Founded in 2017 by Ganesh Gopalan and Ananth Nagaraj, Gnani.ai is a Bengaluru-based startup that claims to facilitate over 1 million conversations daily with its product line, which is meant for contact centres.
The company has over 100 customers in banking and financial services, insurance, telecom, automotive and healthcare industries.
In May, the startup launched a series of voice-first SLM (small language models), trained meticulously on millions of audio hours of proprietary audio datasets and billions of Indic language conversations.
“What we’re doing is something different, like a fusion of voice and text models. It’s a multimodal model but right now we are focused on voice and text,” Ganesh Gopalan, the CEO of Gnani.ai, told AIM in an exclusive interview.
So far, the company has built a series of five models designed for the banking, finance, security and insurance (BFSI) sector.
Gopalan revealed that the models are multilingual. In the US, it supports both English and Spanish, whereas, back home, the model is designed to support 12 Indian languages.
“We also plan to launch a model designed for the automotive industry because we have a lot of customers in this industry. Healthcare is going to be a sector in the future,” Gopalan said.
Building for the Edge
The size of SLMs developed by Gnani.ai are relatively smaller compared to an LLM like GPT-4 or even smaller LLMs like Llama-3 7 billion parameters. Gopalan believes the size of SLMs will come down even further, enabling the deployment of them on the edge.
“In the future, we will deploy these models on the edge because the size is also coming down drastically. We believe the solution to many enterprise problems isn’t always found in the generalised 100 billion+ parameter models that companies often tout.
“These models are excellent for generic applications but may not always address specific enterprise needs effectively,” he said.
Moreover, many enterprises hesitate to adopt third-party models due to concerns about their proprietary nature, uncertainty about the training data used, and security apprehensions. Running a model on the edge where the customers’ data are solves a lot of these problems.
“We think running models on the edge will be a lot cheaper. So, at some point, all these models will be on edge. And that’s something that we are working actively towards,” Gopalan said.
What Gnani.ai considers its strength is the ability to quickly fine-tune the model based on enterprise data and make it production ready.
“It’s one thing to have an SLM for the BFSI sector, but the real value to a company is when you have a model built just for their data. So that’s what we do. We take our model to enterprises and help them build on top of it. We provide them with necessary tools that can quickly launch the model based on their data,” Gopalan said.
AI Agents are Coming
Today, AI agents are believed to be the next big iteration in the AI cycle. Previously, Kailash Nadh, the CTO of Zerodha, have said the prospects of having AI agents are very high but perhaps not in a nice way.
While Gopalan agrees AI agents will be the next big leap, he adds that Gnani has been building AI agents for over four years.
“We automate processes that are done by contact centre agents through our voice bots. We built AI agents that helped one of the largest banks in India in collecting over a billion dollars in the last six months,” Gopalan said.
The bot handled payment reminder calls to customers, ensuring timely payments and helping them find the appropriate payment methods.
“For a US client, our AI agents are assisting the contact centre agents by providing instantaneous answers to customer queries as conversations unfold,” he added.
AI agents will also change how contact centre operations work. The future is multimodal, according to him. “ For instance, if you encounter an issue with your laptop today, calling the contact centre requires providing a tag number and describing the problem, which can be challenging.
“Instead, you can show a video, and an AI bot can analyse the visuals, identify the problem area, suggest solutions, and present options to the agent. The agent, using human intelligence, can then assess the problem and provide a solution,” he said.
IVR, Biggest Impediment to Customer Experience
However, there is a lot of apprehension about AI agents making contact centre agents redundant. Gopalan believes there is a long way to go before AI makes call centre agents redundant, if it ever happens.
“We currently deploy AI bots for numerous use cases. However, systems are still evolving for scenarios that necessitate human intelligence or empathy. In these areas, AI bots are not ready yet,” Gopalan said.
Moreover, he believes interactive voice responses (IVR) could be the biggest impediment to customer experience.
“No one enjoys navigating through automated menus and pressing buttons when contacting a contact centre with an urgent issue,” he pointed out.
Gopalan believes AI will have to bring an end to IVR before it eliminates jobs of call centre agents. Contact centre business is one segment which is being impacted by generative AI, and discussions around AI making contact centre agents redundant is widespread.
“Another reason contact centres will remain relevant for a long time is that it’s not just about having people to answer calls. Many companies lack the integrations with CRM systems, ticketing tools, and other necessary infrastructure. Additionally, a significant portion of customer service knowledge resides within the minds of contact centre employees,” he concluded.
