AI’s data appetite: Can traditional strategies keep up?

While Generative AI is being increasingly favoured for its many benefits, including higher productivity, better revenue generation, and improved customer experiences, there continue to be many challenges that arise in its adoption.

The real question is: are organisations ready for the AI age?

In a fireside chat at YourStory’s India Tech Leaders’ Conclave 2024 held in Bengaluru on June 21, 2024, Shivir Chordia, Managing Director, Couchbase India & SAARC, offered insights on what organisations must be prepared for while deploying GenAI and how the technology can evolve with the demand and supply change.

The conversation was moderated by Shivani Muthanna, Director, Strategic Partnerships and Content, YourStory Media.

Preparedness to adopt AI

Chordia highlighted how at Couchbase, there was a global survey including 1,000 participants who adopted GenAI in some form. There was much excitement around its use, but certain challenges crept up.

“One was data hallucinations, which is a key element. Sometimes, the results may mislead decision-making, so it's important to choose the right kind of data models you are working on,” Chordia said.

The other challenge is around security, since a lot of the data and models could also be proprietary information. While it's important to build more modern applications, it's equally essential to consider the costs since GenAI adoption is expensive.

Organisations need an extremely flexible model to work with, where rapid changes can be made in the application. Also, there's a requirement for an engine that can do real-time analytics across the board.

“Not everyone would need large language models (LLMs) or running large compute infrastructures. You need things that can run on the mobile, or small models running on EDGE …[they] are going to be more interesting, and a larger use case in India,” Chordia added.

If these critical elements are put together while building a GenAI strategy, the problem is solved. From flexibility at the application level to a multimodal database that can give capabilities far beyond purpose-built models, all aspects are important.

A differentiated strategy

As per a survey by Couchbase, almost 80% of CIOs move budgets from existing infrastructure projects to play in the GenAI space. However, it's important to understand that if one tries to pick a traditional architecture or application and attempts to modernise by infusing it with AI, the project will fail.

“Many of the traditional databases aren't meant to run GenAI at scale. The customer experience will decide what kind of models you need, what are your outcomes, and eventually, what kind of infrastructure you need to build,” Chordia said.

The days of purpose-built data are gone–it is the world of multi-model, real-time databases that are going to fuel AI.

“Very specific smaller models are enough to do the job. That's where it's important to keep optimum infrastructure cost,” Chordia explained.

At this juncture, everyone is attempting to play with databases that have vector capabilities. Nobody looks below the hood to say that they can scale but at a certain cost, Chordia said.

“You want to start looking at areas where you can say, ‘can I knock two or three databases and make them slightly more multimodal in nature and reduce my cost yet not compromise on scale’?”

While there are certain aspects that can be addressed, solving for data complexity and volume is something that will never happen. India is a heavily populated country where the consumption of data on mobile phones or social media is tremendous.

“If you start building out even a simple customer scenario, the amount of feeds and the information you get are possibly thousands. In that sense, the traditional application at best has 100 parameters. The learning is: can we make data work for us as individuals and organisations?” Chordia asked.