The goal of analytics should be to improve enterprise value (e.g., revenues, margins, retention) through the generation of relevant, accurate, and appropriately related inferences for the right people at the right time in the right way. The promise of machine learning (ML) for analytics has been the promise of more. More relevant and accurate inferences, and more timely delivery—all gleaned through deeper machine learning.
Though ML can point to many small analytics victories, it has failed to achieve the larger promise of improving analytics and value for the enterprise as a whole. On the contrary, ML has contributed to an increase in analytics siloes thus making it even harder for senior executives to understand what their company knows. Even within a single ML application, knowledge that any person would see as obviously relevant to an ML decision (that exists in the analytics data store) may be invisible to ML (e.g., knowledge of a person’s spending habits as being relevant to an ML decision about fraud).
The Achilles’ heel for ML is its inability to incorporate unanticipated sources of information (which may be the output of other ML). Whereas humans are great at dynamically incorporating all kinds of sensor information and prior knowledge in order to make sense of their current surroundings, ML is not. There is no practical way to combine or reconcile new channels of information (e.g., a new ML algorithm or expert opinion) with an existing ML application in order to improve accuracy or confidence. This kind of multichannel reconciliation capability is not ML per se but is the missing link needed to enable ML to fulfill its original promise to analytics.
In this session you will learn
- Why the performance boost that ML alone can give to analytics is limited
- Why introducing ML to Analytics has worsened analytics fragmentation and made it harder for executives to introspect on what their company knows
- How major companies and leading researchers are trying to solve the problem
- What you can do to provide a multichannel learning environment for your ML
- Where to start