When GenAI is a must, good to have, and when is it an overkill

The rise of Generative AI (GenAI) marks a transformative shift in technology, ushering in a new era of agentic AI. While there is a growing following of advocates excited about its potential, there are also detractors—those who reject it outright or remain hesitant to adopt it. As a result, GenAI use cases are often confined to corporate innovation labs or experimental initiatives.

Because of scale, ROI and the strict necessity of accountability and governance, selecting the proper use case that GenAI can solve is an important consideration for organisations. How can organisations identify if a use case requires GenAI or can be solved using other methods of pattern recognition, predictability, prescription, diagnostic or description?

GenAI: To be or not to be?

Organisations are already deploying use cases such as improving customer experience, boosting employee productivity, accelerating process optimisation, enterprise search and enabling field agents with assistance in the local language.

While most of these use cases are relevant to GenAI usage, there are instances where organisations are not fully aware of the power of the technology and its implications in full-blown form. Many rush to implement GenAI programmes due to board pressure often overlooking where it truly benefits the business and where it may not be necessary.

Business value through GenAI can be measured through direct-value use cases such as new product design, customer voice and feedback actions and fraud prevention by detecting unseen anomaly cases. Indirect value use cases would be insights generation or democratisation of insights to the front-line staff, leading to decisions and sales.

Efficiency creators are the most popular, such as saving time through document parsing, summarisation, SOP search engines and smart contracts. The scaled usage of GenAI cases can lead to multiple combinations of these value drivers, incremental value to the top line and the bottom line of an enterprise.

Identifying the right use case

The first step is for enterprises to make the right choice of use cases to be solved by GenAI. The use cases should be scanned through a spectrum of GenAI dependency to solve the problem, ranging from complete dependence on GenAI to considerable GenAI intervention requirement and minimal or no GenAI intervention required.

· Complete dependence: Content generation, code generation, conversational interfaces such as chatbot and search, knowledge management and automating repetitive tasks

· Considerable dependence: Segmentation, classification, recommendation systems, anomaly detection, intelligent automation and pattern recognition

· No/Minimal dependence: Prediction, forecasting, BI-early warning systems, automation of core processes, etc.

If you take your current or business use cases, including those driving revenue, efficiency and data-led decision-making, and check whether they merit GenAI application, you would be surprised to find that use cases work best in combination.

The choice framework – A must-have for organisations

This paves the way for the need for a framework that assists organisations in classifying use cases based on GenAI dependency and accordingly investing in such solutions. Having a choice framework helps organisations determine when GenAI should be used for a use case by analysing their problem across three pillars –

· Differentiability index

· Ease of implementation

· Impact

The differentiability index indicates the level of uniqueness of the problem, whether it extends beyond the capabilities of traditional AI models and the novelty of use cases. This has three sub-dimensions of rating: high, medium, and low. High differentiability can be when:

- Data for the use case are in different formats, such as multi-modal.

- Reliance on the company's own data is less and external data is more. The core task of the use case is to generate something new, such as a draft report or parts of a financial report, conversation spiels and hyper-personalised images.

- Use cases can be solved by ML techniques, but the time taken will be too high.

Ease of implementation involves the level of capabilities and time needed to implement the solution. Use cases that require the least time and are backed by existing technology and strong internal capabilities will take precedence over the ones that involve significant time and resources and non-availability of technology and capabilities. This also involves the ease of estimating the costs during the production stage which can be easily modularised and removed if they are not functioning in a desirable state. 

Impact indicates the size of the market opportunity and strategic importance for the network.

GenAI is not expensive at the experimental/Proof of Concept (POC) stage when the usage is limited to a small team or to a pilot region. However, scaling the GenAI solution requires investment. Hence, it is pertinent to determine if the use case has the greatest strategic importance and would cater to a large audience.

For example, forecasting is a complex process. It cannot be solved using GenAI because, first, it is a central team of planners, and second, it is a complex use case that has a history and not something that can be generated.

Way forward

The choice framework is a great starting point for organisations to determine use cases that require GenAI as a full-fledged solution versus those that can be combined with traditional AI methods to make them more efficient with GenAI.

It gives a full spectrum of GenAI implementation, helping organisations choose the right ones and removing unwanted GenAI hype. Only when the right use cases are explored will it lead to scaling from POC, enabling organisations to derive value from their investments and realise maximum ROI.

By Moumita Sarker, Partner, Deloitte India