Most organisations treat finding AI opportunities as a technical exercise. They engage vendors to demonstrate what is possible, they send teams to AI conferences, they hire data scientists to tell them where their data could support a model. The assumption is that identifying AI opportunity requires AI expertise, and that without it the organisation cannot know where to look. That assumption is wrong and it creates a dependency that is entirely unnecessary.
AI creates value by improving decisions. Not by processing data, not by automating workflows, not by generating reports — all of those things are means to an end. The end is always a decision that changes because of AI, producing a better outcome than the same decision made without it. Fraud that is blocked because a payment was scored correctly. Credit extended to a customer who was previously declined incorrectly. A claim settled faster because the relevant information was available at the moment the adjuster needed it. In every case, the value lives in the decision, not in the technology. The AI is the mechanism. The broken decision is the opportunity.
This is not a semantic distinction. It is the most practical reframe available to any organisation that is trying to build an AI strategy grounded in business value rather than technology enthusiasm.
A broken decision has three characteristics
A decision is broken when the gap between its current quality and its achievable quality has a cost attached to it that makes improvement worth pursuing. That definition is precise enough to be useful and broad enough to apply across industries, functions, and decision types.
The first characteristic is a measurable gap. A decision is not broken because people feel it could be better. It is broken when the difference between current outcomes and achievable outcomes can be expressed in numbers. Approval rate versus achievable approval rate. False positive rate versus achievable false positive rate. Cycle time versus achievable cycle time. Without a measurable gap, there is no reliable way to know whether improvement is worth pursuing and no way to measure whether AI achieved it.
The second characteristic is a cost. The gap has to be worth closing. Cost of error multiplied by frequency of error is the most direct calculation, but it takes other forms depending on the decision type. Revenue foregone from unnecessarily declined applications. Operational cost of processing alerts that are not fraud. Customer attrition from decisions that take too long. The cost anchors the opportunity in economic terms and is the only legitimate basis for prioritising one broken decision over another.
The third characteristic is an addressable constraint. The decision is not merely suboptimal — it is suboptimal for a reason that AI can address. The two most common reasons are information and speed. The decision is made with incomplete information that a model, trained on historical outcomes, could supply. Or the decision is made too slowly for the context it operates in, and a model scoring at the point of decision could produce a better outcome in the time available. If neither constraint is present, the decision may be suboptimal for reasons AI cannot address — process failure, incentive misalignment, organisational dysfunction — and identifying it as an AI opportunity would be incorrect.
This makes opportunity identification a business skill, not a technical one
The three characteristics of a broken decision require business knowledge to identify, not AI knowledge. Which decisions in this function produce the most costly errors. Where does cycle time create downstream consequences. Where does information arrive too late to change what happens. What does it cost when the wrong customer is approved, the wrong alert is investigated, the wrong claim is declined. Every experienced business leader in an organisation already has intuitions about these questions and most of them have access to the data that would turn those intuitions into quantified observations.
What they typically do not have is a framework that connects those observations to AI. The question they have been asked — where could AI add value — is not a question they are equipped to answer, because it requires AI expertise they do not have. The question they should be asked — where are decisions costing the most money and what is the constraint that makes them poor — is a question they are entirely equipped to answer, because it requires business knowledge they already have.
The practical consequence is significant. Organisations that frame AI opportunity identification as a technical exercise will consistently identify opportunities that are technically interesting rather than economically important, because the people doing the identifying have technical knowledge. Organisations that frame it as a business performance exercise will consistently identify opportunities where improvement is most valuable, because the people doing the identifying have business knowledge. The output of the first process is a technology agenda. The output of the second is an investment case.
The market prefers mystery to simplicity
The framing of AI as a technical exercise is not accidental. Vendors, consultants, and specialist teams all have structural reasons to present AI opportunity identification as something that requires their expertise. Mystery creates dependency. If finding AI opportunities requires specialist knowledge, the people who have that knowledge become essential to any organisation that wants to pursue AI. The market is organised around that dependency.
The alternative, that finding AI opportunities requires asking experienced business people where decisions are costly and what constraints make them poor, is a simpler and more powerful method that threatens the dependency. It does not require a vendor. It does not require a design thinking workshop. It does not require a proof of concept to discover what the opportunity is. It requires asking the right people the right questions and being disciplined enough to quantify the answers before any technology is selected.
I have found more AI opportunity in a two-hour conversation with an experienced operations manager than in a week-long discovery engagement structured around capability demonstration and use case ideation. Not because operations managers know more about AI, but because they know where the decisions are, what goes wrong, and what it costs. That knowledge, structured around the three characteristics of a broken decision, is all that is required to identify and prioritise AI opportunities that will move the commercial needle.
The technology comes after. It always should have.
