BOZEMAN, Mont. – FICO, the company well known to credit unions for its credit scores, is taking a different approach to artificial intelligence than the large tech companies betting on ever-bigger models, according to a new report.
With the launch of its Focused Foundation Model for Financial Services, the company says smaller, specialized models offer more reliable, efficient tools for banks and lenders operating under strict regulations, Forbes said.
Unlike general-purpose large language models from OpenAI, Google or Anthropic, FICO’s Focused Language Models(FLMs) and Focused Sequence Model (FSM) are designed for precision and compliance. “They are not primarily intended for use in open-ended conversational chatbots,” Dr. Scott Zoldi, FICO’s chief analytics officer, told Forbes.
1,000 Times Less Computing Power
The company told Forbes the domain-specific design allows its models to run with 1,000-times less computing power than general models, while still outperforming them on tasks relevant to financial institutions. In one case, a client in Asia Pacific saw a 38% gain in compliance adherence for customer communications, while U.S. clients reported a 35% improvement in credit-card fraud analytics over two years, Forbes reported.
‘Institutional Truth’
FICO’s strategy leans on what it calls Trust Scores, a metric showing how closely an output matches an institution’s required standards. Unlike typical AI confidence scores, these reflect alignment with “institutional truth,” echoing FICO’s roots in credit scoring, according to Forbes.
“Banks set thresholds based on their tolerance of potential hallucinations or damaging outcomes,” Zoldi told the news outlet.
The system is built from inception for regulated environments, with explainable AI features, observability layers, and compliance support for frameworks such as GDPR, Forbes said.
The report explained that models are trained in two stages: first with broad financial knowledge, then localized for regional regulations with synthetic and client data. FICO manages the fine-tuning process itself, including red-teaming, rather than leaving it to clients.
“By narrowing scope, FICO also avoids many of the ethical and operational pitfalls seen with broader AI platforms, including hallucinations and auditing gaps,” Forbes said. “The models are tightly integrated with the company’s existing analytics systems, drawing on decades of credit-risk experience.”
Looking Ahead
FICO plans to expand multilingual support and introduce new base models over the next two years, Forbes said.
Zoldi told Forbes the company also intends to add more autonomous “agentic” functions and to extend its auditable AI blockchain, creating traceable records of every training update and deployment.
“It is critically important for language model providers to walk the walk, not just talk the talk,” Zoldi was quoted as saying.
