Introduction
The most valuable place for AI is at the core of the business, inside the systems that make the decisions which determine revenue, cost, and risk. These decisions happen in high volume and at high speed, in transactions such as payments, credit approvals, claims processing, and identity verification. They define the financial performance of the enterprise.
AI at the Core means placing intelligence where it matters most, in the very transactions that run the business. It is the difference between reacting to events after they occur and shaping outcomes in the moment. With Machine Learning for z/OS (MLz), this intelligence operates in the same environment that processes the world’s most critical workloads, ensuring decisions are made with the most current data and context, in milliseconds, and at the scale the business demands.
To make every transaction smarter is to turn each one into an opportunity to create value, reduce risk, and improve efficiency. By embedding AI into the core of IBM Z, enterprises can influence millions of high-value decisions every day, achieving measurable financial results that compound over time. The opportunity is not hypothetical. It is here, it is proven, and it can be realised now.
AI at the Core
AI delivers the greatest value when it operates at the heart of an organisation’s business, not at the periphery. For most enterprises, the core is formed by high-volume, high-frequency transactions such as credit approvals, payments, insurance claims, and identity checks. These systems, often built on IBM Z, process millions of transactions each day. Each transaction represents an opportunity to create value or incur loss.
In many cases, these transactions still depend on static rules and fixed thresholds. Such logic cannot adapt to emerging fraud patterns, changing market conditions, or shifts in customer behaviour. The result is a steady leakage of value in the form of missed revenue, unnecessary losses, and operational inefficiencies. Even a small percentage of suboptimal decisions, when scaled to millions of daily transactions, can have a dramatic effect on financial performance.
Embedding AI directly into transactional flows changes the dynamic completely. Instead of relying on rigid rules, decisions can be powered by predictive models that learn from historical data, adapt to real-time context, and execute in milliseconds. This allows outcomes to be faster, more accurate, and more closely aligned with business objectives.
Machine Learning for z/OS (MLz) enables this transformation by running AI directly on IBM Z, co-located with the systems that execute the world’s most important transactions. By placing intelligence exactly where decisions are made, MLz eliminates the latency, compliance risks, and operational gaps associated with off-platform processing.
This is not AI as an afterthought or a separate analytics function. It is intelligence embedded into the very fabric of core business applications. When every decision benefits from real-time insight, every transaction becomes smarter. The closer AI is to where value is created, the greater its impact will be.
Why Transactional AI Delivers More Value
AI is now deployed across the enterprise, but its financial impact depends on where it operates and the stakes of the decisions it influences.
Generative AI improves productivity by creating content, summarising documents, assisting customer service representatives, or acting as a chatbot or copilot. These functions can save time and reduce support costs, but they typically affect productivity and experience rather than directly determining financial outcomes.
Agentic AI goes further, executing multi-step tasks and automating workflows across systems. It can coordinate actions, streamline processes, and even remove human touchpoints. This reduces operational friction and can deliver meaningful efficiency gains, but it still acts indirectly on financial outcomes. The value comes from process optimisation rather than influencing the transaction outcomes themselves.
Transactional AI operates in a completely different arena. It sits inside the high-volume, high-value decisions that define business performance. These are credit approvals, payment authorisations, insurance claim validations, and identity checks. They happen millions of times a day, each carrying a direct revenue, cost, or risk implication.
Consider the difference. AI assistants may help a team of analysts review hundreds of claims in a day. Agentic AI might streamline a few thousand back-office processes. Transactional AI can influence millions of transactions in the same time frame, each worth tens, hundreds, or thousands of dollars. Improving accuracy by even one percent can translate into millions in recovered revenue, reduced losses, or avoided costs every single day.
When embedded into IBM Z systems, Transactional AI executes these decisions in sub-milliseconds without moving data off-platform. This ensures compliance, preserves performance, and applies intelligence exactly where value is created or lost. The scale, speed, and financial exposure of these decisions mean that Transactional AI delivers the most direct business impact.
AI Where It Matters Most
Many AI solutions operate outside the moment of decision. They analyse data after the event, produce recommendations, and rely on manual follow-up. In high-volume, high-stakes environments, that delay means the chance to influence the outcome has already passed. A fraudulent payment may be processed, a valid claim rejected, or a customer lost. The financial impact of these missed interventions compounds rapidly.
Some organisations perform off-platform inferencing, as is the case in certain fraud detection implementations. While this can be effective in some scenarios, it introduces additional latency, network dependencies, and data movement. Each of these factors can reduce responsiveness, increase operational risk, and create compliance considerations that must be managed.
MLz takes a different approach by running inferencing in the same z/OS environment as the transaction. Although MLz and CICS (or IMS) run in separate address spaces, IBM Z enables direct memory-to-memory communication. Model scores are passed through WOLA adapters via shared memory rather than across a network. As a result, inferencing behaves as if it were inside the transaction, providing results with no network hop, no external data transfer, and minimal latency.
This design eliminates unnecessary latency, reduces compliance exposure, and preserves predictable performance at scale.
The result is intelligence that operates at mainframe speed, maintaining the performance, reliability, and security required for business-critical workloads. Fraudulent transactions can be intercepted before approval, claims corrected before submission errors, and customer offers tailored while the interaction is still active.
By embedding AI directly alongside transaction execution, MLz ensures that every decision benefits from the most current intelligence with minimal delay. In industries where the difference between winning and losing is measured in milliseconds, this capability secures both competitive advantage and operational confidence.
Where AI Creates Business Value
Every organisation competes on three primary levers: growing revenue, controlling cost, and managing risk. These are the direct drivers of profitability, and they are where Transactional AI delivers the most measurable and sustained results. By embedding intelligence directly into high-frequency, high-value decision points, MLz ensures that improvements in each of these levers are realised at scale and in real time.
For revenue growth, Transactional AI improves approval rates in areas like payments, credit, and claims without increasing exposure. This can unlock millions in incremental revenue each year. Real-time personalisation enables businesses to capture opportunities at the exact moment of customer engagement, while dynamic pricing adjusts margins based on demand, competitive activity, and individual customer profiles. Every additional conversion or upsell at the point of interaction compounds into long-term revenue growth.
On cost efficiency, Transactional AI automates repetitive decision-making, removing unnecessary manual intervention. Predictive insight anticipates exceptions before they occur, reducing escalations and avoiding expensive rework. Operational resources can be dynamically aligned with demand, ensuring the right capacity is applied to the right place at the right time. When these improvements are applied to millions of transactions, the operational savings become material within a single reporting period.
Risk management benefits from the ability to detect, prevent, and respond to threats in milliseconds. AI models can identify fraudulent behaviour before a transaction is approved, spot compliance violations as they occur, and predict operational disruptions before they cause service failures. These interventions not only prevent financial loss but also protect brand reputation and customer trust.
The common thread is focus. The most successful enterprises do not attempt to augment every decision with AI. They identify the handful of decision points in revenue, cost, and risk that carry the highest value and apply Transactional AI to those areas first. By doing so, they realise quick, measurable returns that build both the business case and the operational capability for scaling AI more broadly.
When intelligence is applied to the moments that matter most, each improvement is multiplied by the volume of transactions it touches. This is how AI moves from concept to competitive advantage.
The Case for Action
In high-volume transactional environments, even a minor improvement in decision quality can translate into extraordinary business value. When millions of decisions are made every day, the compounding effect of accuracy, speed, and precision is immediate and measurable. The inverse is also true. Small inefficiencies or errors, when repeated at scale, can silently erode profitability.
Consider a platform that processes 3 million transactions daily. If just 5% of those decisions are suboptimal, that is 150,000 transactions producing a negative financial outcome every day. If the average financial impact per error is $50, the value at risk is $7.5 million per day. Over the course of a year, that is $2.74 billion in lost or unrealised value from a single decision process.
The reality is that most enterprises have multiple high-volume decisions embedded in their core business journeys. Credit approvals, payment authorisations, claims processing, fraud checks, and customer offers each have their own rate of leakage. When aggregated, the value at stake across all of these decision points is often several times larger than leadership anticipates.
This is why the justification for Transactional AI is not grounded in the novelty of the technology. It is grounded in the economics of loss prevention and value capture. By improving the decision process at the exact point where revenue, cost, or risk is determined, the enterprise locks in financial gains that are realised immediately and sustained over time.
Delaying action carries its own cost. Every day without improvement means losses continue to accrue, customers continue to experience friction, and competitors continue to gain advantage. The sooner Transactional AI is embedded into these decision points, the sooner the leakage is stopped and the upside is captured.
When the numbers are this large and the implementation path is this clear, the case for action becomes more than a strategic choice. It becomes a competitive necessity.
AI Adoption Is Driven by Outcomes
AI is often introduced as a technology project, but technology alone rarely drives adoption. Leadership teams do not invest because of a model’s architecture, a training technique, or the novelty of the algorithms. They invest when there is a clear, measurable link between AI and the improvement of critical business metrics. Without that link, AI initiatives risk becoming pilots that never scale, no matter how advanced the technology.
Outcome-led adoption begins with understanding the decision points that have the greatest financial impact. In fraud prevention, this could be the ability to stop losses before they occur, reducing annual fraud costs by hundreds of millions of dollars. In pricing, it might be the capability to achieve a two percent margin uplift across billions in revenue. In operations, it could be the reduction of rework, error handling, or service downtime that frees up millions in resources.
The challenge is that technology-first conversations often start with the capabilities of the AI rather than the size of the opportunity. A “fraud detection model” is interesting to data scientists, but “reducing false positives by thirty percent” is compelling to a chief financial officer. “AI for pricing” sounds like a project; “adding fifty million dollars in annual margin” sounds like a business priority.
Organisations that succeed with AI treat the technology as an enabler, not the end goal. They build the business case around outcomes, quantify the potential gains, and set clear targets for realisation. Progress is measured not by model accuracy in isolation, but by the impact on the profit and loss statement.
When AI is presented in this way, adoption accelerates. Business leaders can see the direct connection between investment and return, which justifies scaling the approach beyond initial deployments. This shifts AI from an experimental tool to an operational capability that consistently delivers value.
In the end, the most advanced AI model is irrelevant if it does not move the business metrics that matter. Focusing on outcomes ensures that it does.
From Use Cases to Business Journeys
Industry use cases can be a useful starting point. They illustrate how AI has been applied in areas such as fraud detection, anti-money laundering, intelligent underwriting, and demand forecasting. They help teams imagine what is possible and give stakeholders a common frame of reference.
Use cases are powerful in showing where AI has proven value and in providing solution blueprints. However, to maximise their impact, they work best when tailored to an organisation’s specific decision points and workflows. The key is to bridge from a market-tested use case to the exact moments in a business where improving decisions will create the greatest financial and strategic return.
A business journey approach begins by mapping the end-to-end flow of activities that create value for the enterprise. This might be a customer applying for credit, a merchant processing a payment, or a policyholder submitting a claim. Within each journey are decision points — moments where a choice determines whether value is created or lost. These could be approval thresholds, routing logic, prioritisation rules, or offer selection criteria.
By working with an organisation’s specific business journeys, it becomes possible to identify where decisions are currently underperforming and quantify the potential upside of improving them. This process links AI directly to measurable outcomes rather than abstract potential. For example, adjusting an approval process in a high-volume lending journey could generate millions in incremental interest income. Improving accuracy in claims adjudication could reduce rework costs and increase customer satisfaction simultaneously.
The strength of the business journey method is that it connects industry inspiration to client-specific action. It starts with what is unique about an organisation’s operations, its customer base, and its competitive context. From there, it prioritises the AI opportunities that will have the largest financial and strategic impact.
Use cases can show you where AI has delivered value in the market. Business journeys show you exactly where to apply it in your environment to deliver measurable returns.
Prioritise Where AI Can Deliver the Most
Not every decision point is worth the same level of investment. Some carry high potential value but are technically difficult to address today. Others may be easy to automate but have little impact on revenue, cost, or risk. Spreading AI investment evenly across all opportunities dilutes impact and slows momentum.
A disciplined approach begins by scoring each decision point against two dimensions: business value and technical feasibility. Business value measures the financial upside — such as revenue gain, cost reduction, or risk avoidance — that can be achieved by improving the decision. Technical feasibility measures how ready that decision is for AI adoption, based on factors such as data availability, latency requirements, and integration complexity.
Plotting decision points across these dimensions creates a clear prioritisation map. Decisions with high value and high feasibility are the quick wins that can be delivered rapidly to generate proof of impact. High-value, lower-feasibility decisions become strategic investments, worth pursuing over time as data, infrastructure, and capabilities mature. Low-value decisions, regardless of feasibility, should be deprioritised or handled through conventional automation.
This method focuses resources on the small number of decisions that can deliver the largest financial return in the shortest timeframe. For example, improving fraud detection in a payment flow that handles millions of daily transactions may offer an immediate payback period measured in weeks. Enhancing credit approval models in a large loan portfolio could deliver long-term margin growth worth hundreds of millions annually.
By applying this lens, AI deployment becomes a repeatable, outcome-focused strategy rather than a collection of disconnected pilots. Early successes provide both the financial results and the organisational confidence to expand into more complex opportunities.
Prioritisation ensures that AI is not just deployed but deployed where it matters most. In competitive markets, the organisations that master this discipline will see faster returns, greater adoption, and sustained advantage over those that attempt to spread their resources too thin.
The Evolution of Business Decisions
The logic behind core business decisions has evolved significantly over time. Early systems relied on hardcoded if-then statements. These rules were simple to execute and easy to understand, but they were inflexible. Any change in market conditions, customer behaviour, or risk patterns required manual updates to the code, making adaptation slow and costly.
Rules engines emerged as the next step. They introduced configurable business logic that could be updated without altering the underlying application code. While more flexible, these systems were still based on static thresholds and deterministic logic. They could only apply the rules they were given, not learn from new data or context.
The introduction of predictive AI represented a major leap forward. Machine learning and deep learning models could be trained on historical data to identify patterns and make more accurate decisions. In MLz, these models can be deployed within the mainframe environment to act in real time. Predictive AI brought the ability to adapt based on experience, but it still lacked the capacity to explain its reasoning or adjust dynamically to situations it had never encountered.
The latest evolution is reasoning AI, made possible by running large language models (LLMs) alongside predictive models within the MLz environment, accelerated by IBM’s Spyre AI Accelerator Cards. In this architecture, predictive models continue to excel at identifying patterns and forecasting outcomes, while LLMs provide contextual understanding, explanations, and flexible decision logic. Together, they form a multi-model AI approach that combines the precision of prediction with the adaptability of reasoning.
This combination allows systems to handle both routine, data-driven decisions and complex, ambiguous situations that require context and interpretation. For example, a predictive model may identify a transaction as high risk, while the LLM explains why and suggests the most appropriate next action. This not only improves decision quality but also builds trust and transparency in AI-driven processes.
From static rules to adaptive prediction to reasoning in real time, each stage in this evolution has brought the decision-making process closer to the way humans think — but at a scale and speed that mainframes are uniquely positioned to deliver.
What Makes MLz Different
Most AI platforms were designed for general-purpose workloads, not for the speed, scale, and reliability required by mainframe transaction processing. MLz is purpose-built for these environments, which gives it a set of advantages that directly translate into business impact.
Real-time inferencing — MLz delivers sub-millisecond inferencing co-located with transactional systems. This ensures decisions are made at the pace the business demands, without slowing the flow of high-volume operations.
On-platform scoring — Models are executed within MLz, with inference results delivered to CICS or IMS through WOLA adapters using shared memory. Although running in separate address spaces, this design eliminates latency and compliance risks, ensuring scoring behaves as if inside the transaction.
Seamless integration — MLz is designed to work directly with CICS, IMS, Java, and batch environments. This means AI can be embedded without re-architecting existing systems, reducing time to value and implementation risk.
Model portability — It supports models built in open frameworks such as scikit-learn, XGBoost, and PyTorch, which can be deployed via ONNX or PMML. This allows organisations to leverage existing AI investments without costly redevelopment.
Extreme scalability — MLz is optimised to handle millions of inference requests per second without disrupting throughput, ensuring consistent performance even during peak load.
Hardware acceleration — It takes full advantage of IBM Z’s hardware roadmap, including Telum processors and Spyre AI Accelerator Cards, to execute models faster and support more complex workloads, including multi-model AI with predictive and reasoning capabilities.
Trustworthy AI by design — MLz includes built-in capabilities for monitoring, explainability, and governance, aligning AI decisions with enterprise standards and regulatory requirements.
By combining these characteristics, MLz brings intelligence into the same operational environment that runs the business’s most critical workloads. It delivers AI performance without compromise, ensuring that the core transaction systems are both smarter and just as dependable as they have always been.
The Strategic Playbook
Successful AI transformation is not the result of isolated projects. It follows a deliberate sequence that moves from insight to measurable impact. The organisations that win with AI apply a consistent method, ensuring that investment is directed toward the decisions that matter most.
Map — Begin by identifying the business journeys where critical decisions are made. These are the points in your operations where outcomes directly determine revenue, cost, or risk. Mapping these journeys reveals where value is created, where it is lost, and where AI could make a measurable difference.
Quantify — For each decision point, determine the volume of decisions made and the financial impact of getting them wrong. This puts a real number on the value at stake and creates a fact-based case for change. Without quantification, AI remains a technology initiative instead of a business priority.
Prioritise — Use a dual lens of business value and technical feasibility to identify the decisions worth pursuing first. High-value, high-feasibility opportunities deliver rapid proof of impact. High-value but lower-feasibility opportunities become strategic investments for the future.
Act — Deploy AI into the highest-priority decision points, focusing on production impact rather than prolonged proof-of-concept exercises. Embed the models into operational workflows so they influence outcomes in real time.
Measure and refine — Track the business results, not just technical metrics like model accuracy. Monitor the value delivered, refine the models as conditions change, and extend the approach to other high-impact decisions across the organisation.
This disciplined sequence ensures that AI is not treated as an experiment but as a repeatable capability that compounds value over time. It aligns technology with measurable outcomes and avoids the common trap of chasing too many ideas without delivering meaningful results.
By following a playbook grounded in mapping, quantification, prioritisation, action, and measurement, AI becomes a driver of business transformation, not just a technical project.
The Bottom Line
Every day, IBM Z processes millions of decisions that directly shape revenue, cost, and risk. Many of these decisions are still governed by static rules or outdated thresholds, leaving significant value untapped. MLz changes this by embedding intelligence into the same environment where these decisions are made, transforming each one into a potential source of competitive advantage.
The opportunity is substantial. Across industries, the value at stake from improving just a handful of high-volume decisions often runs into the billions annually. These are not speculative numbers; they are grounded in the daily economics of payments, lending, claims, fraud prevention, and customer engagement. The scale of IBM Z transactions amplifies the impact, allowing even small improvements in accuracy or efficiency to deliver outsized returns.
The platform is ready. IBM z16 and z17 systems feature Telum I and Telum II processors with on-chip accelerators for high-speed AI inferencing. The introduction of Spyre AI Accelerator Cards unlocks the ability to run more complex models, including multi-model architectures that combine predictive accuracy with reasoning and explanation. With MLz, these capabilities are integrated directly into the operational fabric of the business.
The method is proven. Focusing on business journeys instead of generic use cases ensures AI investment is directed to the decision points with the greatest financial leverage. Prioritisation ensures that resources are applied where they will deliver the highest return in the shortest time. Measurement and refinement turn initial wins into sustained performance gains.
Urgency matters. Competitors are moving to embed AI into their core operations. Those who act now secure early advantages in decision quality, customer experience, and operational efficiency that are difficult to dislodge. Those who wait risk competing on slower, less intelligent decision processes while the market moves ahead.
These transactions already define the performance of the business. With MLz, they can also define its competitive future. The question is not whether the opportunity exists, but how quickly organisations will act.
Conclusion
The path to maximising the value of AI is not about chasing every new technology trend. It is about applying intelligence precisely where it will have the greatest financial and strategic impact — inside the high-volume, high-stakes decisions that run the business. Transactional AI on IBM Z, powered by MLz, brings predictive and reasoning capabilities into the same operational environment that processes the world’s most critical workloads.
This combination of proximity, performance, and precision turns AI from an abstract concept into a measurable business driver. The methodology is clear: map business journeys, quantify the value at stake, prioritise the decisions that matter most, act decisively, and measure results. The technology is ready, proven, and built to scale without compromising speed, reliability, or compliance.
Enterprises that act now will capture economic gains that compound over time, strengthen their competitive position, and set a foundation for continued innovation. Those who delay risk losing opportunities and falling behind faster, more intelligent competitors.
In an environment where milliseconds and accuracy define success, embedding AI into core transactions is no longer optional. With MLz, the capability is within reach. The opportunity is here. The time to realise it is now.
