Why improving business decisions may be a lower-risk path to AI adoption.
For the past two years, the dominant narrative around Artificial Intelligence has been remarkably consistent. Organisations are urged to reinvent the enterprise, redesign operating models and re-engineer business processes end to end. The implication is that meaningful AI adoption begins with large-scale business transformation.
There is merit in that thinking. Many organisations run processes that have evolved over decades, accumulating manual workarounds, duplicated activities, technical debt and organisational complexity. In those environments, AI may well be the catalyst for broader transformation. If the process itself is fundamentally broken, improving individual decisions within it will only deliver limited value.
Is Business Transformation Always Necessary?
This perspective, however, overlooks another class of organisation. Financial institutions, insurers, governments and airlines have spent decades refining their core operational processes. Payment processing, card authorisation, claims adjudication and tax administration are not immature workflows waiting to be reinvented. They are highly optimised systems executing millions or billions of transactions a year with extraordinary reliability, resilience and regulatory oversight.
For these organisations, the challenge is rarely that the process is broken. It is that the quality of the decisions made within it can still be improved. This raises an important question. If a business process already performs efficiently, should the first objective really be to redesign it?
Every Business Process Is Simply a Series of Decisions
Every operational process exists to support a sequence of business decisions. A payment is authorised or declined. A merchant application is approved or referred. A loan is accepted or rejected. A claim is paid or investigated. A customer is authenticated, routed, prioritised or offered a product. These decisions occur continuously throughout every business process, determining both customer outcomes and operational performance.
Viewed through this lens, the process itself is only part of the story. The quality of the process is ultimately determined by the quality of the decisions made within it.
Decision-Centric AI
I refer to this alternative philosophy as Decision-Centric AI. Rather than viewing AI primarily as a catalyst for redesigning entire business processes, Decision-Centric AI focuses on improving the quality of individual business decisions that already exist within those processes. The surrounding workflow remains familiar. Existing operational systems continue to orchestrate activities exactly as before. What changes is the intelligence applied at the moment a decision is made.
Fraud losses decline because suspicious transactions are identified more accurately. Customer experience improves because unnecessary declines are avoided. Operational costs reduce because fewer exceptions require manual intervention. Risk decisions become more consistent because they are informed by richer data and predictive insight. The process may remain largely unchanged, yet its outcomes improve significantly because the underlying decisions become better.
Is This Really New?
A fair challenge deserves to be addressed directly. Banks have deployed machine learning in fraud detection and credit scoring for more than twenty years, and analysts have written about “decision intelligence” as a discipline for some time. What is different now? I would argue that three things have changed, namely capability, reach and discipline.
The first is capability. Modern AI can reason over unstructured information such as documents, correspondence, conversation history and images, which previous generations of decision models could not economically incorporate. Decisions that once relied on a narrow set of structured variables can now draw on the full context surrounding a transaction, a claim or a customer.
The second is reach. Historically, only a handful of decisions justified the cost of building and maintaining a bespoke model. Fraud and credit scoring earned that investment; merchant onboarding referrals, claims triage, collections treatment and thousands of mid-tier decisions did not. That economic threshold has fallen dramatically, bringing a far longer tail of decisions within range.
The third is discipline. Decision-Centric AI is not simply “deploy a model”. It is a deliberate practice of ranking an organisation’s decisions by economic value, improving the most valuable one first, measuring the outcome, and reinvesting the evidence in the next. The novelty lies less in any individual technique than in treating the decision portfolio, rather than the process map, as the unit of AI strategy.
What This Looks Like in Practice
Consider an illustrative example. A card issuer processes 500 million transactions a year and declines 4% of them. Industry experience suggests a meaningful share of those declines are false positives, meaning legitimate customers turned away. If improved decision intelligence recovered even one in ten of those declined transactions at an average value of £40, that would represent roughly £80 million in recovered annual payment volume, alongside reduced fraud losses and fewer customers reaching for a competitor’s card.
Critically, nothing about the authorisation process changes. The same systems route the same transactions through the same steps in the same milliseconds. Only the decision at the centre becomes better informed. The numbers above are illustrative rather than drawn from a specific institution, but the arithmetic is the point. When a decision executes hundreds of millions of times a year, even fractional improvements compound into material economic value, with no process redesign required.
Deciding Whether to Start with the Process or the Decision
None of this means process redesign is never the answer. The honest question is a diagnostic one. Which situation is your organisation actually in? Several tests can help. If a process produces the right outcome but too slowly or expensively because of hand-offs, rework and duplicated effort, the constraint is the process, and redesign is the answer. If a process executes quickly and reliably but produces outcomes you would not choose with better information, such as payments wrongly declined, claims wrongly paid or customers wrongly prioritised, the constraint is the decision. If the people operating a process spend most of their time compensating for its structure, transformation is likely overdue. If they spend most of their time exercising judgement with incomplete information, the decision is where the value lies.
There is also a caution worth stating plainly. Improving decisions within a genuinely flawed process risks entrenching that process, making it harder to justify redesign later. The diagnostic matters precisely because Decision-Centric AI is powerful enough to make a mediocre process perform acceptably. For mature, mission-critical operations, however, such as the payment networks, claims platforms and tax systems described above, the diagnosis usually points the same way. The process is sound, and the decision is the opportunity.
Why This Lowers Risk
This distinction has important economic implications. Large-scale transformation programmes require significant investment, organisational change, governance redesign, technology modernisation and extensive retraining. They typically span multiple years and depend upon the successful coordination of numerous initiatives. While they can deliver substantial benefits, they also introduce significant execution risk.
Decision-Centric AI offers a different starting point. By improving a single economically significant decision, organisations can establish measurable business value before committing to broader transformation. Existing processes remain intact, minimising disruption to customers and employees. Investment can be targeted, outcomes measured, and lessons learned before expanding into adjacent business areas. Rather than asking an organisation to transform everything at once, Decision-Centric AI enables transformation to occur incrementally, building confidence through evidence rather than assumption.
A Foundation for Broader Transformation
Decision-Centric AI should not be viewed as an alternative to enterprise transformation, but as a practical way of beginning it. An organisation may improve payment authorisation today, merchant underwriting tomorrow, fraud detection next quarter, and customer servicing thereafter. Each successful deployment strengthens the business case, develops organisational capability and demonstrates measurable economic value. Over time, these incremental improvements may naturally lead to broader changes in operating models and business processes, but they do so with experience, executive confidence and proven outcomes already established.
Enterprise transformation and Decision-Centric AI are therefore not competing philosophies. One seeks to redesign the enterprise from the outset. The other seeks to improve the decisions that drive it, allowing transformation to evolve through measurable success.
A Different Starting Question
Perhaps the question organisations should begin with is not “How do we redesign our business for AI?” but rather “Which business decision would create the greatest economic value if we could improve it?”
For organisations operating mature, mission-critical systems, that subtle shift in perspective may prove to be the lowest-risk and most economically attractive path to AI adoption. By improving one decision at a time, they can unlock measurable business value today while laying the foundations for broader AI transformation tomorrow.
