Inside the 'perfect storm' disrupting fraud and risk management

The tension between speed and safety and Customer experience has long been the defining paradox of digital payments and their adoption. approve. Fraud detection systems that lag even by a fraction of a second can alienate legitimate customers, while overly aggressive filters can trigger false rejections, eroding trust and revenue.

“The value of catching fraud is minimized if legitimate customers are trapped in the network.” Matthew PierceVice President, Fraud Risk Management and Dispute Operations i2ctold PYMNTS in a discussion B2B Payments 2025 Campaign.

Pearce highlighted the three key metrics i2c monitors most closely: fraud loss rate, fraud decline rate and False alarm rate. These metrics define the delicate balance between vigilance and availability. The goal is to minimize losses and friction, and this harmony requires constant recalibration.

Balancing capture rate and friction

Fortunately, the future looks promising; and This is largely Thanks to innovations in artificial intelligence (AI). as agentive, generative and Predictive AI becomes central to financial operations, and their application proves that performance and accountability can coexist.

“Leading organizations measure performance across multiple dimensions and continually adjust models to maintain balance. … Modern defenses combine real-time anomaly detection with controlled retraining cycles,” Pierce said, noting that agility can be a true differentiator.

The result is what he calls “agility without volatility,” or systems that evolve quickly enough to keep up with fraudsters but not overreact to the point of destabilizing existing portfolios.

“Agility without volatility is the new definition of resilience,” Pierce said. “Adaptability is as important as accuracy.”

Build explainability

At the same time, while artificial intelligence becomes integral to financial operations, it also raises troubling questions about trust and responsibility. Regulators have increased their focus on “black box” decision-making systems, requiring explainability in areas such as credit scoring and dispute resolution. and Fraud detection.

For Pierce, these are not check boxes; They are design principles. At i2c, every AI model is version controlled and documented and Tested for fairness before deployment. When regulators or customers ask why decisions were made, companies can provide a clear narrative: data lineage, rationale and The governance path that led to this outcome.

“Every result [must] “It's all traceable from the functionality and rules behind it to the business impact it has,” Pierce said. “We build explainability into the model and the model lifecycle. It's not an afterthought, it's part of the process.”

key For this “complete story”, explainability is the data. The future of AI in payments depends on the data that powers it.

“We draw insights from a broad portfolio of transaction data, dispute outcomes and behavioral patterns,” he continued. “Each data set is subject to pattern checking, drift tracking and Conduct challenger testing before models go into production. “

Federated Learning and Data Integrity

The key is jointness: a local/global hybrid design maintains predictive power without overfitting any single data source.

“The model learns from global trends but is tailored to local conditions,” Pierce said. “This allows us to maintain performance accuracy without biasing the model towards a single portfolio.”

Equally important are the things not to do Enter the system. Personally identifiable information is never part of the i2c training process. Instead, the company tags or hashes the identifiers at the schema level, ensuring that “the model only sees the attributes relevant to the predictions, not the customers behind them.” When explanations are generated, they are built from structured metadata rather than raw personal details, Pierce said.

This approach may soon become table stakes. As regulators like the Federal Reserve and the Consumer Financial Protection Bureau continue to refine algorithmic accountability frameworks, financial institutions not only need systems that perform well, but also systems that can show they perform well.

“Transparency will never be a cost of security,” Pierce said. “Privacy protection really starts upstream.”

From pilot to proof of impact

Even the most complex AI systems can get stuck without a clear path to implementation. For banks and fintech companies, the challenge is often not what to build; but how arrive Put it into practice.

“Effective AI adoption follows a strict 90-day cycle,” Pierce said. “Scope and success criteria first, integration and configuration second, then limited rollout. … The hardest hurdle is not the technology – it’s the organization. Governance, approvals, data quality and Regulatory comfort tends to slow down AI more than coding. “

The goal is not a “proof of concept,” he added. This is “proof of influence.”

By shifting the focus from feasibility to results, i2c aims to redefine AI as a strategic asset rather than an experimental enterprise.

This distinction can resonate with financial institutions looking for clearer ROI on their digital transformation efforts and also informed Why integration of solutions like i2c is handled through APIs designed to co-exist with legacy core and CRM, thereby easing the resource burden on customers.

“Customer resources are still smaller,” Pierce noted. “They have data access, compliance oversight and The technical liaison, while the provider is responsible for setup and governance. “

His team’s work hints at what the next phase of fintech development will look like: smart systems that are not only faster and more adaptable, but also more ethical and auditable.