How do you identify AI use cases?

AI is the buzz word right now and people are too excited to use AI in every possible opportunity. Like how the "Big Data", "Data Science" Hype was there in the market , right now it is the AI or Gen AI hype. Whatever Google does my company also can do that is the belief among Tech people who try to convince executives on the possibilities and the magic of AI.

But if you just take a step back ward and see how do you identify use cases of AI when there is a limited budget within a company to pursue AI opportunities. Two dimensions can be used

1. Business Impact
2. Feasibility

Business impact is usually from the business leaders and the Feasibility part comes from the Tech people. Even you can consider many factors for the feasibility dimension, i just restrict to the broader ones. Data , Infrastructure/Architecture and Ethics.

Data:

Data is paramount for an AI use case. Compared to a traditional BI use case which can be done through a waterfall model most of the times. An AI use case is more iterative in nature. In most AI project the CRISP-DM methodology is followed as shown below.

As shown above the steps of business understanding and data understanding is an iterative one. Unless you get a proper understanding of the data, don't proceed further. Ask the following questions before proceeding further.

• What data do you need for your use case?
• Is this data available? If so, what exactly do you have (tables, media files, text)?
• Do you have access to this data?
• Who can access it?
• Does your company own it, or do you have to buy it?
• Do you have legal authority to use this data?

Once the above questions about data is answered then it is wise to come up with a prototype or MVP. Again depending on the type of problem we are trying to solve the data requirements might vary.

Infrastructure/Architecture:

Following is a high level overview of an AI system.

Like any other IT system, you could decide on "Make or Buy" decision. The entire system could be on a single platform, or you could build few components on your own and leverage a few services from a provider or just get the infrastructure from a cloud service provider and build the system on your own. All of these choices depends on the resources available, time to market and overall fit into the existing ecosystem.

The architectural decisions include whether you go for a serverless approach or a containerized solution. Also how is the user consuming the solution. Architectural decisions all depend on the use case requirements.

Ethical Considerations:

AI ethics, we refer about AI services that are highly specialized tools, not self-aware entities that think about the consequences of their actions (and are thus exempt from responsibility for themselves). Rather than assessing the AI service itself, therefore, we need to assess the use case or application for which AI technologies are being adapted. And it is the people who are responsible for them who ultimately bear the responsibility and must be held accountable.

In almost every AI use case, potential ethical considerations must be addressed both before and after the solution is implemented. In some cases, these issues are so serious that the use case cannot proceed. In other cases, you need to make changes to the use cases to be morally safe. And in still other use cases, the risk of discriminating against people on a large scale is very low. The point is that as a business leader or someone trying to think about developing AI solutions, you are responsible for going through this thought process before you try your first prototype in the wild.

Prioritized Usecases:

Once you have the business impact and feasibility, Just map them as shown in the following buckets.

Champions

Use cases that have high impact and high feasibility. These should be your first priority and drive your ML roadmap.

Quick wins

Use cases that have similar high feasibility to champions, but lower business impact. Because these use cases are often less complex, they are easier to implement. However, they can be a good showcase and demonstrate the value of ML and AI applications. Think of this as an area where off-the-shelf AI services can be used, for example.

Research cases

These areas have a high impact but are generally not feasible in the short term. Either you do not have the data or you are not allowed to use it (yet), or certain technical limitations exist. However, because these use cases potentially have such a big impact on your business, do not neglect them; keep them on your radar.

Reassess later

These use cases have the potential to consume enormous resources without contributing much business value. They should be put on the back burner and reevaluated later. Use cases with low impact and low feasibility can turn into quick wins as technology changes. For example, some time ago, creating an automated priority ranking for customer support tickets was complex. However, with new developments in AI services, this could suddenly be feasible overnight, making it a potentially interesting quick win for your business.

Reference:

AI Powered Business Intelligence, Tobias Zwingmann Oreilly

Disclaimer: Extracts from the book are just for knowledge sharing purpose.