From Scoring to Reasoning: Operationalizing Trusted AI with Cloudera, NVIDIA, and Dell Technologies for Financial Services.
Enterprise AI has crossed an important line over the past two years. The conversation is no longer about whether AI works or which models to evaluate; it is about how to deploy AI reliably to production, run it at scale, and stand behind every decision it makes. The questions that occupy IT and business leaders today are operational. Where should AI workloads live, how should they connect to the data that makes them useful, and how can the organization remain in control of the answers AI produces?
Underneath that operational shift is a deeper one. For financial services companies managing fraud detection and credit risk, earlier AI implementations focused on isolated predictions and binary outcomes. Today’s requirements demand systems that can reason across multiple data sources, explain their decisions, and take instantaneous coordinated action. This move from scoring to reasoning is reshaping how organizations approach their most consequential workflows.
That shift requires a different foundation. Trusted data that AI can rely on, purpose-built AI infrastructure, or AI Factories that can deliver real-time inference at scale, and the ability to keep sensitive data within sovereign, secure environments. Cloudera AI, accelerated by NVIDIA and deployed on Dell Technologies infrastructure, gives enterprises a practical foundation for this moment, with financial services as a leading proof point and a value proposition that extends to any organization moving AI from pilot to production.
From Experimentation to Operation
Research from Enterprise Strategy Group, now part of Omdia, indicates that 78% of financial services organizations consider AI agents a top or high priority compared to other AI initiatives. That elevation is not unique to financial services. Across regulated industries, AI has moved from experimental projects to board-level priority, with a focus on the most consequential business processes. Anti-money laundering, credit underwriting, risk management, and fraud detection rank among the top use cases, but the same pattern holds across healthcare, telecommunications, the public sector, and manufacturing.
The same research surfaces friction. 51% of financial services organizations cite concerns about data privacy and the handling of sensitive information when using AI agents. 45% are concerned about the inaccuracy of AI agent decision-making. These are not adoption hesitations from the experimentation era. They are operational requirements that any production AI system must satisfy, and the architectural choices made today will either support or undermine those requirements for years.
A New Architecture: From Scoring to Reasoning
Traditional machine learning continues to provide enormous value for high-volume, high-speed scoring. Models trained on structured data evaluate transactions in under 50 milliseconds and identify anomalies at scale. What machine learning does not do well is explain itself. A score tells an investigator what happened. It does not tell them why, and it does not give a human reviewer the context to make a confident decision.
Generative AI fills that gap. Large language models can consume thousands of pages of documents, extract relevant signals, and generate narrative summaries that explain complex situations in plain language. For a fraud case, generative AI produces an investigation brief that contextualizes why a transaction was flagged. For a credit decision, it drafts a preliminary memo that synthesizes structured financials, market intelligence, and unstructured analysis.
Agentic AI orchestrates these capabilities into coordinated workflows. Rather than asking human analysts to gather information from disconnected systems, agents autonomously retrieve the relevant data, invoke the appropriate models, and assemble a comprehensive assessment that a human reviews and acts on. The result is faster decisions and better-explained outcomes, with the human staying in control of the final call. This is the architecture NVIDIA calls an AI Factory for financial services: a full-stack, production-ready system where governed, high-throughput AI agents operate across core financial workflows, from fraud detection and credit underwriting to risk modeling and customer intelligence, transforming traditional data centers into industrial-scale environments for producing intelligence.
Where AI Belongs: With the Data
The most decisive operational question for enterprise AI is also the simplest. Where should the AI run? The answer is increasingly clear. AI delivers its highest business value when it runs where your data lives, whether on premises, in the cloud, or across hybrid environments. For financial institutions, this is not an architectural preference; it is a regulatory and competitive imperative. Banks and payment providers that train custom, domain-adapted models on their own proprietary transaction data build an intelligence asset that competitors cannot replicate. Cloudera makes AI deployable wherever that data lives.
What changes when AI runs alongside enterprise data is the quality of context. A risk score in isolation is a number. A risk score with full access to recent transactions, customer history, device signals, and account behavior becomes a contextualized recommendation that a human can act on with confidence. This contextual depth, the ability to extract predictive signals from the multimodal datasets that financial institutions already hold, is what separates institutions building proprietary AI intelligence from those relying on generic models.
Cloudera AI, Accelerated by NVIDIA, on Dell Technologies for Financial Services
Cloudera provides the trusted data foundation. The Cloudera hybrid data and AI platform unifies structured and unstructured data in an open Lakehouse architecture. Cloudera’s Data in Motion captures real-time streams alongside historical data, while the Shared Data Experience delivers consistent governance and security across on-premises, private cloud, and public cloud environments. The platform is engineered to keep sensitive data inside its jurisdiction while still making it available to the AI systems that need it.
NVIDIA accelerates the AI pipeline through a full-stack AI Factory platform. NVIDIA Blackwell GPUs reduce training time for traditional machine learning and enable large-scale inference for neural networks and language models. For risk modeling, NVIDIA acceleration transforms simulations that previously ran overnight into near real-time analyses. Cloudera AI Inference, powered by NVIDIA, allows organizations to deploy and scale models on premises, with the NVIDIA AI stack including NVIDIA Nemotron, NVIDIA's family of open, domain-adaptable models that institutions can fine-tune on their own proprietary data, NVIDIA Dynamo-Triton Inference Server, and NVIDIA AI Enterprise, the operating system of the AI Factory, which provides NIM microservices for optimized low-latency inference and NeMo for building and fine-tuning custom models. The NVIDIA AI Blueprint for financial fraud detection is based on graph neural networks and the Build Your Own Transaction Foundation Model developer example. The blueprints are available from NVIDIA and provide organizations with a concrete starting point for building AI applications on financial data.
Dell Technologies provides the enterprise infrastructure that ties the solution together. Dell PowerEdge servers configured with NVIDIA Blackwell GPUs deliver the compute density that demanding AI workloads require, while keeping that compute in the same physical environment as the sensitive data it operates on. Together, this solution can deliver substantially lower total cost of ownership than equivalent public cloud inferencing, with predictable economics that scale with usage rather than against it.
Two Use Cases in Action: Fraud Detection and Credit Risk
Financial services illustrate the value of this architecture. In real-time fraud detection, Cloudera captures transactions alongside customer interaction history, device metadata, and behavioral signals. Transaction foundation models, transformer-based AI trained on billions of financial events, interpret behavior in context, where timing, device, location, and prior activity all shape meaning. For example, a payment at midnight means something different when it is the fourth in ten minutes, on an unfamiliar device, in a city the customer has never transacted from before. NVIDIA-accelerated machine learning models, such as Nemotron or open-source foundation models, score each transaction within the latency envelope required by instant payment networks. When a transaction is flagged, an agentic workflow on Cloudera AI retrieves the relevant context, and a language model generates a case file with a natural-language explanation of the risk factors. The investigator no longer receives a score and a workload. They receive a preliminary investigation brief and a faster, better-supported decision.
Credit risk assessment shows the same shift. Bureau scores and structured payment history have always been narrow. The unstructured information that could enrich a credit decision, including financial statements, market analyses, and news coverage, has historically been too labor-intensive to incorporate at scale. The combined platform unifies these sources in a governed Lakehouse that uses NVIDIA-accelerated Monte Carlo simulations and risk modeling, workloads that NVIDIA's full-stack platform runs at a fraction of legacy infrastructure cost, uses NVIDIA acceleration to run sophisticated simulations that previously would have extended decision timelines, and produces preliminary credit memos that loan officers refine rather than draft from scratch.
The same logic applies to claims adjudication in insurance, clinical decision support in healthcare, supply chain risk in manufacturing, and intelligence operations in the public sector. The use case differs by industry. The architecture does not.
From Insight to Outcome
Organizations adopting this approach see a consistent set of outcomes:
Faster decisions on the workflows that matter most, with sub-50 millisecond inference for real-time scoring and dramatically compressed cycles for analytical workloads
Better-explained decisions backed by data lineage, context, and business metadata that satisfy both internal review and regulatory scrutiny
Lower total cost of ownership relative to public cloud inference, with predictable economics as usage scales
Reduced risk through architectural alignment with data sovereignty, compliance, and governance requirements
These benefits compound. The same architecture that delivers operational AI today becomes the foundation for the next generation of agents, the next set of use cases, and the next phase of automation, without rebuilding the data and governance layer underneath.
The Path for Financial Services to Operational AI
The next phase of enterprise AI will be defined less by model capability and more by operational discipline. The institutions that succeed will be those that evolve from data-centric to intelligence-centric, building proprietary AI on their own data, deploying it within secure and sovereign environments, and governing it with the auditability that regulators and risk committees demand. Cloudera AI, accelerated by NVIDIA and deployed on Dell Technologies infrastructure, gives enterprises a practical path to that destination, turning the conversation from what AI can do to what AI is doing, every day, in the systems that matter most.
To learn more, download the: Fraud Detection and Credit Risk Assessment with Cloudera AI, Accelerated by NVIDIA on Dell Technologies Omdia Showcase Brief and BrightTALK On-Demand Webinar.