotes both the ability to capture such voluminous data and also the ability to store and process them.
Devise strategies for optimal infrastructure. Experiments usually tend to go overboard with allocated infrastructure. That is when it becomes imperative to be perpetually cognizant of the volume of data at hand and finding optimal ways to process more data.
Have right people and partners in place. Datasets and skillsets are in abundance, but the most critical piece of the puzzle is the mindset. Having the right partners in place ensures that the thought process is enhanced and the output is consumable.
Enable right communication to drive right consumption. The cognitive bias against an experiment or a new project can be removed only with consistent and engaged communication with the target audience. It is important to establish a communication channel with the right frequency easily consumable content.
Conclusion
Ever since the first group of analysts set out to comprehend a complete picture of a customer’s usage behavior and their attitude toward a product, the analysts relied heavily on surveys to understand customer perceptions with little or no visibility into how customers use or interact with products.
In reality, however, it is usage that influences perceptions that in turn leads to customer actions such as repurchase and recommendation. Thus, it is imperative to not look at behavior and attitude from individual lenses, but understand them using a combinatorial approach that links behavior and attitudinal data. The sooner organizations realize the significance of this exercise, the better are their chances of influencing and/or modeling a customer’s usage and brand perception. With advancement in technology and with right analytical partners in place this integration exercise would be one of the most worthwhile investments for any company looking to understand its customers better.