How evaluative AI is changing the economics of trust

For years, businesses have treated fraud prevention and the customer experience as opposing priorities. Tighten controls and introduce more friction to legitimate customers. Relax them and allow fraud to slip through. While many businesses have long accepted this tradeoff as unavoidable, that assumption is now increasingly challenged. 

As fraud grows more sophisticated and businesses gain access to richer sources of identity, behavioral, and transaction data, the question isn’t just whether businesses can stop more fraud. It is whether they can do so without creating unnecessary obstacles and putting off legitimate customers. 

A newer category of AI is emerging to answer that question. Most of the AI conversation has focused on generative AI, which creates content, and agentic AI, which carries out tasks. A third category, evaluative AI, serves a different purpose: helping organizations assess information and determine what it suggests about trust and risk. 

This shift comes at a critical moment as fraudsters increasingly leverage AI to create more compelling scams and synthetic identities, making it harder to separate legitimate activity from fraud. INTERPOL’s 2026 Global Financial Fraud Threat Assessment estimates that global losses from financial fraud reached $442 billion in 2025 alone. In the United States, the FBI’s Internet Crime Complaint Center reported over 22,000 complaints involving AI-related cybercrime and nearly $900 million in associated losses last year. 

At the same time, from ecommerce checkouts to account openings, businesses are under pressure to deliver seamless digital experiences. Each added verification step, account review, or declined transaction creates friction and frustration that can do lasting damage to customer relationships and conversion rates. 

As a result, organizations are starting to view AI as more than an automation and productivity tool. They’re looking at AI to help improve decision-making in moments that directly affect both security and the customer experience. 

Why traditional fraud controls are no longer cutting it 

Fraud teams rarely lack data. Most organizations already collect extensive identity, device, behavioral, and transaction data. The problem is that the information may sit in separate systems or be considered in isolation, making it difficult to determine how those signals relate to one another. 

Without a holistic, contextual view of the customer and the activity, organizations are left making important decisions from incomplete information. The result is a familiar cycle of tighter fraud controls, heavier manual reviews, and increased customer friction. 

While these precautions might reduce fraud, they can also create costs that are more difficult to measure. A declined transaction might cost a retailer a customer, not just a purchase, while a loan applicant asked to complete additional verification steps might turn to another lender entirely. Datos Insights estimated that false declines cost the card industry $213 billion globally in 2025. Although these costs receive less attention than direct fraud losses, their real impact is measured in lifetime value, not single transactions. 

How connecting signals changes the decision 

The challenge for fraud and risk teams is no longer just identifying suspicious activity but rather determining whether activity can be trusted. 

Consider two banking customers who appear similar. Both log in from a different location and an unfamiliar device and then initiate larger-than-usual transactions. 

These scenarios appear nearly identical at first glance. However, a closer look might reveal that the first customer recently traveled or purchased a new phone, making the activity consistent with the broader customer profile. 

The second customer’s activity might be a different story altogether. The device may be connected to previously identified fraud, the identity information may overlap with known fraud patterns, or the transaction may conflict with established behaviors. 

Two transactions that appear similar on the surface may carry very different levels of risk when viewed contextually. The distinction comes not from any one signal but from the relationships between them. 

Where evaluative AI fits 

Agentic AI may dominate today’s headlines, but every task an AI agent carries out ultimately rests on a judgment about trust: is this user, this device, this transaction legitimate? That judgment is where evaluative AI does its work. 

In fraud prevention, evaluative AI collectively analyzes identity, behavioral, transaction, and device data to determine whether activity aligns with the broader customer relationship or points to elevated risk. By surfacing patterns and context that may otherwise be missed, it helps organizations make more informed decisions. 

The goal is not to replace human judgment or determine an organization’s tolerance for risk. Fraud and risk teams still play the critical role of establishing policies and deciding when activity should be approved, challenged, or declined. Evaluative AI simply provides a more complete picture before those decisions are made. 

Rather than focusing on automation and efficiency, evaluative AI focuses on improving decision quality. In a landscape where organizations must continuously decide who and what to trust, that capability is proving increasingly valuable. 

Measuring the full business impact 

Fraud programs have long been evaluated primarily through the lens of prevented losses. While this metric remains important, it only tells part of the story. False declines, abandoned applications, manual review volume, customer complaints, and lost transaction revenue should be considered as well. After all, a system that stops fraud but drives away legitimate customers may not be performing as well as its fraud loss numbers suggest. 

When businesses can evaluate activity with greater confidence, trusted customers can move forward with fewer unnecessary interruptions, creating benefits that extend beyond fraud reduction. Conversion rates improve, operational costs fall, and customer trust and loyalty deepen. 

As evaluative AI improves businesses’ ability to assess risk in context, the longstanding tradeoff between security and customer experience is becoming less absolute. The businesses that gain a competitive advantage won’t necessarily be those with the most data or the most AI tools but will instead be those that can make the most confident decisions about trust. 

Questions leaders should be asking 

A few questions reveal how much room there is to improve: 

• Do we measure the cost of false declines, abandoned applications, and manual reviews with the same rigor as direct fraud losses? 

• Can our systems evaluate identity, behavioral, device, and transaction signals in context, or do they still sit in separate silos? 

• When a transaction is flagged, do our teams see the full customer relationship or just a single suspicious event? 

• Does our AI inform human judgment or attempt to replace it? 

For years, organizations assumed better security required greater friction. Increasingly, evaluative AI is proving those goals don’t have to be at odds.