AI transformation often begins with a familiar kind of excitement. A new platform is launched. A promising tool is demonstrated. A team sees what could be faster, easier or more efficient, and the natural response is to ask where it might fit.
That excitement is understandable. It is also incomplete.
The organizations that make meaningful progress with AI are rarely the ones that start by asking which tools they should buy. They are the ones that first take time to understand the work itself: how it moves, where it gets stuck, what judgement is required, which decisions depend on context, and where people are spending energy on work that may no longer need to be done in the same way.
In other words, the starting point for AI transformation is not technology. It is work.
This matters because AI does not improve a process simply by being added to it. When organizations introduce AI into unclear, outdated or poorly understood workflows, they can make those workflows faster without making them better. The result may be more output, but not necessarily more value; more activity, but not necessarily more clarity; more automation, but not necessarily better decisions.
Before we automate, we need to understand.
The risk of doing the wrong work faster
It is tempting to begin with tools because tools are tangible. They can be demonstrated, purchased, piloted and announced. They create momentum, and in many organizations momentum is reassuring.
But tool-first transformation narrows the conversation very quickly. The focus shifts to features: what can this summarize, generate, classify, schedule, draft or analyze? Those capabilities may be genuinely useful, but they do not tell us whether a tool is being applied to the right problem, at the right point in the work, with the right level of human judgement.
This is where AI initiatives can lose their way. A team may automate a reporting process before asking whether the report is still needed. A department may use AI to draft communications before clarifying what the communication is meant to accomplish. A leader may search for efficiency inside a workflow without first examining why that workflow has become so cumbersome.
Years ago, in writing about effectiveness, we returned repeatedly to an important distinction: doing things right is not the same as doing the right things.1 Selecting the right work requires deliberate choices about priorities before attention turns to tasks and execution. The danger, otherwise, is that considerable effort can be spent doing the wrong things very well.
AI makes that old problem newly consequential.
The issue is not that AI tools are unhelpful. Many are remarkably useful. The issue is that they can amplify the quality — or the dysfunction — of the thinking, processes and decisions around them. If the underlying work is fragmented, unclear or misaligned, AI may simply help an organization do the wrong things faster.
Work is more than a collection of tasks
One common mistake in automation efforts is to treat work as a list of activities. If a task is repetitive, it becomes a candidate for automation. If it takes time, it becomes a candidate for acceleration. If it involves writing, analysis or coordination, it becomes a candidate for AI support.
That lens can be useful, but it is not enough.
Work is also a system of relationships, decisions, constraints, expertise and accountability. A task that appears simple on paper may carry important context in practice, while a step that looks inefficient may exist because it manages risk, builds trust or allows an experienced person to notice something that is not captured anywhere in the formal process.
This is especially true in knowledge work. A briefing note is not simply a document; it reflects decisions about what matters, what can be left out, what risks should be surfaced and what someone needs to know in order to act. A client update is not just information transfer; it may also be a relationship moment. Even a meeting, much maligned in conversations about productivity, may be where alignment, confidence or accountability is actually built.
AI can support parts of this work, and in many cases it should. But we need to understand what the work is doing before deciding what to automate, augment, redesign or deliberately keep human.
Start with how the work really happens
A practical AI transformation therefore begins by looking closely at how work actually happens — not simply how a process is described in a policy or procedure, and not how leaders assume it happens from a distance.
Where does information enter? Who touches it? What decisions are made along the way? Where does work wait, repeat or get reinterpreted? Where are people re-entering information or creating workarounds? Which steps require expertise, judgement, empathy or negotiation, and which exist largely because “that is how we have always done it”?
These may seem like basic questions, but they often reveal more than starting with a technology assessment.
When teams map work this way, they often discover that the most obvious AI opportunity is not the most important problem. A time-consuming task may be a symptom rather than the constraint. The deeper issue may be unclear inputs, duplicated approvals, inconsistent decision criteria, fragmented systems, or the absence of a shared understanding of what good looks like.
In those situations, AI may still form part of the solution, but not necessarily the first part.
Sometimes work should be eliminated. Sometimes it should be simplified or standardized. Sometimes a decision needs a clearer owner. Sometimes existing systems need to communicate with one another. Sometimes the workflow itself needs to be redesigned. And sometimes AI is exactly the right choice for drafting, synthesis, pattern recognition, triage or repetitive processing.
The order matters.
Reclaiming capacity for what?
This brings us back to time.
Every person gets the same 1,440 minutes in a day. No technology changes that constraint.
In my earlier writing on time and effectiveness, we argued that the challenge is not really to “manage” time but to make conscious choices about what deserves it. Time is finite while the possible demands on it are almost endless; effectiveness therefore depends on connecting what we do to what matters most.2
The same principle applies to organizations adopting AI.
It is easy to measure time saved. It is harder, and more important, to ask what happens to that time next.
If AI saves a leader five hours a week, does that leader simply absorb five more hours of activity? If a team produces reports twice as quickly, does it produce twice as many reports? If employees become more efficient, is the reward simply more work?
That would be a remarkably limited vision for a technology with this much potential.
Reclaimed capacity can instead create room for better decisions, stronger relationships, deeper thinking, creativity, learning, and leadership. A business owner might spend less time answering routine questions and more time looking ahead. An advisor might spend less time finding information and more time advising. An instructor might spend less time marking and more time re-imagining their course content.
The value is not the automation itself.
The value is what the automation makes possible.
A better first question
Understanding before automating does not mean endlessly studying the organization while technology races ahead. Thoughtful action and experimentation matter. In fact, a well-understood, bounded workflow can be one of the best places to begin because the problem, people and desired outcome are clear enough to learn from.
But the purpose of an early AI initiative should not be to prove that AI is impressive. That has already been proven.
The purpose should be to learn whether AI can meaningfully improve work in a particular context — reducing friction, strengthening decisions, improving quality or reclaiming capacity for something more valuable.
So perhaps the better first question is not: What should we automate?
It is: What work are we trying to make better?
That question directs attention toward value rather than novelty, toward judgement rather than features, and toward the lived reality of work rather than technology in the abstract.
AI will continue to change quickly. Tools will improve, capabilities will expand and the pressure to adopt will not disappear. But the organizations that benefit most will be those able to connect that technological possibility to a deep understanding of their own work.
The tool is not the transformation. The work is.
References
- Clark, A. M., & Sousa, B. J. (2018). How to Be a Happy Academic: A Guide to Being Effective in Research, Writing and Teaching. SAGE Publications.
- Sousa, B. J., & Clark, A. M. (2021). Why academics never have enough time … and what to do about it. University Affairs.