Where AI actually creates value – and where it doesn’t

One of the stranger effects of the current AI moment is that it has become possible to sound strategic simply by sounding interested. Leaders talk about adoption, transformation, experimentation and capability. Teams are asked to identify use cases. Vendors promise productivity, efficiency, better decisions and new growth. Somewhere inside all this activity, the technology can begin to feel less like a tool and more like an obligation. The implicit question shifts from whether AI is useful in a given context to where it can be inserted quickly enough that the organisation does not appear to be falling behind.

That is rarely the most helpful place to start.

For most organisations, the question is not whether AI matters. It clearly does. The more useful question is where it actually creates value, and where it does not. Those are not the same thing, and the distinction matters more than many organisations are currently willing to admit.

One reason this is hard is that AI value is often discussed too generically. It is talked about as though the benefits are obvious and universal: more speed, more insight, more productivity, more scale. Sometimes that is true. AI can make certain kinds of work much faster. It can help teams synthesise information, automate repetitive tasks, generate first drafts, surface patterns in large datasets, improve discoverability, and reduce the friction involved in moving from question to rough answer. In the right context, those gains are real.

But value does not emerge simply because AI has been added. It emerges when the technology is applied to the right kind of work, in the right kind of way, with enough clarity about what improvement actually means.

Broadly speaking, AI tends to create the most value where work is high-volume, repeatable, information-heavy, and currently slowed down by manual effort that does not need a great deal of nuance. That might include things like summarising large volumes of material, drafting internal documentation, automating routine customer interactions, supporting research, improving knowledge retrieval, speeding up first-pass analysis, or reducing the time spent on repetitive operational tasks. In these cases, the gains are often fairly visible. People save time. Work gets moved forward faster. Teams can spend less energy on lower-value activity and more on the work that still requires human judgment.

This is where many organisations first see the appeal of AI. It helps with tasks that are necessary but not especially differentiating. It can improve productivity without asking the business to reinvent itself. It can make work less cumbersome. For leaders under pressure to deliver efficiency or capability gains, that is a meaningful proposition.

But this is also where organisations can start overstating what AI is actually doing. Saving time is not the same as creating value unless the time saved is genuinely redirected into something more useful. Automating a repetitive task may be worthwhile, but if the underlying process is already unclear, poorly designed or low-value to begin with, AI may simply help the organisation do the wrong thing faster. In that sense, AI often amplifies whatever is already there. Well-designed workflows can become more efficient. Weak workflows can become more confusing at greater speed.

The same principle applies to decision-making. AI can absolutely support better decisions in some settings. It can help synthesise evidence, surface patterns, improve access to relevant information and widen the set of options under consideration. It can reduce the time it takes to get from a question to a working hypothesis. But that is not the same as replacing judgment. In fact, one of the clearer dividing lines in AI value is between work that benefits from augmentation and work that still depends heavily on context, discernment, trust and accountability.

This is where many organisations start forcing AI into the wrong places. There is a tendency to assume that if AI can generate an answer, it is therefore useful to let it lead. But in work that involves ambiguity, sensitive trade-offs, relationship management, ethics, brand judgment, leadership, or decisions where accountability matters, the value of AI is often much more limited unless it is being used carefully and with strong human oversight. A plausible answer is not the same thing as a reliable one. A fast answer is not automatically a good one. In some cases, overreliance on AI can actively erode value by reducing the amount of human thinking applied to the work.

That is why one of the least helpful ways to approach AI is to ask where it can replace people most aggressively. The better question is where it can meaningfully support human capability without introducing new forms of risk, error, confusion or trust erosion. In some areas, that support is substantial. In others, the costs are less visible but more serious.

There is also a category of AI activity that creates more noise than value simply because it is being pursued for symbolic reasons. This happens when organisations adopt tools or launch pilots mainly because they want to signal progress, reassure a board, satisfy a strategy conversation or keep pace with competitors. The activity may look impressive, but the connection to a material business problem is weak. Teams experiment because experimentation itself has become the goal. Use cases are collected without a clear sense of which ones matter. The result is often a lot of motion without much leverage.

This is one reason the organisations getting the most from AI are not always the ones using the most of it. They are often the ones being more selective. They understand that the existence of a use case does not automatically make it a good one. They are clearer about where AI can create real gains, and just as clear about where the technology is likely to create distraction, poor-quality output, governance challenges or misplaced confidence. They do not assume that technical possibility and business value are the same thing.

That selectiveness matters because value is contextual. A use case that is worthwhile in one organisation may be pointless in another. The same tool can create very different outcomes depending on data quality, workflow design, staff capability, leadership expectations and the type of work involved. AI does not arrive in a vacuum. It enters an organisation with its own habits, constraints, systems and blind spots. Whether value appears depends as much on that environment as on the tool itself.

This is also why better AI decisions tend to come from leaders who are willing to ask narrower, harder questions. Where are we currently losing time on work that does not need to be so manual? Where is information hard to access or use? Which tasks genuinely benefit from speed and synthesis? Where do we still need strong human judgment no matter how capable the tool seems? What parts of the workflow are mature enough for automation, and which are too messy for AI to improve anything meaningfully? Those questions are less glamorous than broad transformation language, but they tend to produce much better outcomes.

For many organisations, the challenge is not discovering that AI can do impressive things. It is becoming disciplined enough to separate useful from performative, high-value from low-value, and augmentation from overreach. That is where real strategy begins. Not in adopting AI widely for the sake of it, but in understanding where it can make work better, faster or more effective in ways that actually matter, and where it is still more likely to create noise than value.