Gila: Banking Blends Generative and Deterministic AI

AI-powered virtual assistants are becoming standard across sectors such as retail and hospitality. In these environments, fully generative models are often acceptable because occasional hallucinations pose minimal risk. 

According to the Gartner 2025 CIO Agenda, the top technology that banks will increase investment in for 2025 is generative AI, with a 90% rate of CIOs reporting spend of up to 39% more than the previous year.

This planned investment increase brings the need for tailored AI solutions into sharp focus.

Dan Michaeli, Chief Executive Officer at Glia, explains his company has taken a unique approach to developing virtual assistants by blending the understanding of generative AI with deterministic responses designed for regulated environments.

This hybrid method begins by harnessing the deep comprehension capabilities of generative models. As Dan notes, the "magic of these models" lies in their ability to understand intent. 

Their system achieves more than a 92% understanding rate across all banking interactions routed to their virtual assistants. This foundation allows conversations to be reliably contained and guided without losing context.

Understanding and response must be separated

The key principle behind the company’s approach is the separation of understanding from response. While the AI interprets customer intent with high accuracy, the final response is controlled entirely by the financial institution.

This ensures "a 0% chance of hallucinations,” says Dan, not a reduced chance, but none at all. In a regulated sector, this level of certainty is essential.

Concerns about the risk of new technologies like Agentic AI are top of mind for banking leaders, according to the American Bankers Association, and a deterministic model could directly address those fears when paired with generative AI.

By allowing institutions to fix the response layer, virtual assistants can deliver safe, compliant and predictable interactions.

Overcoming AI implementation and integration hurdles

Despite substantial planned investment, successful AI implementation remains a major obstacle. According to MIT, NANDA 95% of generative AI pilots are failing, which is primarily attributed to flawed implementation and integration challenges rather than the technology's capabilities.

A controlled approach could help sidestep these common pitfalls. This is critical as the CSBS Annual Survey of Community Banks finds the number one risk they face, besides increasing community bank competition, is cybersecurity, followed by costs of technology implementation.

A system where the AI’s understanding continually improves while responses remain controlled helps mitigate both implementation and security risks enabling higher containment rates and greater automation.

Automating high-volume, low-value interactions

Dan highlights the operational pressure that frontline banking staff face. 

Large call volumes, long queues and poorly designed IVR systems often force employees to rush customers off the phone. This reduces the quality of service and increases stress for staff and customers alike.

AI offers a path to reverse this effect. By automating high-volume, low-value interactions, routine questions, simple account requests and standard updates, staff can spend more meaningful time with customers who require assistance. 

Rather than detracting from the human experience, automation becomes a tool to enhance it.

Dan argues that to "feel more human", institutions must automate the repetitive parts of the customer journey.

Once those conversations are handled by the virtual assistant, human agents can focus on empathy, problem-solving and complex queries that truly benefit from a personal touch.

Reframing AI adoption fears

Many financial institutions hesitate to adopt conversational AI because they fear losing human connection. Yet, as Dan explains, AI can actually strengthen that connection. 

By removing the pressures caused by volume spikes and inefficient call routing, staff gain the time and energy to engage more thoughtfully with customers.

Ironically, the very technology institutions worry will make them feel less human is the same technology that enables a more human experience. 

In a sector where accuracy, compliance and empathy all matter, intentionally blending deterministic AI, grounded in a strong understanding and controlled responses, with the power of generative understanding, is emerging as the model that balances innovation with responsibility.