(US & Canada) Sudip Ghose, VP, Enterprise Data, Analytics, and AI at Equifax, speaks with Dennis Allen, Director of Security Programs, Strategy, and Risk at Stratascale, in a video interview about key initiatives, the collaboration between human and AI agents, and incorporating security measures.

Equifax Inc. is an American multinational consumer credit reporting agency headquartered in Atlanta, Georgia.

Speaking of key initiatives, Ghose lists real-time analytics as a prime topic, as these enable faster reaction times. He adds that it sits well with his financial planners when it comes to delivering quick results that they can compare and correct if needed. The other initiative Ghose refers to is around AI, ML, and GenAI. He highlights building the AI portfolio using Vertex AI, which includes mastering the platform with an AI team comprising engineers, legal, and governance people.

Additionally, Ghose mentions that the organization is also developing applications with data scientists and hinges on executing AI/ML projects to drive better organizational outcomes. Regarding GenAI, the focus is on deploying AI agents internally. Referring to Looker’s cloud-based feature “Looker Core,” he reveals the intention to adopt a similar approach as the organization gradually transitions to the cloud.

Further, Ghose states that the organization will incorporate agentic experience in its visualization platform itself. Google search and conversational agents are also being built internally, which is a massive area of attention, he adds.

Next, he emphasizes the expected collaboration between human and AI agents. Explaining with an example, he says, in case of any revenue dispute, a human agent would have to go through the back-end procedures to resolve it. If an LLM is purpose-built based on the well-connected data in the data lake, it would significantly reduce the investigation time from hours to minutes.

Adding on, Ghose states that Equifax is very close to building it and is recording the experiences of human agents for better work performance of their AI counterparts. He remarks that the AI agents can either work as co-pilots for human agents or can be rolled out to users to solve queries.

Then, Ghose sheds light on the process of incorporating policies of security and data governance in his data analytics mission. Stating that it spans two ways, he first mentions addressing the security concerns surrounding LLMs and prompts for which building a secured platform is under process.

Secondly, Ghose affirms that Equifax is carefully building its models on Google Cloud. Appreciating Google’s security and privacy measures, he stresses the search giant’s assurance of data security and accessibility, and the guarantee that Google will not use the company’s data to train its own models.

Building models internally is relatively secure; however, the organization is still finessing the data security and access security parts, says Ghose. He considers implementing a RAG (retrieve, augment, and generate) structure to secure model data assets. For instance, if a salesperson asks a question intended for finance, it will be blocked by the system with polite responses.

Still exploring the approach, Ghose notes that the structure would be organized at the site or function level, with functions having their applications separated by models. In conclusion, he states that the overall plan is that RAG will assess and control model access, though the company is still refining the intricacies.

CDO Magazine appreciates Sudip Ghose for sharing his insights with our global community.