How Intuit Developed and Rolled Out Custom Financial LLMs That Reduce Latency by 50% While Also Increasing Accuracy—And What Enterprise AI Teams Can Learn

In the high-speed world of fintech, speed and accuracy are everything. Intuit, which is behind the likes of TurboTax, QuickBooks, Credit Karma, and Mailchimp, has gained significant advantages in recent years from building out its own AI. Intuit created proprietary bank-trained financial LLMs and has realized an incredible 50% latency reduction while delivering improved accuracy by 5% for critical accounting workflows. This offers valuable insights for enterprise AI teams looking to improve their systems.

The Birth of GenOS: A Personalized AI Strategy

Intuit’s road to these innovations started with the launch of its Generative AI Operating System (GenOS). Unlike general AI systems, GenOS was tailored for financial use cases. It provides a unified framework that integrates different tools and algorithms, facilitating developers to create AI solutions with high performance and accuracy.

One of the elements that make GenOS unique is its in-house trained finance LLMs. These models are further fine-tuned on Intuit’s proprietary financial dataset to understand financial language and related tasks. This specialization is necessary because general-purpose LLMs may not recognize all nuances of financial terms and steps. By emphasizing domain-specific training, Intuit ensures its AI is not just smart but financially literate.

Latency Reduction and Accuracy Improvement

These custom LLMs have had a significant impact:

• Streamlined processing of financial queries, cutting in half the time it takes for a response.
• 5% improvement in accuracy, ensuring the delivery of information that is both trustworthy and contextually relevant.

These enhancements are most pronounced in platforms like QuickBooks Online and the Intuit Enterprise Suite. Customers report significant time savings and productivity improvements, thanks to the advanced AI features from GenOS. For businesses processing large amounts of financial data, such efficiency gains are game-changing.

The Role of Expert-in-the-Loop Integration

Intuit employs an “expert-in-the-loop” system, allowing humans to step in when the AI encounters complex or ambiguous scenarios. By combining human oversight with AI automation, Intuit ensures outputs are quick, accurate, and contextually relevant.

This combination of AI and human expertise is particularly useful in areas such as tax preparation and financial consulting, where precision matters. It illustrates a critical best practice for enterprise AI teams: marrying the scalability of AI with human practitioners’ nuanced understanding, guaranteeing better results while retaining automation benefits.

Lessons for Enterprise AI Teams

Enterprise AI teams can derive several lessons from Intuit’s experience:

1. Domain-Specific Training Is Important
   • Adapting AI models to the language and tasks of a specific industry greatly improves performance.
   • Generalized models may miss key nuances in unique workflows.
2. Balance Latency and Accuracy
   • Fast but inaccurate models lose user trust, while slow but precise models frustrate users.
   • Achieving the right balance improves user experience and operational efficiency.
3. Incorporate Human Expertise
   • Machine learning should be treated as an “expert-on-tap system”, not standalone.
   • Direct human input ensures AI is context-aware and reliable.
4. Continuous Monitoring and Evaluation
   • Use evaluation frameworks or leaderboards to track AI performance and gaps.
   • Essential in sectors like finance, where regulations and user expectations constantly evolve.

Scalable Infrastructure and Model Design for Efficiency

Intuit also invested in scalable backend infrastructure to support custom models:

• Efficient model architecture
• Optimized training pipelines
• High-throughput deployment strategies

These measures were essential for achieving latency reduction without compromising accuracy. By emphasizing infrastructure and engineering alongside model training, Intuit ensured its AI solutions could scale to meet enterprise demands.

The Future of AI in Finance

Intuit’s approach provides a playbook for transforming financial services:

• Specialized models + human oversight may become standard practice.
• Off-the-shelf models are no longer sufficient; customization, fine-tuning, and integration into workflows are crucial.

For enterprise AI teams, AI must be both intelligent and pragmatic. Deploying models that are fast, accurate, and contextually aware requires:

• Carefully crafted design
• Rigorous testing
• Sustained human oversight

Teams following these principles are more likely to achieve operational excellence, better customer outcomes, and informed decision-making.

Conclusion

Intuit’s development of custom financial LLMs highlights the importance of specialization, optimization, and collaboration in AI systems. By reducing latency by 50% and achieving higher accuracy, Intuit has shown how custom-trained models can dramatically improve performance in high-stakes, transactional environments.

Enterprise AI teams can leverage these insights to create AI systems that are:

• Smart
• Trustworthy
• Efficient
• Aligned with user needs

The combination of domain-specific training, human oversight, and robust infrastructure provides a blueprint for any organization aiming to scale AI in financial services—or other complex industries.