‘Crossing the Chasm’ – Bridging the gap between Machine Learning & Business Decision Making

Machine Learning, Deep Learning and in general Data Science have made tremendous strides over the last few years in a business setting. Having said that, in my view, there exists a significant gap in utilizing the output of Machine Learning for business decision making due to certain limitations. This talk will discuss those limitations and introduce “Business Simulations” as a powerful tool to be leveraged along with machine learning techniques to help business stakeholders confidently execute data-driven decisions.

The talk will be in 2 parts:

Part 1: Decision Gap – What is it? Why does it exist?

  1. Models are approximations of the real world. While data science practitioners approach model building from the data perspective, business decision-makers have a process-centered view of the world built over a period of time through experience & knowledge (mental models).

  2. Machine learning & deep learning models created from data have certain limitations. They are:

    • Black boxes – Many of the powerful algorithms are black boxes which result in business users not trusting the output of these models

    • Lacks measurement of uncertainty – Classical Machine Learning algorithms are mathematical formulations and they rely on optimization to find out model parameters, resulting in point estimates

    • No causal structure – Machine Learning predictions are based on correlations and does not provide causation. Business decision making is very much about causation “What happens if I intervene”? and hence that is a limitation

    • Lacks Process View – ML does not take into account non-linear relationships among entities, multiple actors, delayed feedback loops etc. and hence acts on an over-simplified version of reality

Part 2: Business Simulations & Data Science – A Powerful Combination

This part of the talk will introduce business simulations as a solution to the decision gap explained in Part 1 above. We will briefly touch upon the 3 types of simulations (viz. Agent-Based Modeling, Discrete Event Simulation, and System Dynamics) and discuss the scenarios which leverage the powerful combination of simulations & data science techniques. Karthikeyan will demonstrate a sample business scenario (using Python package called PySD) in which machine learning output can be utilized to drive simulations leading to better results.



Karthikeyan Sankaran (Karthik) has over 20+ years of experience in the software industry with specific focus on Business Intelligence & Advanced Analytics. He is a passionate data science practitioner with the ability to effectively use technology to solve business problems. Currently in his role at Latentview Analytics, he works at the intersection of Business, Data, Math and Technology to answer business questions using an approach that is simple, pragmatic & effective. Karthik has spoken at many conferences, conducted data science bootcamps and is a regular guest lecturer in campuses. Karthik finished his engineering at IIT Madras and is an MBA from IIM Calcutta.

Buy Ticket
Social media & sharing icons powered by UltimatelySocial