Why Most Financial Chatbots Are Broken by Design — And What It Actually Takes to Fix Them

2026-04-22 · Source: HackerNoon · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, long

Summary

Vijay Kumar Sridharan, VP of Software Engineering at Goldman Sachs, focuses on bridging the gap between technically capable chatbots and systems that genuinely serve user needs. His work, spanning over fifteen years, emphasizes designing conversational AI around user mental models rather than company information architecture. During the COVID-19 pandemic, Sridharan led the development of a production-grade multi-channel chatbot at OneMain Financial, which successfully handled increased customer demand by implementing robust intent classification for edge cases, graceful ambiguity handling, and context management. He also spearheaded the "CheckCapture" project, challenging assumptions about technical feasibility by building solutions alongside existing architecture. His research extends to adaptive chatbot systems, LLM-based IVR, and verification layers for LLMs in regulated financial environments, aiming for systems that anticipate user needs and ensure accuracy.

Key takeaway

For AI Product Managers and CTOs evaluating conversational AI deployments, prioritize system architecture that aligns with user mental models and incorporates robust failure handling. Your teams should focus on building verification layers and context management to ensure reliability and trust, especially in regulated environments where accuracy is non-negotiable. This approach will drive higher user adoption and mitigate regulatory risks.

Key insights

Effective conversational AI prioritizes user mental models and robust system design over raw NLP capabilities.

Principles

• Design for user journeys, not isolated inputs.
• Challenge assumptions of technical unfeasibility.
• Build verification layers for LLM accuracy.

Method

Develop classification models trained on edge cases, implement graduated confidence thresholds for graceful failure handling, and design persistent user modeling for adaptive intelligence across sessions.

In practice

• Analyze user interaction logs for failure patterns.
• Implement clarifying questions for ambiguous intent.
• Consider building alongside existing architecture.

Topics

• Conversational AI
• User-Centric Design
• Financial Chatbots
• Large Language Models
• Context Management

Best for: AI Product Manager, CTO, VP of Engineering/Data, AI Engineer, NLP Engineer, AI Architect

Related on AIssential

• When a chatbot runs your store — Large language models struggle with reality adherence, making them unreliable for autonomous transactional roles.
• We Don’t Have an AI Problem. We Have a Context Problem. (And MCP Exposes It) — Unstructured context, not model limitations, is the primary bottleneck in scaling and maintaining LLM-based systems.
• AI Is Not Your Friend. It’s Designed to Make You Feel Like It Is. — AI's conversational design can induce user delusions, fostering an illusion of friendship and trust.
• Building Reliable Agentic AI Systems — Agentic RAG systems require robust context and harness engineering for reliability and trust in regulated environments.
• US startup advertises ‘AI bully’ role to test patience of leading chatbots — AI chatbots frequently lose context and hallucinate in sustained conversations, leading to significant real-world risks.
• Agents need more than a chat - Jacob Lauritzen, CTO Legora — Complex AI agents require high-bandwidth interfaces and structured human collaboration to overcome context rot and verification challenges.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by HackerNoon.