AI & Technology: The Adoption Gap

By FRISS —

Almost universally, practitioners understand that AI represents a step-change in fraud detection capability. The proof of value exists, and the threat landscape that demands it is well understood. Yet the operational reality remains: the overwhelming majority of insurers are not yet equipped to respond. The cost of that gap is not abstract; it is being paid in claim losses, manual inefficiency, and competitive disadvantage every single day.

What makes the adoption gap particularly striking is that this is not a story of ignorance. The issue is not awareness, but execution. The barriers are real, the path is understood, and early adopters have already demonstrated what is possible. What is missing in most organizations is the collective will and the structural conditions required to close the distance between knowing and doing.

Where Organizations Actually Stand

• 9% – Fully implemented AI or ML in fraud detection (this is the only bracket that represents genuine, scaled operational protection)
• 36% – In pilot / partial deployment
• 36% – Planning to implement
• 19% – No current plans

The adoption picture breaks into four distinct groups. Thirty-six percent of insurers are in the pilot or partial deployment phase, meaning concrete steps have been taken but the technology is not yet operationalized at scale. A further 36% are planning to implement, indicating a strategic decision has been made but execution has not yet started. Meanwhile, 19% report no current plans, and just 9% have fully implemented AI or machine learning (ML).

The combined 72% who are planning or piloting can sound like progress until you examine what it actually means in practice. Planning without execution is not protection, and a pilot that never scales is not a defense. These organizations are not yet receiving the fraud detection benefits of AI; instead, they are at various stages of approaching the starting line. The only benchmark that matters operationally is the 9% who have fully implemented. That number is a clear indictment of the distance still to travel, particularly given the threat timeline that will be laid out.

There is also a hidden risk in the planning cohort that deserves naming: organizational inertia dressed as due diligence. When planning stretches indefinitely, when pilots run without defined scale-up conditions, and when budgets are studied but never committed, the planning stage becomes a comfortable alternative to action. The window for preparation is already closing, and the organizations that remain in planning mode longest will find themselves with the least time to close the gap when the threat accelerates.

What is Blocking Adoption and What the Barriers Reveal

• 29% – Budget constraints
• 27% – Data quality issues
• 18% – Lack of internal buy-in
• 16% – Regulatory / privacy concerns

Budget constraints (29%) and data quality issues (27%) dominate as the primary barriers to AI adoption. Critically, these are not independent problems; instead, they form a compounding cycle that keeps many organizations permanently trapped at the planning stage. Poor data quality makes it harder to build reliable AI models. Unreliable models make it harder to demonstrate return on investment (ROI). Insufficient ROI evidence makes it harder to secure budget approval, and limited budgets make it harder to invest in data infrastructure. The cycle continues.

Breaking this cycle requires recognizing that it is an interconnected loop and that addressing either constraint in isolation is insufficient. Organizations that invest in data quality without securing a budget for AI tools will end up with clean data but no models to use it. Conversely, organizations that secure a budget without improving data quality will have tools without the necessary inputs to make them work. The two investments must move together, guided by leadership that understands both are prerequisites rather than sequentially optional steps.

Lack of internal buy-in (18%) is the barrier that most directly reflects an organizational culture problem rather than a resource problem. In nearly one in five organizations, the obstacle is not capability but will. This stems either from leadership skepticism about AI’s value or from cultural resistance among practitioners who have built expertise in existing workflows and perceive AI as a threat to that expertise rather than a force multiplier for it. Fortunately, both are solvable problems. The performance data from AI adopters, detailed in the next section, provides the clearest possible counterargument to internal skepticism.

Regulatory and privacy concerns (16%) complete the picture. While these are real constraints, their lower ranking suggests that most practitioners view them as manageable, seeing them more as a factor to navigate than a reason to stop. This sends an important signal to vendors, consultants, and industry bodies: the organizations that most need help are not waiting for regulatory clarity. They are waiting for budget, data, and organizational alignment.

The Proof of Value is Already There

• 70% – Faster fraud detection
• 45% – Improved portfolio quality
• 35% – Reduced false positives
• 25% – Better customer experience
• 25% – Lower operational costs

Among the organizations that have adopted AI, whether in a pilot phase or full implementation, the results are unambiguous. Faster fraud detection leads at 70%, followed by improved portfolio quality at 45% and reduced false positives at 35%. Better customer experience and lower operational costs both sit at 25%. (Percentages reflect respondents who have adopted AI (pilot or full implementation). ‘Not applicable’ responses excluded.)

Each of these benefits deserves to be understood not just as a performance metric, but as a strategic asset:

• Faster Detection: Fraudsters are caught earlier in the claim lifecycle, before payouts are made and before patterns can be replicated.
• Improved Portfolio Quality: The underwriting risk base improves over time, reducing exposure systematically rather than reactively.
• Reduced False Positives: This is perhaps the most underappreciated benefit in the set. Because every false positive subjects a legitimate customer to unnecessary delay, scrutiny, or rejection, AI-powered precision ensures fewer honest customers experience unearned friction.

The customer experience benefit (25%) deserves particular emphasis for organizations building the internal case for investment. One of the most persistent arguments against aggressive fraud prevention is that it creates friction for legitimate claimants. The data from AI adopters directly challenges that narrative. Precision and speed are not in tension when the detection system is properly designed. This is a powerful secondary argument that speaks to stakeholders beyond the fraud team, including customer experience leaders, C-suite executives concerned about Net Promoter Scores (NPS), and regulators monitoring the fair treatment of policyholders.

A Tool Portfolio Weighted Toward the Past

• 60% – Automated red flags / business rules
• 60% – Cross-reference / external databases
• 44% – Social media / online investigation
• 36% – Document / media verification
• 20% – AI / ML models

The tools most widely deployed today tell a revealing story about where the industry’s defenses are anchored. Automated red flags and cross-reference databases each stand at 60%. These are fundamentally rules-based or lookup-based approaches that are powerful for catching known patterns. However, they are structurally incapable of detecting novel schemes because they can only find what they were specifically configured to find.

Social media and online investigation (44%), document and media verification (36%), and AI or machine learning models (20%) complete the picture. 81% of insurers who have not yet fully deployed AI models are entering the most dangerous phase of the fraud arms race with legacy weaponry. This delay comes at precisely the moment when fraudsters are beginning to deploy generative AI, synthetic identities, and agentic systems that no rules-based system was ever designed to detect.

Sixty percent of respondents rate their current fraud detection tools as only somewhat effective, while just 15% say they are very effective. This is not practitioners dismissing their tools; rather, it is practitioners honestly recognizing their inherent limits. The gap between “somewhat effective” and “very effective” is not a matter of operator skill or workflow optimization. It is a fundamental capability gap that only better technology can close. The industry is aware of this reality. The critical question remaining is whether structural conditions will align to allow insurers to act before the threat arrives in full force.

Key Takeaway

The adoption gap is an organizational problem rather than a technology problem. Budget and data quality are real constraints, but they are entirely solvable. The 9% of insurers who have fully implemented AI are not outliers with unique resources; they simply made a strategic decision to prioritize. Organizations still in the planning phase are not protected by their intent. Every month spent in the planning stage is a month spent operating with tools built to catch yesterday’s fraud.

About FRISS

FRISS is 100% focused on automated fraud, risk and compliance solutions for P&C insurance companies worldwide. Their AI-powered solutions are available for Underwriting, Claims, and SIU, offering support for full end-to-end digital processing. With over 200 implementations across more than 40 countries, FRISS is seen as a trusted advisor, guaranteeing a safe digital transformation for all of their customers, and unique tailoring of solutions to fit their specific needs. Carriers can expect a seamless integration and products that provide a quick time to value (TTV). Now, with $65 Million from their Series B funding round in 2021, FRISS will be able to continue offering their customers state-of-the-art technology to guide carriers through an ever-changing fraud landscape. For more information, please visit FRISS.com.

Source: FRISS