Generative AI can transform customer service: Sendbird's Shailesh Nalawadi

"From what was primarily a one-way brand to user communication, today it is a seamless transition into a two way conversation. And it doesn't always need to have a live human on the other side," says Nalawadi. Earlier, a bot could never make sense of a company policy and handle customer interactions to the level of a real human, which isn't the case for bots trained using large language (LLM) models.

In fact, an interaction with a customer service executive could take a lot of time to reply or address customer queries as they cannot possibly learn all of these details, which could change every few weeks depending on the offer or the specific detail of the incident.

If this sounds easy, training LLMs with specific details is also not easy and often will require assistance from a service provider like Sendbird, says Nalawadi. The LLMs are often trained on large corpus of internet data and is a generalist and is not a specialist and has to be trained to service a company's specific needs. The models will probably will need to be trained on the customer profile details, different nature of products and services of a company, past customer transactions, orders, customer service interactions and tickets raised.

"An LLM bot is cleverer than the bots from five years ago, because it is very conversational in nature. It is able to understand semantically what it means and is not just doing key words. It is still a superb conversationalist, but trained on that corpus of data, and taking that corpus of data and training only on that data and not on anything else, that is a technique, that's an enhancement on top of large language models that currently Sendbird is actually focusing on," says Nalawadi.

Sendbird is a unicorn and has raised a total of $221 million from investors such as Iconiq Capital, Tiger Global, SoftBank and Steadfast Capital among others. The company has been offering LLM-based services to its customers since April this year. While the company used to focus on running chat platforms for customer engagement, today the Generative AI models is throwing open a larger opportunity in powering customer service requirements.

The development of ChatGPT and Google's Bard can probably help a dating platform's customers with very specific connections based on their interests which can be curated with extensive rounds of conversations and filtering of potential matches, which was hitherto impossible earlier.
Nalawadi says that there is a lot of hype around LLM models such as ChatGPT as the technology is in the initial phase of development and people are not clear about what are the boundaries of its capabilities.

"It wasn't clear to the creators of LLMs, that it could solve that math problems. It is almost like, they have birthed this creature, and this creature is kind of starting to learn, which is surprising for computer scientists. Because our general computers are very deterministic systems," adds Nalawadi. While it is starting to get beyond the capabilities of the designers to anticipate, and its capabilities are expanding. "Which is why there is equal parts up hype and equal parts hysteria. The fear of the unknown," he says.

However, the expectations has to be trumped as the hype around past advances in AI has died down. For instance, in the healthcare sector, there were expectations that the software trained on a million x rays or blood samples can predict cancer cell growth and so on. With hindsight, these has been proved to be at best an assistant to a doctor. This has been the case in other fields as well.

"Companies will have to roll out these features first and work with service partners and need to see whether this works for their industries," says Nalawadi. According to him, Indian companies are more adventurous in terms of pushing the boundaries of customer experience.