FICO

Decision logic in FICO Platform can run on rules and optimization alone. Decision Modeler, the successor to Blaze Advisor, executes business rules, and FICO Xpress solves optimization problems without machine learning. Models drive much of what FICO sells around that logic.

Falcon scores payment fraud with neural networks trained across a consortium of more than 10,000 institutions, the FICO Score is itself an analytic model, the Analytics Workbench builds machine learning models for lenders, and FICO Standard Models offer ready built AI for fraud, risk and customer engagement. FICO Assistant brings generative AI into the platform to help build decision components.

The Focused Foundation Model, introduced in September 2025, adds a Focused Language Model for compliance and text tasks and a Focused Sequence Model for transaction sequences, each returning a trust score with its output.

Lenders set the strategies FICO Platform executes. At UnionBank, low risk applicants pass straight through while higher risk segments get extra verification, and more than 80 percent of card and personal loan applications are processed without manual handling. For FICO's own models, Scott Zoldi, its Chief Analytics Officer, describes each model's permitted use, monitoring logic and alert thresholds being recorded on a blockchain during development, with no release until they are in place.

Alerts go to human operators, who can adjust strategies, disable a latent feature or fall back to a secondary model. The Focused Foundation Model's trust scores are meant for setting risk thresholds on its outputs. Where a lender's strategy hands a case to a person is left to the lender's configuration.

FICO builds interpretable neural networks in which each latent feature combines no more than two inputs, so every relationship the model learns can be monitored. The monitoring configuration, with expected firing rates and limits for each latent feature, is fixed during development and recorded on a blockchain that FICO says is covered by a US patent. Before a strategy goes live, FICO Platform can simulate thousands of combinations as a digital twin to compare expected outcomes.

The Focused Foundation Model attaches a trust score to each output as a quantitative measure of reliability, and FICO reports a 38 percent lift on compliance adherence tasks and more than 35 percent on transaction analytics, without the test method. FICO Assistant claims more than 90 percent accuracy for the components it generates, also without a method. There is no validation package or mapping to a named supervisory standard for FICO Platform models.

UnionBank of the Philippines moved about a million customers from Citibank's local consumer business onto FICO Platform in nine months. It now processes more than 80 percent of card and personal loan applications straight through, onboards customers in 5 to 15 minutes and reached 50,000 new accounts a month at peak. Grab Finance built 22 decision workflows across six Southeast Asian countries in under eight months and raised credit offer eligibility by nearly 50 percent for its users.

UnionBank won a 2025 FICO Decision Award and Grab Finance a 2026 one, with judges from outside FICO. The annual report puts financial services at 92 percent of revenue and FICO Platform annual recurring revenue at $263.6 million in September 2025.

The privacy policy, effective July 2026, says FICO generally uses only depersonalized data to develop and update its analytic models. For Scores and Software it acts as a service provider, so a lender's own privacy terms govern its data. Pooling is part of the design in fraud, where Falcon learns from a consortium of more than 10,000 institutions. FICO does not say whose data trained the Focused Foundation Model, and gives no retention schedule for client data used in model development.

The privacy policy separates FICO's own processing from the work it does for clients. For Scores and Software, FICO acts as a service provider and the client's policy applies. FICO certifies under the EU, UK and Swiss Data Privacy Frameworks, uses standard contractual clauses where required, names an EU representative, and states retention periods for its consumer service.

Its public subprocessor list names 28 service providers and 19 affiliates, with Amazon Web Services hosting in twelve countries including Singapore, Indonesia and India. Affiliates in the region include Eighth Intuition in Malaysia and Fair Isaac entities in Singapore and Thailand.

The Cloud Security Alliance STAR registry lists FICO with a Level 2 STAR Attestation, renewed in February 2026, and a CAIQ self assessment from April 2026, both downloadable by anyone. FICO's own trust center shows ISO, PCI, SOC 2 and GDPR badges without naming the standards or report types behind them, and sends buyers to their client partner or the security team for detail. For the FICO Analytic Cloud it describes a DevSecOps model, penetration testing and a security operations center, and offers a Secure by Design white paper.

FICO Scores sit inside US mortgage regulation. Fannie Mae and Freddie Mac require them for eligible loans, subject to validation by the Federal Housing Finance Agency under 12 CFR Part 1254. Since July 2025 lenders may choose which approved score to submit, according to FICO's annual report. The software business has no comparable standing, and FICO claims no supervisory status for FICO Platform. The same report flags the EU AI Act, most of whose provisions take effect in 2026.

Model governance at FICO is built around a development standard whose adherence is recorded on a blockchain. Zoldi describes the monitoring it enables as supporting ethics and stability testing, since nonlinear relationships between latent features can produce biased decisions, and interpretable neural networks keep each learned relationship visible for that monitoring. FICO's score research has aimed at people with thin files, through FICO Score XD and the UltraFICO Score. FICO gives no subgroup results, disparate impact testing or fairness metrics for its credit models.

The blockchain record gives FICO and its clients an immutable trail of how each model was built, what it may be used for and how it is monitored, and trust scores flag less reliable outputs. FICO's annual report describes the Fair Credit Reporting Act as governing the bureaus, the firms that furnish data and the users of reports, so correcting the data behind a score runs through those parties rather than FICO. Contract terms for FICO Platform, including any warranty or indemnity for model errors, are private.

FICO's public subprocessor list names 28 service providers and 19 affiliates, with Amazon Web Services as the public cloud in twelve countries and Google and AWS providing text to speech for customer communication services. No outside language model provider appears on it. The Focused Foundation Model is FICO's own, trained on task specific data sets, with patent applications filed on trust scoring, training and monitoring. A strategic collaboration agreement with AWS, signed in May 2025, covers FICO Platform, which FICO says is built and deployed exclusively on AWS.

At UnionBank, FICO Platform combined bureau data with alternative data to score customers with thin files, and it integrated Citibank's systems after the acquisition. Strategies built in it can be deployed into card issuers' account management systems, and Loan Pricing Optimizer and Credit Line Optimizer are listed on AWS Marketplace. Grab Finance runs workflows on ride frequency, merchant revenue and payment history from its own apps. Connections to specific core banking or loan origination systems are not named.

Software is sold as SaaS or for installation on a customer's own infrastructure, according to FICO's annual report. FICO Platform itself is built and deployed exclusively on Amazon Web Services, and the subprocessor list places hosting in twelve countries, including Singapore, Indonesia, India and Japan. The platform launched in Indonesia in 2023 once AWS's arrival there met data sovereignty requirements, with Superbank as the first customer. Grab Finance's workflows build in local credit, banking and data protection rules across Southeast Asia.

FICO lists per score prices for US mortgage lending. Under the direct license program announced in October 2025, scores distributed through tri merge resellers cost $4.95 per score plus a $33 fee per borrower per score when a loan closes, or $10 per score with no loan fee. FICO Platform and its applications are priced privately, and AWS Marketplace lists Loan Pricing Optimizer and Credit Line Optimizer as private offers only. A lender in Southeast Asia has no listed rate for the software.

Financial services produced 92 percent of FICO's revenue in fiscal 2025, and its products are split across originations, customer management, collections, fraud, loan pricing and credit line management, with separate material for card issuers, mortgage lenders and collectors. Scores are sold in more than 40 countries. Asian customers include UnionBank and Bank of Commerce in the Philippines, Superbank in Indonesia and Grab Finance across the region. FICO employs 3,811 people in 28 countries, about 40 percent of them in India.