Maximizing Business Value with Generative AI

Have we ever seen something get adopted so quickly as generative AI (GenAI) compared to the past? Think about it: ChatGPT launched in 2022 and gained 100 million users in two months. In comparison, we have been hearing about AI for a few years, but the adoption rates of AI have varied from 25% to 35% (based on different research). This in itself shows the ease with which GenAI tools are being used to augment existing business processes, boost productivity, and provide benefits at most of the organizational nodes.  

Given that we’re roughly dealing with something whose age is quantifiable in months rather than years, the final goal and tangible solutions might be a little far, but the initial POCs and experiments have sparked innovation and are forcing people to think fundamentally differently about how businesses work.  

The global generative AI market is approaching an inflection point, with a valuation of $8 billion and an estimated CAGR of 34.6% by 2030. Even Gartner has placed GenAI at the peak of inflated expectations in their hype cycle. Most research reports highlight nearly 50-70% adoption of generative AI at least in the exploration stage.  

But people forget that generative AI is still AI; it needs the capability that any AI would have in terms of strategy, technology, tools, and people. So, it is highly recommended that executives think about this as part of their overall AI strategy. One of the ways the strategy has changed is the need for better governance, shorter roadmaps, and more use cases.  

If not treated this way, generative AI and LLMs would remain as isolated and siloed implementations without much interconnectedness and proper governance.  

For now, the three key facets of any GenAI playbook should be: Strategy, Infrastructure, and Roadmap. 

Generative AI Strategy

One of the biggest challenges we are seeing organizations face right now is that even before reaching the Proof of Concept (POC) stage, most lack a well-defined strategy and a set of identified use cases. It’s important to identify use cases with low complexity but high impact and low validation. This might be possible by creating a matrix of feasibility vs. business value each of them provides and choosing the right one based on your priorities. Based on which function to start, which department to work on, and the type of implementation you want, prioritize the ones that have maximum value. Some of the types of LLM implementations include: 

• Use LLM “as-is”  
• Embed LLM Into an application frame 
• LLM as chatbot 
• Use to generate training data for conversational AI  
• Embed LLM Into a workflow 
• Document retrieval 

Please note all these have different levels of complexity for implementation. This is where the analysis of current infrastructure and teams comes into consideration.  

Infrastructure

Even if organizations want to leverage generative AI for insights, they might not have the right data infrastructure and business processes in place when implementing AI for practical use. Data quality issues might impact the output, or your organization’s systems and platforms might have unique needs and capabilities. What’s more, there might be specific training practices needed for success. As everyone knows, GenAI is prone to bias and hallucinations; if there are data quality or training issues, the output would be manifested incorrectly, to put it lightly.  

Because of this, your organization should organize a team with some related expertise. A team consisting of business experts, engineers, and AI specialists would work. Obviously, you can’t expect people to have 10 years’ experience with generative AI, but something close to that – such as people on your data science team, with LLM or NLP or prompt engineering experience – would be beneficial. Such expertise will be needed while transitioning from individual inquiries to production-level applications. Accuracy will be critical, and training on extensive data sets will become foundational.  

One of the main reasons AI projects don’t take off is the lack of leadership, so you should also ensure someone can pilot the program.  

Roadmap for Deployment 

Once you have the pilot objectives and risks, you can consider the deployment approach. In the spectrum of build vs. buy, you can either consume generative AI embedded in applications like Adobe Firefly, Canva’s Editor, Hubspot Magic Assistant, etc., or build custom models from scratch with the help of open APIs (for any confidential or sensitive data).  

Since each of these approaches has its pros and cons from both a flexibility and cost point of view, if the objective of building an MVP is to quickly validate the hypothesis, it is preferable to embed APIs or use applications with GenAI. 

Moving from Proof-of-Concept (PoC) to Deriving Maximum Value 

As C-Suite leaders begin to understand GenAI, they are starting to uncover some questions: Which use cases will deliver the most value for my business? And how do we transition from a Proof of Concept (PoC) to full-scale implementation or enterprise-level deployment?  

A lot of the work currently remains in the PoC stage, though some industries are ahead of the curve, such as chatbots for HR and legal contracts, which have become relatively common. So, now what remains to be seen is how enterprises move toward widespread adoption by integrating GenAI into other business processes.  

To move from the PoC to the deployment stage, organizations must identify their strategy, as we covered earlier, as well as the use cases with high impact. Prioritizing these use cases based on their impact, cost, data readiness, and resistance to adoption is essential. Becoming familiar with the limitations and capabilities will also be important for decision-makers. A roadmap must be developed, and you must leave room for the possibility of failure. Once this is done, various PoCs and pilots can be launched, based on the problems an organization genuinely wants to solve.  

Additionally, transparency with your internal stakeholders is key. Communicate how these changes and the cultural shift will impact them, augment their capabilities, ensure productivity, and make them more effective. A change management program right from the start is necessary.  

The Way Ahead  

Going forward, the main differentiator will be how enterprises deploy foundational models responsibly. Given the cost to train and maintain foundational models, enterprises will have to decide how they want to deploy them for use cases. At the PoC level, a lot can be overlooked, but at the enterprise level, solving business problems will require a degree of certainty and decision-making around cost, time effort, data privacy, intellectual property, etc. If a foundational model lacks business context, it can result in outputs that make it challenging to identify accuracy. Accuracy is a critical factor – biases and hallucinations are real concerns because most foundational models are trained on extensive datasets. If there’s an inherent bias in that data, it will manifest in the outcomes.  

There is a huge paradigm shift taking place, one of the biggest in recent history, and it will impact every aspect of business in the near future. The way we see it is that at the organizational level, there must be an understanding of the power of foundational models in a frictionless environment – and this will be the key to success for many enterprises going forward.