Generative AI – Where do I Start?

Generative AI has caught the imagination of the world. The adoption of the technology has been unprecedented. From the looks of it, while the technology is still in its hype cycle, it shows promise. It is the next step in the evolution from Data Analytics to Machine Learning to AI to Gen AI for now. The ecosystem has been already in development for years now. There are a plethora of Gen AI tools already flooding the market. All major enterprises have been investing in the technology either for their use or to create Gen AI products and services.

The explosion of Gen AI is posing a “where do I start” dilemma”. With a buzz around Gen AI, the real understanding of how to build a Gen AI strategy and execute it is fuzzy. It took me a while to figure out where to start. Reading reports, talking, and attending seminars has not answered my questions. So, the next best alternative, talk to the experts in AI technology and understand what it can do to your own business. The logical steps that one can put in place are as follows:

• Start using the Technology – Gen AI is very intuitive to use, which is evident from the fast uptake. The reason is that all the infrastructure needed to run Gen AI already exists. The cloud, software-as-a-service, application programming interfaces, app stores, and other advances keep lowering the amount of time, effort, expertise, and expense needed to acquire and start using new information systems. While I am writing this article Grammarly AI is already improving my structures and removing errors. The image in this article is generated by Stable diffusion. In addition, learn about Prompt engineering (Got recently introduced to this term. Believe me, it is not as esoteric as it sounds). It is very useful.

• Gen AI is new, and organizations are experimenting with refining individual processes. The future would be where Gen AI could be involved with wider Business processes interacting with each other. An important start that can be made is to assess the organizational “DATA”. “Good Data” is the key ingredient for any AI application. “Good Data” is Accurate, Complete with all needed parameters, Interpretable, and Accessible when needed. It is widely accepted now that Data is a key asset for any organization. AI-powered systems consuming that data will drive new competitive advantage. While this is known strategically, implementation is key. Data has to be viewed not only as an input to AI algorithms but also as output from the operations in a usable form. Automation of manual processes requires considering the future use of that data in AI models. Early-stage enterprises usually do not have mature processes. Emphasis is on quick go to Market with Minimum Viable Products, and rightly so, the focus is on survival, cost saving, and flexibility. Entrepreneurial spirit takes precedence over structured data gathering. Data quality is critical. An early data capture strategy will be useful.

• Application in your Business environment – It is crucial to keep in mind that just because a technology is hyped does not mean it is relevant. Assess your business and take a view of what areas of your business Gen AI can make an immediate impact. It could be a revenue generator, Customer experience enhancer, Efficiency generator, cost reducer, or generally makes life easier for you and your team. The human aspect of this technology is not talked about a lot. The most obvious areas may be the most impactful ones. Or you may get insight into areas that could be a source of competitive advantage.

Building Gen AI for your business: Once comfortable with using the technology the next step would be to assess building one for your business. The following steps are recommended to get going:

Brainstorm – Brainstorm with your team on some possible Use cases that could give an immediate edge in your business. Select one that can be done quickly.

Build a Proof of Concept – Data scientists create trained, tested, and iterated models to consistently improve results. This is where it becomes daunting for an organization with limited Data science capabilities. In today’s world, one organization can’t do everything. A good Partner ecosystem is essential. A good partner Data/ AI company could be a great partner. It could be a mutual win for both with you providing the data and the Partner building the model and executing it. In all likelihood, your Partner may also be looking for references. By working together, you may help each other, and commercials and risks may not be huge to start with.

Build a Minimum Viable Product – The next step would be to build a basic product without extensive features or bells and whistles. That would keep the efforts and investments low. Any enhanced features can always be added later in subsequent phases.

Finally put the model in Production and reap the benefits. The steps above may seem to be simple their execution can get complex, especially in larger organizations with multiple stakeholders. Like any transformational project the larger the scope the more complex it is to get off the ground. Early experiments could be short, simple, and easy Proof of concepts for larger projects subsequently. Organizations need to learn first.

Do you see it the same way I see it? Comments welcome.

References:

The Data Problem Stalling AI: Gregory Vial, Jinglu Jiang, Tanya Giannelia, Ann-Frances Cameron: MIT Sloan review. (winter 2021)
How to Train Generative AI using your companies data: Tom Davenport and Maryam Alavi –: Harvard Business Review