Statistical analytics are leveraged in a seemingly endless number of ways. They power dashboards, customer scoring, forecasting, and investment optimizations. They reveal patterns, outliers, and data observations that are out of place. They inform what has worked, what is expected to work, and how businesses can drive improvements. However, most statistical analytics are based on some form of average, which at times renders them less powerful then they could be,
- Coefficients in marketing mix models express the average relationship between sales and marketing across time, not the actual relationship for any specific subset of time.
- Customer segmentation is based on similarities between actual customers and average customers, averaged over a span of time, not the actual changing relationship between customers in any specific subset of time or geography.
- Test and control studies look for the average differences between test and control groups, and thus the wider the time period or group size, the more the results are averaged out.
- Order-to-cash metrics, such as costs per order, days payables and receivables are outstanding, and inventory days, are based on averages across orders, customers, and time periods.
- Queueing optimization is based on average arrival and service rates, not based on expected rates for specific situations.
- Correlations, covariances, and most other statistics are themselves expressive of the average relationship between variables.
Beyond these generalizations, from a value-add perspective, companies and decision-makers often only get a limited amount of business-driving value from analytics, in part due to some of the above issues.
The return on the investment in analytics is on average quite limited. For many companies, in fact, this return is so low that it can even be questioned as to why analytics should even be invested in. The simplest, yet most common, application of analytics is as a means to fuel dashboards. Less common, but further up the value chain, analytics are sometimes used to uncover interesting patterns and insights. Yet these do not always change quickly and thus these analytics are generally performed only on an ad hoc basis, and thus on average are not fully value-add. Although all businesses invest in analytics, the state of analytics in general is that it is not embedded into business processes. There are, of course, exceptions, such as credit card transaction analytics, which inform in real time the likelihood of fraudulent activity. Yet beyond several specific use cases, analytics are not embedded into decision making but instead used on an ad hoc basis.
Two issues thus arise: 1) how to make analytics less average, and therefore more impactful, value-add, and critical for companies; and 2) how to embed them into business processes so they indeed provide that impact, value-add, and critical insight for companies. The answers to these may be related.
By investing in better analytics beyond just average analytics increases its value beyond average. From a mathematical perspective, time varying coefficients improve value over the standard regression models, which generate time-invariant (average) coefficients. Time series customer segmentation shows how customer segments change over time, and which customers are shifting, instead of creating average-based fixed customer segments. Purpose-building the mathematics that underlie the analytics for a business problem, instead of using standardized analytic methodologies built on average analytics' results, will lead to more realistic and above-average value. Ironically, defining analytics' best practices, although often a selling point to those less familiar with analytics, sometimes reduces the value of analytics, since best practices, by definition, means practices that work in all similar situations, and thus are average. Purpose-building analytics explicitly to the nuances of each and every business problem — from a data perspective, from a statistical and mathematics perspective, and from a value-add perspective — means that instead of just using the standard average-based approaches, methodologies would be designed to address each and every nuance. When this is done, the value of the analytic output is far above average.
Once the analytics are improved, then the business case can be made to embed them into the business processes. When business decision-makers struggle to find deep value in their analytics, when the result of the analytics are themselves only average, when they do not provide insights not already known, when the opportunity for disruption based on analytics is low: in all of these cases the value of embedding analytics is not obvious. But when analytics are upgraded from the current state of the industry to explicit drivers of value, insight, and opportunity, then the investment in embedding them into standard business process is all the more likely to generate a return far above average.
The ultimate value of analytics is to upgrade from nice-to-have to need-to-have. Analytics, like all investments, should ideally always provide a return. Companies spend huge amounts of money, time, and infrastructure for their data. But too often companies do not leverage the majority of this data beyond reporting. On average, the value is at most average. When they leverage the industry standard of average-centric analytics, the return tends to be underwhelming. However, when companies leverage their data asset to the fullest extent with purpose-built analytics, then they see a return on both the analytics investment and data investment, and on the investment in embedding analytics into business processes.