AI in decision-making

AI can be applied to decision-making using data through the building, training and deployment of AI models to the production environment.

A typical AI model deployment process for decision is an understanding of the business domain coupled with the understanding of data, followed by the preparation of data for building AI model(s), training and evaluation of the model(s) and deployment. 

Creating REST APIs using Flask Webserver, Dockers to reuse the AI models, Kubernetes orchestration platform that enables fault tolerance, auto-scaling, load balancing, rolling service updates among others, model monitoring & recalibration for concept drift, data drift and assessing the model to decide recalibration need – are other essential elements of how AI can be used for decision making using data in the context of digital transformation.

Most of the cloud services providers viz. GCP, AWS, and Microsoft Azure have AI services built-in on their offerings and can be used on a pay-as-you-go model. But then building a robust and resilient data pipeline is crucial as multiple compliance and regulations around data privacy and cloud administration are in place by respective authorities that one needs to consider for a successful AI strategy formulation.

While AI-based DevSecOps, and multiple COTS products on the same are available, LCNC platforms like Pega have built-in AI decision engines into their platforms that could be leveraged in SaaS or PaaS models depending on the specific customer’s need.

How does AI help in decision-making:

A decision may be categorized as simple, complicated, complex or chaotic. Based on the type of decision an AI model assesses, it aids a human decision-maker or supports a decision or augments the human knowledge and capability using predictive and prescriptive analysis. Mostly for simple decisions, may AI decides autonomously, but nevertheless, human intervention is planned for handling exceptions and errors in the decision. 

Few use cases like payroll processing that is simple and predictable with less margin of error may be left to AI for decision automation. 

A transaction monitoring system in a bank, where erroneous transactions can be classified based on a classification and regression or any other supervised machine learning model for a human analyst to pick high probability fraud transaction – is AI-based decision augmentation. 

Complex decisions like a diagnosis of say mouth cancer based on multiple image analysis and historical medical data analysis of patients using AI is a decision support and here finally a doctor decides whether a case is positive or not.