Only 5.2% of Tech Leaders Have Full Confidence Explaining Their AI Decisions, New FICO Research Finds

The survey expanded beyond financial-services C-suite executives to senior leaders spanning seven industries.

See more from StockTitan in Google Search and AI answers. Adds StockTitan as a preferred source · opens Google

Add on Google Not now

This year’s State of Responsible AI report examines untapped opportunities that would drive increased trust, transparency, and returns from AI deployments

BOZEMAN, Mont.--(BUSINESS WIRE)-- Global analytics software leader FICO (NYSE: FICO) today announced the results of its 2026 State of Responsible AI study, developed in collaboration with Corinium Global Intelligence. Now in its fifth year, the report tracks how organizations are adopting responsible AI practices, the early returns they’re seeing from deployments, and the remaining barriers to fully unlocking the technology’s full potential.

This year’s report surveyed 1,004 senior technology, risk, and data leaders, surfacing insights about AI and enterprise decision making. While organizations have poured billions into AI, the study reveals 71.5% of customer-impacting decisions are still made without it. This, coupled with the fact that only 5.2% of respondents are very confident explaining their AI-driven decisions today, signals the need for Responsible AI practices to improve trust and transparency. When asked which Responsible AI practice would make the biggest difference, respondents pointed most often to AI model development standards (71.9%).

“Most organizations can tell you where and what their AI decided. Far fewer can tell you why," said Scott Zoldi, chief analytics officer at FICO. "This points to a lack of responsible AI standards and techniques, which directly impacts organizations' willingness to use AI in customer-impacting decisions. As regulators and customers scrutinize how AI is used in decision-making, auditability and explainability are nonnegotiable."

Explainability Is Now Considered an Impact Driver

AI regulation is intensifying across the globe, with policymakers proposing new frameworks that mandate Responsible AI practices. Yet, only 5.2% of tech leaders are very confident explaining their AI-driven decisions to regulators and customers today, and more than a third (37.3%) are neutral or not confident at all. Left unaddressed, this gap could cost leaders the trust of regulators and customers alike.

Leaders recognize the importance of transparency, with most saying responsible practices such as model development standards (71.9%) and interpretable model architectures (67.9%) have the greatest potential impact for AI. They’re also actively investing in Responsible AI practices with most using bias detection tools (58.8%) and regularly auditing their AI models (53.5%).

Untapped Opportunities Remain to Maximizing AI’s ROI

While 85.1% say AI has met or exceeded their ROI expectations, most of today’s enterprise decisions depend on legacy processes: humans (33.3%), rules-based systems (21.4%), and standard analytics (16.8%). Bridging that divide requires responsible standards that earn trust for AI and give the technology a seat in decisions that still run on instinct and legacy rules.

Part of the problem is a lack of shared technical infrastructure to orchestrate responsible AI capabilities. Only 8.2% of organizations have a single shared AI deployment platform fully in place, while 71.2% have one partially in place. Almost all respondents (96.4%) believe closer collaboration between AI and technology leadership on a shared platform would unlock additional returns. Organizational education and shared vision is another barrier. Resistance to change (61.4%) and a lack of shared understanding of AI (58.4%) are the two leading barriers to internal AI alignment.

Looking ahead, 63.2% cite workforce skills falling behind the pace of AI adoption as their top five-year risk, while loss of control over AI systems ranks last (10.4%). This suggests leaders are most concerned about their people keeping pace.

“A year ago, there were serious questions about whether AI would deliver returns. Today, the top question facing executives is how to scale the technology enterprise-wide," said Lariza Carrera, Production Director, Corinium Global Intelligence. "Increasingly, the answer has less to do with the technology itself and more to do with the people running it.”

Practical Barriers Keep Agentic AI Stuck in Pilots

The promise of agentic AI is real, but the numbers show most organizations are still easing their way into the technology. Only 2.4% of tech leaders have broadly deployed agentic AI across customer-facing use cases, while 71.5% remain in early exploration or pilot phases. Security concerns (40.1%) and data integration or quality challenges (38.3%) are the most-cited barriers to achieving ROI with agentic AI.

Agents aren't where leaders expect near-term returns. Over the next 18 months, domain-specific language models (49.5%), Gen AI (48.7%), and interpretable model architectures (46.6%) top the list of technologies expected to drive ROI, as compared to agents (27.5%).

"Agentic AI is still in early experimentation mode, and that's exactly when the groundwork matters most," said Rachael Hadaway, VP of AI Product at FICO. "The leaders investing in security and data integration today are building the foundation for autonomy at scale."

Methodology

Conducted by Corinium on behalf of FICO, this year's report surveyed more than 1,000 senior technology, data, and risk leaders, including CTOs, CIOs, Chief Data Officers, Chief Digital Officers, Chief Analytics Officers, Chief AI Officers, Chief Risk Officers, and VP/SVP/EVP-level executives across analytics, data, and AI, spanning seven industries. The sample marks an expansion from prior reports, which surveyed C-suite executives in the financial services sector. The respondent pool is global, with two-thirds based in North America.

To download the full report, visit: https://www.fico.com/en/latest-thinking/survey-results/state-responsible-ai-2026

As a leader in predictive analytics and decisioning technology, FICO helps organizations turn their data into decisions they can explain, audit, and stand behind. Learn more at https://www.fico.com/en/fico-platform

About FICO

FICO (NYSE: FICO) powers decisions that help people and businesses around the world prosper. Founded in 1956, the company is a pioneer in the use of predictive analytics and data science to improve operational decisions. FICO holds more than 200 U.S. and foreign patents on technologies that increase profitability, customer satisfaction and growth for businesses in financial services, insurance, telecommunications, health care, retail and many other industries. Using FICO solutions, businesses in more than 80 countries do everything from protecting 4 billion payment cards from fraud, to improving financial inclusion, to increasing supply chain resiliency. The FICO® Score, used by 90% of top US lenders, is the standard measure of consumer credit risk in the US and has been made available in over 40 other countries, improving risk management, credit access and transparency.

Learn more at fico.com/en.

Join the conversation at x.com/FICO_corp & fico.com/blogs.

For FICO news and media resources, visit fico.com/en/newsroom.

FICO is a registered trademark of Fair Isaac Corporation in the U.S. and other countries.

View source version on businesswire.com: https://www.businesswire.com/news/home/20261007280824/en/

press@FICO.com

Source: FICO

Key Terms

domain-specific language models technical

A domain-specific language model is a machine learning model designed to understand and generate text for a particular field by being trained or fine-tuned on data from that field (for example, medical records, legal documents, or financial filings). It works like a general language model but focuses its vocabulary, facts, and patterns on the jargon, formats, and typical questions of one domain, which improves accuracy on domain tasks but can limit knowledge outside that domain and retain biases present in its training data.

interpretable model architectures technical

Model architectures designed so a human can follow how inputs produce outputs; they use structures or constraints that make the model’s internal logic transparent rather than hidden in many parameters. Examples include shallow decision trees, rule lists, sparse linear models, scorecards, or networks with monotonic or additive components; these architectures expose explicit rules, feature weights, or simple pathways that map features to predictions. A key limitation is that interpretability depends on the chosen representation and the user’s background — an architecture that is interpretable to a statistician may still be opaque to a lay reader — and design choices that improve transparency can restrict model complexity.