Across finance and digital commerce, speed is no longer just a customer-experience feature. It is becoming an operating condition. Instant payments, automated credit decisions, algorithmic trading, AI-led service workflows and 24-hour digital channels can move value or trigger actions before a traditional review cycle has time to intervene. That creates a structural change: organisations must increasingly manage risk at the same speed at which transactions and decisions occur. The Bank for International Settlements has highlighted this shift in work on real-time payment analytics, where the value of network-level transaction analysis lies partly in the ability to surface suspicious patterns while payment activity is still unfolding. The broader implication goes beyond payments. As systems become faster, the window for detection, escalation and recovery becomes shorter.
Speed changes the economics of control
For much of the digital era, organisations could separate the moment a transaction occurred from the moment it was reviewed. Batch monitoring, end-of-day reconciliation and retrospective exception analysis were imperfect but workable because many processes moved slowly enough to allow intervention after the fact. Real-time systems weaken that assumption. When funds settle in seconds, when software can approve a workflow instantly, or when an AI agent can trigger a downstream action without waiting for a human checkpoint, the value of a control depends increasingly on whether it can operate before or during the event.
This does not mean every decision should be automated. It means that the architecture of control must become more selective. BIS work on digital fraud points to the tension between the efficiency of digital finance and the risks created by fraud, identity theft and deception. The practical response is not to slow every transaction equally, but to differentiate between low-risk activity that can move frictionlessly and higher-risk activity that deserves stronger authentication, additional data or a deliberate pause.
The result is a move from universal friction to targeted friction. That is strategically important because customer expectations have been trained by fast digital services. A bank or platform that responds to risk by making every user wait may reduce fraud at the cost of competitiveness. A better model is to use data, context and behavioural signals to decide where speed is safe and where speed itself becomes part of the risk.
Real-time monitoring becomes an infrastructure question
Real-time risk management is often described as an analytics problem, but analytics is only one layer. The underlying system also needs clean data, low-latency connectivity, consistent identity signals, event-driven architecture and the ability to enforce a decision across multiple channels. A fraud model that identifies an anomalous payment is only useful if the payment system can act on that signal quickly enough. Likewise, an AI governance rule is only meaningful if it can constrain an agent before a prohibited action is completed.
The G20 TechSprint work hosted through the BIS Innovation Hub has focused on trust and integrity in scalable open finance, including the challenge of detecting fraud in fast-payment environments with low latency. That framing is useful because it treats risk controls as infrastructure rather than an external compliance layer.
As this model spreads, financial institutions and large digital businesses may need to invest less in isolated monitoring tools and more in shared control layers. These can include streaming data platforms, decision engines, policy orchestration, identity verification, model monitoring and real-time case escalation. The value comes from connecting them, not merely owning them.
The rise of network-level intelligence
Many forms of modern fraud are difficult to identify from a single account or a single company. Coordinated scams can distribute activity across multiple users, institutions and payment channels. That weakens controls based only on one customer history or one firm's dataset. Network-level analysis can reveal relationships, repeated behaviours and clusters that are invisible locally.
Project Hertha tested this idea using synthetic data representing a large retail-payment ecosystem. Its published findings suggest payment-system analytics can complement the monitoring performed by individual banks and payment service providers, particularly when identifying new patterns of suspicious behaviour. The project also stresses limits, including legal, governance and practical questions that remain outside a purely technical solution.
The broader lesson is that risk may increasingly be managed through ecosystems. Banks, payment networks, technology providers and regulators may need better mechanisms for sharing high-quality signals without creating unacceptable privacy or competition concerns. This is a governance challenge as much as a data-science challenge.
AI accelerates both sides of the contest
Artificial intelligence increases the speed of both legitimate and malicious activity. It can improve anomaly detection, triage alerts and prioritise investigations. At the same time, it can make social engineering, synthetic identities and automated attack campaigns more convincing and scalable. That creates an arms race in which static rules are likely to age faster.
Regulatory and supervisory commentary has increasingly recognised this dual effect. Remarks collected by the BIS on innovation and financial crime describe how AI can be used both to bypass controls and to strengthen them. For institutions, the challenge is not simply adopting more AI but building feedback loops that let models learn from new patterns while maintaining explainability, oversight and model-risk controls.
This will raise the importance of data lineage, labelled outcomes and post-event learning. Every confirmed fraud case can become a training signal. Every false positive can help tune thresholds. But firms will need governance around how these signals are created, validated and shared, particularly where models influence customer access or payment execution.
Operational resilience has to keep pace with automation
Real-time decisioning can concentrate operational risk. If a rules engine, identity service or cloud dependency becomes unavailable, thousands of decisions may fail simultaneously. If a faulty model is deployed, errors can propagate quickly. Speed therefore increases the value of resilience, redundancy and controlled degradation.
Organisations may need to design for graceful fallback: a payment can be delayed rather than rejected, a risky transaction can require secondary approval, or an AI workflow can revert to a human queue if a control service is unavailable. These are not merely business-continuity features. They are part of the risk logic of a real-time system.
The strategic question is shifting from how quickly a company can automate to how safely it can operate at automated speed. The firms that answer that well may be able to deliver faster service without treating security, fraud prevention and control as afterthoughts.
The new management metric is time to intervention
Traditional risk metrics often focus on losses, alert volumes, false positives and recovery rates. Real-time systems add another dimension: how long does it take to detect an issue and intervene before value moves or a decision becomes irreversible? That measure can matter across fraud, cyber risk, treasury, trading, operations and AI governance.
A shorter time to intervention does not always require more automation. Sometimes it requires clearer decision rights, better routing, fewer handoffs or better data. The important point is that delay itself becomes measurable risk. As businesses become more digital, organisations may increasingly treat latency in control systems the same way they treat latency in customer-facing systems: as an operating variable that can affect outcomes.
This is why the next phase of digital transformation may be less about making systems faster and more about making control systems equally fast. Real-time growth without real-time risk management can create fragile scale. Real-time growth with selective, well-designed controls can create a more durable operating advantage.
Key questions
Why are real-time systems changing risk management?
Because transactions and automated actions can complete before traditional batch reviews or manual escalation processes have time to intervene.
Does faster risk management mean more customer friction?
Not necessarily. The goal is increasingly to apply friction selectively to higher-risk activity while allowing low-risk transactions to remain fast.
What capabilities matter most?
High-quality streaming data, identity signals, real-time analytics, policy orchestration, resilient infrastructure and clear escalation paths are increasingly important.
What is the main management implication?
Time to detection and time to intervention are becoming operational risk variables in their own right.