Taking Models to Data and Decisioning to Models!

Artificial Intelligence is radically changing the way organizations make decisions across their business. Previously, once the decision that had to be data optimized was identified, the traditional approach involved 3 steps:

1.      Extracting the relevant historical data and labeling them for model consumption

2.      Building probabilistic models on the historical data to predict future behavior

3.      Based on these predictions, make decisions such that it improves business outcomes; i.e. manage credit losses etc.

However, AI is now fostering an integrated approach to decision making. First, there was the advent of big data and data lakes, which allowed the models to be built without overly spending time on step 1. Predictive models can be trained from multiple data sources without overly structuring the data in the first place. Now, with the advancement of machine learning/ AI, we can build models in a way that optimize the decisions algorithmically; i.e. merging steps 2 and 3. The business outcomes to optimize could range from managing risk control, hyper personalization to increasing operational efficiency and process automation.  

Let’s take an example of lending & pricing decision for prospective new customers for a retail finance company. AI driven approach would let us work directly with business outcomes we are trying to optimize. Assuming the business objective in this case is to increase net returns on the underwritten portfolio whilst maintaining default risk and customer indebtedness under a certain level. AI lets us build an integrated machine learning model (lets assume a deep learning model) that directly predicts for the new returns should a customer be acquired and optimizes the pricing (interest rate) on the lending product to maximize net returns whilst still constraining for overall customer indebtedness and default risks.  A well designed deep neural network with simulation can deliver this.

As AI evolves, it is certain to collapse the step approach of data>> model>> decision making to integrated self-learning applications that can run seamlessly across the enterprise. The pre requisites for this however are a solid governance framework that ensures data/ models and decisions follow the law and ethics of the land.

At AIYDEC, we continue to be excited about working towards this BHAG (Big Hairy Audacious Goal)!