For anti-money laundering and know your customer systems, durable advantage will come from cleaner data, stronger feedback loops and better governance
By Craig Muir, Ian O’Neal and Brendan Kirk
Solomon Partners Technology Group
Artificial intelligence is reshaping AML and KYC, but the center of value is shifting in a way that is still underappreciated. What began as an effort to improve transaction monitoring accuracy is evolving into a deeper re-architecture of financial crime systems — one where control, decisioning, and governance are increasingly centralized within software platforms. The critical question now is not simply which vendors have better models, but which ones control the data that makes those models durable and defensible.
The stakes are significant. The United Nations estimates that money laundering accounts for between 2% and 5% of global GDP annually, underscoring the scale of illicit financial activity that banks, regulators, and compliance technology providers are attempting to detect and prevent. As financial crime grows more complex and cross-border in nature, institutions are increasingly turning to AI-driven systems to strengthen monitoring and risk management capabilities.
The scale of the challenge helps explain why AI adoption accelerated so quickly. Industry studies estimate that traditional transaction-monitoring systems can generate false-positive rates exceeding 90%, creating substantial operational costs as compliance teams investigate large volumes of alerts that ultimately prove benign. Reducing this burden was one of the earliest and most compelling use cases for machine learning in financial crime compliance.
What has changed is the role AI plays in the control environment. As models move upstream — driving onboarding decisions, generating alerts, and shaping investigative workflows — they become embedded in the mechanisms that determine how risk is measured and acted upon. This introduces a new class of challenges, well understood by regulators: model drift, lack of explainability, and bias all sit squarely within the compliance perimeter.
The financial stakes are substantial. Global spending on AML compliance is estimated to exceed $200 billion annually across financial institutions, reflecting the growing complexity of financial crime controls, regulatory expectations, and cross-border compliance requirements. As compliance costs continue to rise, institutions are increasingly seeking technologies that can improve efficiency without compromising governance or risk oversight.
As a result, the industry's focus is converging around governance. Regulators in both Europe and the United States are increasingly requiring that AI systems be explainable, auditable, and continuously monitored. This trend is reflected in frameworks such as the EU AI Act, which places stricter obligations on high-risk AI systems, and U.S. banking regulators' SR 11-7 model risk management guidance, which emphasizes validation, monitoring, and governance of models used in critical decision-making. What is less frequently articulated is that governance without data advantage is ultimately fragile.
In practice, model performance in financial crime is heavily dependent on three categories of data: proprietary transaction data, labeled outcomes (e.g., confirmed suspicious activity), and contextual enrichment data (e.g., identity, network relationships, adverse media). Most vendors can build models; far fewer can access, aggregate, and continuously refine these data sets at scale.
This is where the market is beginning to bifurcate.
A subset of platforms is increasingly positioned as data-driven control layers, rather than analytics tools. These companies derive strength not only from their algorithms, but from their ability to sit inside the transaction flow, capture feedback loops, and improve continuously as new signals emerge.
This dynamic can be seen in payment networks such as Visa and Mastercard, which process enormous volumes of transaction data across global payment ecosystems. Their ability to combine real-time transaction visibility with AI-driven fraud and risk analytics creates a powerful feedback loop that continuously improves detection capabilities while benefiting from data sets that are difficult for standalone vendors to replicate.
A similar advantage is emerging among specialist financial crime platforms such as Feedzai, Quantexa, and NICE Actimize, which aggregate transaction, customer, and investigative data across banking environments. By integrating directly into compliance workflows and incorporating investigative outcomes into future risk assessments, these platforms can strengthen model performance over time through closed-loop learning and increasingly rich data environments.
The more deeply embedded they are in customer workflows, the more proprietary data they accumulate — creating a compounding advantage that is difficult for point solutions to replicate.
This dynamic is most visible at the network layer. When a major payments network internalizes a behavioral analytics engine, the logic is not simply a bet on AI-enabled fraud detection, it is a move to place real-time modeling directly on top of global payments data. The strategic value lies in the combination: a real-time model, continuously trained on high-quality, high-volume proprietary transaction data. That combination is inherently difficult to displace.
A different starting point can produce a comparable advantage in the application layer. Next-generation vendors are designed to integrate across fragmented banking environments, pulling together transaction, customer, and case data into unified models that improve as they ingest more activity. Given sufficient scale and integration, such platforms can build defensible data positions even without owning a payments network.
The tension between these two models — network-based data advantage versus application-layer aggregation — is likely to define the next phase of competition. In both cases, however, the underlying principle is the same: data is not an input; it is the asset.
This has direct implications for how assets in the sector should be evaluated. Vendors that rely on generic data inputs or operate as thin analytics layers are structurally disadvantaged, regardless of model sophistication. By contrast, platforms that can demonstrate access to proprietary data flows, closed-loop learning (i.e., integration of investigative outcomes back into models), and persistence within core workflows are far better positioned to sustain performance and pricing power over time.
Explainability, audit trails, and model monitoring are necessary — and increasingly aligned with international expectations reflected in FATF recommendations and evolving supervisory guidance on financial crime controls — but they are not, on their own, sources of differentiation.
The real advantage lies in being able to generate better decisions because the underlying data is richer, cleaner, and continuously improving. Governance, in that context, becomes both a regulatory requirement and a reinforcing mechanism for maintaining trust in that data.
The emergence of agentic AI further intensifies this dynamic. Systems that can autonomously gather data, surface insights, and initiate investigative steps will only be as effective as the data they can access and interpret. As these systems move closer to decision-making, the importance of high-quality, proprietary data — and the controls governing its use — becomes even more pronounced.
From an investment standpoint, this points to a clear conclusion. The most valuable assets in AML and KYC are unlikely to be those with the most advanced models in isolation. They will be the platforms that combine:
deep integration into transaction and onboarding workflows
access to proprietary and continuously refreshed data sets
closed-loop learning that improves outcomes over time
and embedded governance that makes those outcomes defensible
These characteristics are what transform AI from a feature into infrastructure.
The broader implication is that AML and KYC are evolving into data-centric control layers embedded within financial institutions’ core architecture. The next phase of value creation will not be driven by incremental gains in detection accuracy, but by the ability to build, own, and continuously refine the data that underpins financial crime decisioning. In that environment, data is not just a differentiator, it is the primary source of durability.
