Résultats du processus de synthèse de l'article arXiv : 1702.01742v1

Résumé : The biggest problem with the methods of machine learning used today in business analytics is that they do not generalize well and often fail when applied to new data. One of the possible approaches to this problem is to enrich these methods (which are almost exclusively based on statistical algorithms) with some intrinsically deterministic add-ons borrowed from theoretical physics. The idea proposed in this note is to divide the set of Key Performance Indicators (KPIs) characterizing an individual business into the following two distinct groups: 1) highly volatile KPIs mostly determined by external factors and thus poorly controllable by a business, and 2) relatively stable KPIs identified and controlled by a business itself. It looks like, whereas the dynamics of the first group can, as before, be studied using statistical methods, for studying and optimizing the dynamics of the second group it is better to use deterministic principles similar to the Principle of Least Action of classical mechanics. Such approach opens a whole bunch of new interesting opportunities in business analytics, with numerous practical applications including diverse aspects of operational and strategic planning, change management, ROI optimization, etc. Uncovering and utilizing dynamical laws of the controllable KPIs would also allow one to use dynamical invariants of business as the most natural sets of risk and performance indicators, and facilitate business growth by using effects of parametric resonance with natural business cycles.