A Data Science Central Community
Originally posted on sctr7.
Analytics professionals should keep a persistent eye open for opportunities to create cross-functional insights from data. Insights can frequently be leveraged across multiple business domains. However, business silos can restrict ‘forest for the trees’ visibility. To the degree analytics has cross-functional exposure in the organization, for instance via a C-level champion and an Analytics Center of Excellence (ACE), leveraging such insights is easier.
Janus was the two-faced Roman god of beginnings and transitions. Often associated with doors and gates, Janus was also represented in ports and trading. Janus recognized the notion that all beginnings are an ending and that all movements and change leaves something else behind. The depiction as having two faces, one facing backwards and one forwards, symbolized that all things have a hidden dualism. In business, we can adopt this paradigm by recognizing that all core business functions address multiple domains: robust fraud protection also is a platform for credit approval and customer service, risk management provides insights into strategic opportunities, clear financial planning can clarify new market opportunities and open innovation paths.
Whereas a targeted analytics inquiry addresses a particular business domain, by taking a cross-functional business perspective analytics professionals can often identify insights which benefit the broader organization. Analytics professionals should seek opportunities for optimizing the benefits of an inquiry by reframing results in facing domains.
The paradigm of risk management, often represented via the corporate Enterprise Risk Management (ERM) function, is instructive in this regard. From the perspective of risk management, each risk presents an opportunity, and each new initiative creates risks. The art and science lies in quantifying and optimizing these flip-sides of business. Analytics, as a discipline dedicated to clarifying both risks and opportunities via data science, can be instrumental in providing such insights.
Unfortunately departmental silos often constrain the leverage of an analytics inquiry to a specific business domain. An analytics initiative may be funded by a marketing department and the stakeholders may have little (or even adversarial) interest in assisting product development, customer service, financial planning, or production planning.
Knowledge is power; unfortunately agency interests often align to tight business silos, meaning insights can be protected by those less eager to share. This is unfortunate as the additional overhead to realizing wider analytics insights is often quite modest once the proper data has been collected and analyzed. Not everyone believes or understands the notion of economic optimization via cooperation, especially when short-term targets and bonuses are primary motivators.
By framing analytics insights from cross-domain perspectives, greater systemic insights are created, for instance: how shifting consumer interests indicate a potential new product opportunity, how operations planning can be optimized to meet shifting market demand, how innovation initiatives can be budgeted to ensure financial targets are met, and how customer service performance can be improved through customer, product, and market insights.
For example, we take the case of credit risk reduction – automated mortgage application assessments via machine learning. As a starting point we collect a large dataset of past mortgage applications and known resulting defaults. Analytics is benefitted by a large, rich dataset to start with, which improves domain understanding, efforts to focus a data model, and, thereafter, model training and testing. Typically we would want to enrich base transactional data with demographics and supplemental 3rd party data to inform the assessment.
After a primary assessment of the data and data cleansing, we would typically conduct unsupervised analysis to examine natural categories amongst the domain data. This allows for the collapsing of variables into measured categories, where feasible and appropriate (according to accepted statistical and algorithmic approaches to validate the identified categories). For instance, we might apply cluster, factor, or principal component analysis to identify natural categories amongst mortgage applicants. These approaches are associated with particular statistical tests which can establish validity and context for the identified categories.
Subsequently, we would be interested to apply supervised techniques: to build a machine learning model from a dataset of known mortgages, including defaults. With a sufficiently representative dataset, a machine learning model with demonstrated predictive power can be established (via testing using multiple ML algorithmic approaches utilizing training, validation, and test sets). The resulting model allows identification of potentially at-risk customers as well as an automatic assessment of new applicants. Such a model also gives insight into value-creating customers – those stable and profitable customers and who may be eligible for new products.
Such insights quickly benefit more than the credit approval domain: they inform product strategy, customer service, budgeting, and sales & marketing. With each business silo, there exist flip-side benefits to analytics inquiries. Many examples exist, for instance:
What is not to like? Investments in analytics insight, with proper positioning, can be leveraged in multiple domains. What often stands in the way is ‘silo thinking’. If a particular business unit funded or launched an analytics inquiry, they may be reluctant to share the insights.
This is where having an intra-organizational analytics function can be invaluable. Whereas research has shown that may companies develop analytics functions in disparate business units (i.e. finance, manufacturing, sales & marketing, and customer service), by establishing an Analytics Center of Excellence (ACE) supported by a C-level champion, cross-silo insights gain appropriate exposure and visibility.
With the great interest in ‘big data analytics’, there is a tendency to launch cross-analytics as an IT initiative, focused on technology and engineering. For instance, there is a great interest in establishing Hadoop repositories to store large sets of data. However, without tying the technology perspective to cross-organizational business processes to unite people in business silos, such efforts are quickly moribund. Successful analytics depends upon a unified people-process-technology context. This can only come from suitable organizational visibility and exposure. Those organizations that establish an ACE and appoint a C-level champion to improve the success of such efforts.