AI in Business Application
Anand Banerjee
Transformation Strategist ? Digital Strategy ?Creating a smooth experience for customers across channels | Head of Engineering - Digital & MarTech | Continuous Improvement | Techno-commercial Negotiation | AI for CX
In previous article, AI for Digital Success, we looked at how the 2x2 matrix of Domain Knowledge and AI Complexity should drive the prioritization framework for journeys that are suitable to have AI based solutions. Let's dive into looking at business applications of AI and some its limitations.
To revisit the explosion of AI in the last few years, it would not have been an understatement that while Data & AI technology innovations are at the heart of this AI explosion , cheaper data storage and availability of relatively cheaper computation power have also fueled the growth of AI systems.?
To simply put, an AI systems’ complexity is based on data needs (quality and quantity). More complex the AI solution (let's read it as a model) , the more the need for data.
AI and Computation Power
Additionally, the process of collecting data, analyzing it and making decisions has been the age-old process. This might have been manual or semi automated and quite likely an expensive process but it did exist always. What has changed, is the development in machine learning techniques.?
Now the question arises, why does an AI system require so much computation power???
To understand this let's look below:
This clarifies and connects the dots on technology advancement in AI and its current mass availability of usage, is a lot tied together on availability of cheaper computation and storage.
AI in Business Applications
With availability of AI systems more readily available to businesses, its usage can be categorized into following business scenarios:
领英推荐
Limitations
With availability and use of AI solutions, organizations and especially leaders responsible for rolling out these solutions, need to be vigilant of few existing limitations:
Data Needs
Advanced AI models require a lot of data. Not only quantity matters, quality is quality important. Additionally, it also should be looked at for the allowable use of data. Privacy impact is quite an important puzzle for leaders to be on top of while deploying such a solution.
Explainability
Business should be able to trace the result and explain the outcome to both its internal and external stakeholders. Imagine a fantastic deep learning AI model that has processed and rejected a mortgage application. If the customer or regulators ask about such decisions, businesses should be able to provide the decision traceability of the model..
Generalizability?
Ability to generalize the solution. Most cases, today’s AI systems are trained on a specific process and by using a certain type of data. AI systems cannot use this learning and transfer it to work on another process. This requires a retraining with the required set of data. Thus any changes in the business process needs to be continuously reviewed to ensure validity of the model. An AI system trained in a chess game cannot leverage or transfer its skills to learn another board game (unlike humans), at least as of today.
Bias
It is linked with the Explainability case as mentioned by adobe. Model is as good as the data that was fed to it. Any existence of bias would require constant monitoring and ability to explain the outcome.