The conversation about AI value in large enterprises is dominated by workforce productivity. How many hours of knowledge worker time does AI save, how many support interactions does it deflect, how much faster do development teams ship code when they have AI assistance. These productivity gains are real, and the investment cases for productivity-focused AI are increasingly well-supported by operational evidence.
The productivity conversation is not wrong. It is incomplete. It accounts for the AI value available at the workflow layer of enterprise operations and is largely silent about the AI value available at the transaction layer, which in organisations whose core operations run on IBM Z is a different order of magnitude.
Value density as the investment framework
AI value density is the financial consequence of a decision improvement multiplied by the volume of decisions that improvement applies to. It is the metric that should determine where AI investment generates the highest return, and it is a metric that many AI investment processes do not apply because it requires connecting model performance to financial outcomes at a level of specificity that most AI teams and most business cases do not achieve.
At the workflow level, the value density calculation involves the productivity value of the time saved by a knowledge worker using AI assistance, multiplied by the number of knowledge workers in the relevant population. For a large enterprise with a significant investment in copilot-style AI, this can produce a substantial aggregate figure. The unit value is small and the volume is large: many workers, each saving a modest amount of time per day.
At the transaction level, the value density calculation involves the financial consequence of an improvement in decision quality for a single transaction, multiplied by the number of transactions that decision applies to per year. For a large bank or payment network, the transaction volume is measured in billions per year. The unit value of a decision quality improvement depends on the decision type: for fraud detection, it includes the fraud value of transactions that improved detection captures; for credit adjudication, it includes the revenue and loss impact of more accurate default prediction; for authorisation optimisation, it includes the revenue impact of reducing false declines on legitimate transactions.
The calculation is specific to each organisation and each decision type, but the scale relationship typically holds: the value density of a one-percent improvement in transaction-level decision quality, applied across the transaction volume of a large IBM Z deployment, frequently exceeds the value density of a ten-percent productivity improvement for a knowledge worker population of comparable size.
Why the transaction-level opportunity is underinvested
The transaction-level AI opportunity is underinvested in most enterprises for a structural reason that has nothing to do with its financial merits: the people who own the AI investment process are not the people who understand what runs on IBM Z.
Enterprise AI strategy is typically owned by digital leadership, data leadership, or a dedicated AI function. Those leaders are well-positioned to identify workflow-level AI opportunities because they interact daily with the knowledge worker populations those opportunities serve. They are not typically positioned to identify transaction-level AI opportunities because those opportunities live in core banking, payments, and insurance processing systems that are owned by different organisational functions and often treated as infrastructure rather than investment targets.
The organisational gap between AI strategy ownership and IBM Z operational knowledge is the primary reason the transaction-level opportunity is underinvested, not a lack of financial merit or technical feasibility. Closing that gap requires connecting the AI investment process to the operational system owners, which in most enterprises requires a deliberate cross-functional effort that has not yet been made.
The portfolio implication
The right AI portfolio allocation for an enterprise with significant IBM Z transaction volume is not a choice between workflow-level AI and transaction-level AI. It is a sequencing and weighting decision that allocates investment across both levels in proportion to their value density.
A portfolio weighted heavily toward transaction-level decision improvement on IBM Z is likely to generate higher total return than an equally-sized portfolio weighted toward workflow-level productivity AI, for most organisations with significant transaction volume. This is not an argument against workflow-level AI investment. It is an argument for including transaction-level IBM Z AI in the portfolio calculation, which many current AI investment processes do not do.
The first step toward that portfolio allocation is the decision inventory described elsewhere in this series: identifying the high-volume operational decisions running on IBM Z, measuring their current quality, and quantifying the financial value of improving them. That analysis will typically show that the transaction-level opportunity belongs in the AI portfolio alongside, and often ahead of, the workflow-level productivity investments that currently dominate it.
The value is already there. It is running at billions of transactions per year on IBM Z. The question is whether the AI investment process is looking at it.
