Build some, buy some, lock into none: Aditya Birla Capital's Pankaj Gadgil on buying AI

Aditya Birla Capital (ABCL) is combining AI tools it has built in-house with solutions from external partners, while looking to avoid dependence on any one model or platform, Pankaj Gadgil, head of ABCL’s digital platforms and payments strategy, and managing director (MD) and chief executive (CEO) of Aditya Birla Housing Finance Limited, told ET AI in an interview.

Gadgil said companies should retain the ability to build their own tools without waiting to recreate solutions available in the market.

Build and buy together

“It always has to be a hybrid between build and buy,” he said, outlining a strategy that draws on fintechs, big technology companies, consultants and internal teams.


The company has built a platform for sales and service assistance, with separate databases for its businesses and multiple large language models (LLMs) and components, he said.

The approach extends to voice technology, where Gadgil said the company has developed its own speech-to-text and other speech tools; they also use solutions built by their partners.

“So, it makes sense for you to ride the bus, not wait for building everything from scratch,” he said.

He cited two internally-developed tools – WorkIO, a tool for managing internal operations, and TraceIO, that replays a customer's journey through a digital platform, a task that previously required an outside vendor. Another tool, FinEngage, lets employees trigger a bot to contact customers at particular stages of a loan or insurance process, he said.

However, Gadgil said companies must look beyond their development teams for useful tools.

“You should have your eyes and ears open for solutions coming in,” he said.

Keep the platforms open

Keeping the option open to change technology is equally important, Gadgil said, because AI models and platforms continue to evolve. “Keeping our architecture and platforms open, I think, is very important rather than locking ourselves into something and then possibly rewriting it.”

Gadgil added the company uses both open-source and closed models.

Also read: What are open weight AI models and why are they dividing Big Tech?

That flexibility also applies to model size. He said smaller language models (SLMs) could handle some operational tasks companies give to LLMs.

“I think a lot will happen in the area of SLMs replacing LLMs, because some of the operational problems we are trying to solve with LLMs do not require them. SLMs can handle them,” Gadgil said.

He said most of ABCL's AI solutions sit on existing workflows, citing Salesforce and loan platforms such as Lentra and FinBox as examples of systems AI tools need to work with. Technology partners must provide application programming interfaces (APIs) that allow agents to read information and write back to these systems, he said.

The tools that cannot access the required information will lead to “dumb AI use cases”.

Also read: AI on bad data is a faster way off a cliff, warns Elastic CEO Ash Kulkarni

Data sets the limits

That dependence on information also explains why selecting an AI model is only part of the work, according to Gadgil. ABCL is working towards an enterprise-wide “knowledge IQ layer” to connect data with its meaning and relationships, including information in emails, PDFs and images, he said.

“We increasingly realise that to get greater impact from AI, most of the work that now needs to be done, I think, is on data,” he said.

Gadgil said scaling AI also requires employee adoption and governance covering tool access, accuracy, cost and incident management.