Big Data: Way to Customer Centricity?

To be strong competitors, businesses aim to achieve higher levels of performance by having a deeper insight of the customers. Big data for business is the opportunity to achieve customer centricity by systematically extracting the most valuable analytical insights.

­­­Four analytic imperatives for next generation learning:

A. Design and automate smart experiments that enables causal prediction:

Causal relationships are the most valuable for customer centricity as companies can better understand the individual customer behavior and sensitivities, and predicts likely responses to a specific offer or treatment.

1. Action-Effect Model is regarded as an important component of decision modeling where effects of alternative treatments are evaluated. Moreover, well constructed action effect models help in predicting effects of treatments on customer behavior.
2. Overlapping – essential for finding casual relationships: Overlapping alternative treatments across similar customers provides an opportunity to compare potential outcomes and casual effects. FICO employed the propensity scoring technique to find useable overlap in the available data. In this case, the likelihood of a customer being assigned to a specific treatment was predicted, unlike other familiar application to predict propensity the customer will buy.
3. Designing smart next generation learning experiments: To expand overlap, next generation boundary-hugging test design technique was used in the case study. It does work by randomly assigning individual customers to a treatment, not to a strategy. Customers close to decision boundaries were more likely to receive the treatment and customer farther away were less likely to receive the treatment.
4. Automation stimulates the learning cycle, by rapidly generating high-overlap data using machine-learning algorithms, and balance cost of testing. In the case study, simulation is being used in finding “sweet spot”, balancing learning speed and investment.

B. Analyze and learn from customer behavior on-the-fly

Analysis of streaming data is based on FICO’s self calibrating models under which usual behavior is compared with unusual behavior of peers to detect outlier to score them on the degree of deviance from the standard. A customer in outlier range could be committing or experiencing fraud. On the other side, an individual could be experiencing prosperity. Therefore, outlier model can also point opportunities to be offered by the companies to its customers. This analysis is based on real time updates so that detection of outliers is based on current distributions to facilitate better understanding of outlier behavior.

1. Outlier zone is not constant as it dynamically readjusts the range of outlier’s values against the moving range of normalcy
2. Human expertise and machine learning: Human expertise is required in developing unsupervised self -calibrating outlier model as in determining highly predictive customer characteristics and to incorporate them in the model requires deeper human insight. Therefore, when a supervised model is trained to identify usual and unusual values for those characteristics, machine learning takes place
3. Multi-layered self-calibrating outlier model: In this model, the nodes – separate self-calibrating outlier model, and the set of related characteristic variable hooks up in different ways. Then, the separate scores are combined into single scores, which incorporates the multiple views, and therefore increasing score accuracy and usefulness
4. Informing steaming analytics with other insights: rapidly employed, the self calibrating outlier model can be informed by other insights. Case study states that “this can be fed with outputs from neutral network models and the result of link analysis and semantic scorecards”. Therefore, adding the flexibility leads to a variety of analytical solutions for cross-selling and pre-delinquency treatment.

C. Get the machine learning / human expertise balance right

Identifying correlations in the big data is only one part of the search. Understanding and linking the unstructured data to more structured, meaningful information needs human expertise.

1. Machine learning: speeds up by crunching through big data to test large volume of characteristics variables. FICO found that a Tree Ensemble Model was nearly twice as effective in predicting the insurance fraud as compare to traditional models.
2. Human involvement: to make insights useful in operations needs human involvement. Therefore, to incorporate domain knowledge into predictions and customer treatments requires human expertise
3. Text Analysis: It has been demonstrated that, the text analysis is more effective when folded into structured data models. This model includes

• Data cleansing and standardization
• Extract text features
• Transform them into numerically based customer characteristics

D. Turn every customer touch into an opportunity for more service and learning

By entering into each interaction, the company knows who the customer is and has made the decision about the range of suitable treatments. During the interaction, the company makes additional decisions on the fly based on customer responses and new data. Moreover, these capabilities enable the business to transform every customer touch point into an opportunity for more service.

Conclusion:

In nutshell, Big data yields its forth V – Value, when it is systematically employed to learn more and more about our customers. Accomplishing this depends on judicious use of machine learning and human expertise that can apply big data analytics in new and more competitive ways.