FICO unveils tailored AI models to boost trust in financial sector

FICO has launched a new foundation model tailored for financial services, designed to improve the accuracy and transparency of generative AI applications within the sector.

The product suite, known as the FICO Focused Foundation Model for Financial Services (FICO FFM), includes the FICO Focused Language Model (FICO FLM) and FICO Focused Sequence Model (FICO FSM). Both models have been designed to address specific challenges in financial decision-making, compliance, and fraud prevention by providing domain, data, and problem-specific solutions.

Domain-specific design

The FICO FFM was built as a response to the limitations of general-purpose large language models (LLMs) within highly regulated sectors. Unlike LLMs trained on vast volumes of general world knowledge, the FICO FFM employs tailored data and training for each business problem, with a focus on auditability and compliance requirements. The organisation states that its focused models use up to 1,000 times fewer resources compared to broad LLMs, which results in models that are more cost-effective to train and adapt.

"The focused foundation model represents a practitioner's approach to GenAI in financial services, moving beyond trying to refine universal knowledge models," Dr. Scott Zoldi, Chief Analytics Officer at FICO, said.

FICO FFM enables enterprises to use small language models built for their specific business problems, significantly helping to mitigate hallucinations, provide transparency, auditability, and adaptability. The model complies with regulations through the transparency of data and a decreased risk of hallucinations through Trust Scores and business owner-defined knowledge anchors. These domain-specific models can result in 38% lifts in compliance adherence use cases and more than 35% lifts in world-class transaction analytic models, in areas such as in fraud detection.

Improving transparency and trust

FICO FLM is trained using business-specific and task-driven datasets, which the company says allow it to notably lower the occurrence of hallucinations compared to more generalised systems. These models are smaller in size yet intended to be more accurate and accessible, which the company says will benefit financial institutions seeking a lower barrier to entry for generative AI.

The FICO FSM is designed to extract insight from transaction sequences, identifying relationships in data that traditional analytics might miss. The model supports real-time accuracy improvements in areas such as payment fraud detection, risk assessment, and next-best action forecasting, and enables the identification of complex transactional patterns that were previously costly or difficult to detect.

Megha Kumar, Research Vice President of Analytics and AI Analyst at IDC, commented on the impact of these focused models:

FLMs are essentially similar to SLMs (Small Language Models) and these are transforming how GenAI is used in financial risk management and compliance by providing highly accurate, domain-specific insights and reducing misinformation. Built on curated data and Responsible AI principles, these models are becoming essential tools for institutions that require precision, transparency, and scalable trust.

Responsible AI and trust frameworks

Both FICO FLM and FSM incorporate Trust Scores, which FICO claims offer a quantitative assessment of model output reliability. These Trust Scores enable organisations to set risk thresholds, operationalise model outputs with confidence, and oversee ongoing risk more efficiently in AI-powered decisioning.

The company's recent survey, conducted with Corinium Global Intelligence, highlights industry opinion on trustworthy AI. According to the survey, 56.8% of respondents said Responsible AI standards are the most crucial element for fostering reliable, consistent business returns in financial services, while 40% viewed GenAI and LLM innovation as a key driver of return on investment.

Dr. Zoldi further elaborated on the position of LLMs and domain-specific models in enterprise settings:

While LLMs have been a significant enabler in the rise of AI – and a major disruptor across industries – their adoption has been more limited in financial services. In highly regulated environments where accountability and precision are critical, current LLMs often lack the reliability, transparency, and governance required for enterprise deployment. FLMs, designed with domain-specific data, auditable model focus, and task specific auditable data, enable development of GenAI that is built for purpose, auditable, and able to be monitored and constrained to business guidelines.

Patent filings and ongoing development

FICO has filed several patent applications in relation to its FLM and FSM technology, covering core techniques such as trust scoring, model training, content tracking, real-time monitoring, and modelling of transaction sequences. These patent filings indicate FICO's ongoing investment in the principles of Responsible AI for financial services applications.