Enterprise investment in AI has concentrated on technical capacity. Organizations have acquired models, platforms, development tools and data infrastructure. They have established governance functions, trained specialists and launched growing numbers of experiments. The ability to build and run AI has improved considerably, yet greater technical capacity has not produced adoption at the scale many organizations expected.
The usual explanations focus on data quality, integration complexity, governance, skills and executive sponsorship. Each can become a genuine obstacle, but these explanations begin too late in the process. They assume that the organization has already identified a sufficiently important business problem and established why AI should be used to solve it. In practice, many AI initiatives begin before those questions have been answered.
Before an organization can build, govern or deploy an AI solution, someone must identify a business outcome worth improving. They must understand what produces the current outcome, where the organization has an opportunity to intervene and whether better information or judgment could materially change the result. They must also establish whether the potential improvement is valuable enough to justify the investment required to capture it.
This is not primarily a model-development problem. It is a business-understanding problem.
We solved only one scaling problem
The technology industry has spent years making AI execution more scalable. Models can be accessed through APIs, deployed across multiple platforms and incorporated into applications more quickly than before. Cloud services, open-source frameworks and increasingly capable infrastructure have lowered many of the technical barriers to experimentation and deployment.
This progress matters, but it addresses only one side of AI adoption. An organization must be able to execute AI, but it must also be able to form high-quality opportunities for that capability. Execution determines whether a proposed solution can be built. Opportunity formation determines whether it deserves to be built in the first place.
The first problem has attracted enormous investment. The second remains heavily dependent on individual experience and chance. In many organizations, an AI opportunity emerges only when an industry expert who understands the business happens to work with a technical specialist who recognizes how AI could improve it. If that combination never occurs, the opportunity may never be identified.
This dependence on particular individuals can produce successful projects, but it cannot support adoption at enterprise scale. It leaves the quality of the AI pipeline determined by who participates in each conversation rather than by a repeatable organizational capability.
Opportunity discovery remains inconsistent
Most client-facing professionals are expected to discuss AI across businesses they cannot possibly understand in equal depth. A seller may work with banks, insurers, healthcare organizations and government agencies. A consultant may move between functions with different processes, economics and regulatory obligations. A technical specialist may understand what a model can do without knowing which operational decision deserves attention.
When business understanding is limited, teams tend to begin with the technology. Conversations start with generative AI, agents, predictive models or a particular platform capability. The team then searches for somewhere to apply it, reversing the logic through which a sound investment opportunity should be formed.
This capability-first approach can generate plausible ideas, but technical possibility is a poor substitute for business relevance. A model may perform well without materially improving an outcome. An agent may automate a task that was never expensive enough to justify the effort. A technically successful proof of concept may have no credible path to production because the original opportunity was not sufficiently grounded in the operation of the business.
A technically valid use case is therefore not necessarily an investable opportunity. The distinction is important because organizations frequently discover it only after committing scarce technical resources.
More use cases do not create better opportunities
Many organizations have responded by creating increasingly large AI use-case libraries. These catalogues can demonstrate the breadth of AI and provide useful inspiration, but they do not replace the business understanding required to select an opportunity. They describe where AI might be relevant without establishing where it will create sufficient value for a particular organization.
A use-case library might identify fraud detection as an opportunity in banking, claims automation in insurance or demand forecasting in retail. These categories remain too broad to support an investment decision. They do not explain which specific outcome is underperforming, what causes the loss, whether the organization can intervene or how the proposed improvement would be captured economically.
The result is a fundamental disconnect. An enterprise can possess hundreds of documented use cases and still struggle to identify its next credible AI investment. The list expands the number of possibilities, but it does not necessarily improve the quality of selection.
This is why the objective should not be to generate the largest possible inventory of AI ideas. It should be to identify a smaller number of opportunities connected to material outcomes, addressable problems and credible routes into production.
Business understanding changes the starting point
A business-first approach begins from the premise that AI creates value only when it improves an outcome that matters. Those outcomes are shaped by decisions made throughout the operation of a business. Some decisions are made by people, while others are embedded in applications, workflows, policies or static rules. Some occur occasionally, while others are repeated millions of times across operational systems.
The opportunity for AI emerges when one of those decisions can be materially improved. Intelligence may arrive too late to affect the outcome. A decision may depend on incomplete information or rules that no longer reflect the environment in which the business operates. Another may remain manual because the organization does not yet have sufficient confidence to automate it.
These conditions cannot be recognized by examining AI capabilities in isolation. They become visible only when the organization understands how the business operates, which outcomes are unacceptable and where better judgment could change the result. Technology should enter the conversation after that context has been established.
This distinction also changes how value is assessed. A modest improvement to a high-frequency operational decision may be worth more than a dramatic improvement to an infrequent task. A highly accurate model may create little value if its recommendation arrives after the decision has already been made. An attractive use case may be impossible to operationalize if the organization cannot act on the result.
Understanding the business is therefore not preliminary research undertaken before the real AI work begins. It determines whether that work has a credible reason to exist.
The missing enterprise capability
Organizations often describe AI skills as the main constraint on adoption. Model-development expertise is important, but it is only one form of scarcity. Industry judgment can be harder to acquire and more difficult to scale because it develops through years of exposure to business processes, operating constraints and recurring patterns of value creation and loss.
Experienced industry specialists understand more than the facts of a market. They can recognize when a client is describing a symptom rather than a cause. They know which outcomes are economically significant, which constraints are real and which apparently attractive opportunities are unlikely to survive contact with the operating environment.
Those experts cannot participate in every client conversation or assess every possible application. Hiring more of them may increase capacity, but it does not fundamentally change the operating model. Expertise remains concentrated in individuals, while the wider organization continues to depend on their availability.
Scaling AI adoption therefore requires a way to extend business-first judgment across the organization without pretending that every employee can become an industry expert. Client-facing teams need enough structured understanding to begin in the right place, recognize potentially consequential problems and know when deeper specialist involvement is justified.
What I set out to solve at IBM
I encountered this problem while working to expand AI adoption at IBM. The technical capability existed, and IBM had specialists capable of explaining it in considerable depth. The harder challenge was enabling client-facing teams across different industries and markets to identify the business situations in which that capability could create material value.
Traditional product enablement could teach teams what the technology did. It could not give them years of industry experience or ensure that every client conversation began with a sound understanding of the business. As a result, the ability to identify strong AI opportunities remained uneven and heavily dependent on access to a relatively small number of experienced people.
I developed a business-first AI discovery methodology to address that constraint. Its purpose was to improve the quality and consistency of opportunity discovery, allowing teams to begin with the client’s business and progress towards a credible AI opportunity. The significance was not any single opportunity produced through the approach, but the possibility of turning discovery from an individual skill into a repeatable enterprise capability.
The methodology did not remove the need for industry experts, technical specialists or direct engagement with the client. It allowed those scarce resources to be applied later and more selectively, once a meaningful business hypothesis had begun to emerge.
Better adoption begins with better selection
Organizations frequently measure AI adoption through activity. They count ideas, workshops, experiments, models and proofs of concept. These measures demonstrate that work is occurring, but they do not establish whether the organization is becoming better at selecting opportunities.
A stronger view of adoption would examine the quality of the opportunities entering development. The business outcome should be material, the current problem should be understood and the organization should have a credible point of intervention. The expected improvement must also justify the cost and disruption involved in putting the solution into production.
Improving opportunity formation changes the economics of AI delivery. Technical specialists spend less time qualifying weak ideas, proofs of concept begin with clearer production intent and investment decisions can compare opportunities on business value rather than enthusiasm. Scarce engineering capacity is directed towards problems with stronger reasons to proceed.
This does not guarantee that every AI initiative will succeed. It improves the quality of the decisions about which initiatives should exist.
Scaling understanding alongside technology
The next phase of AI adoption will not be determined solely by access to increasingly capable models. Those capabilities will continue to spread across the market, making technical access a weaker source of differentiation. Competitive advantage will depend increasingly on how effectively organizations connect AI to their own processes, knowledge and economically consequential decisions.
That connection cannot remain accidental. Enterprises need the ability to discover credible AI opportunities repeatedly across industries, functions and client teams. They must scale business judgment alongside technical execution, creating a stronger path from technical possibility to operational value.
The organizations that develop this capability will not simply produce more AI projects. They will make better choices about where AI belongs, commit resources with greater confidence and move a higher proportion of worthwhile opportunities towards production.
Scaling AI adoption starts with understanding the business. The real breakthrough comes when that understanding no longer depends entirely on who happens to be in the room.
