Leonardo Da Vinci: What Leaders Can Learn From Observation

A weekly report can make a problem look settled before anyone has looked at the work that produced it. Revenue is down, so the problem must be sales. A delivery date slipped, so the problem must be effort. An AI demonstration looks impressive, so the problem must be a shortage of automation. Those are explanations, not observations. They may be right. They may also be convenient enough to prevent the question that would change the next decision.

Leonardo da Vinci is useful here for a reason more modest than the usual “genius” story. His surviving notebooks and drawings show sustained attention to bodies, water, light, machines and motion. The Royal Collection Trust’s account of Leonardo’s drawings and notebooks makes that breadth visible. The transferable lesson is not that a manager should become a Renaissance polymath. It is that close looking can be a form of reasoning: a way to slow a premature label, make a system visible and choose a better test.

That distinction matters most when a decision is expensive, difficult to reverse or likely to affect people who have more direct knowledge than the person naming the problem. Observation is not a ritual visit to the front line. It is an attempt to discover what the first explanation misses, and then let that discovery alter the next move.

Start with the decision in front of you

Observation becomes serious when it is attached to a decision. “Let’s understand the customer” is too broad to guide attention. “Before changing the onboarding flow, we need to know where people stop, what they expected and what information they did not have” is a real job. It gives the observer a boundary and makes it possible to tell whether the work changed anything.

Begin with a decision that is already tempting the team: replace a tool, change a policy, hire a role, redesign a handoff, introduce an AI assistant, or respond to a sharp customer complaint. Write the current explanation in one plain sentence. Then write the decision it appears to justify. This separates a fact from the story around it. “Customers abandon the form at step three” is a finding. “The form is too long” is a hypothesis. “We should remove fields” is a proposed intervention. Teams often collapse all three into one confident sentence and call it evidence.

The live work gives that sentence a chance to fail productively. Follow one real case from beginning to end. Watch what someone has to find, remember, translate, wait for, re-enter or quietly correct. Notice where the official process ends and the actual process begins. A customer may leave a form because a field is long, but also because the surrounding explanation makes the request feel risky. An employee may take too long to produce a report because a data source is unreliable, because a manager changes the question late, or because the report answers a question no one has clearly owned. Similar metrics can conceal different causes.

This is why direct observation should not be confused with a preference for anecdote over data. A dashboard can show a pattern across thousands of cases. A single case can reveal the sequence that the dashboard cannot show. They do different jobs. The question is whether the information you have can support the decision you are about to make. If it cannot, look closer before acting larger.

Leonardo’s drawings offer an apt image for this work because a drawing forces choices into relation. Direction, proportion, force and interruption become visible together. A leader does not need artistic skill to use the equivalent. A timeline, annotated screenshot, customer journey, process map, decision log or side-by-side comparison can put a system on a surface where people can inspect it together. The value is not the neatness of the artifact. It is whether it makes a previously invisible question available.

What the evidence can and cannot tell you

The historical record supports saying that Leonardo worked across several fields and used drawings and notes as part of his investigations. It does not support turning every unfinished page into a modern leadership rule. Nor does it let us know what he would recommend for a product launch, an annual plan or an AI policy. Historical examples can sharpen a question; they cannot validate a management prescription by association.

That restraint improves the present-day lesson. Observation is valuable not because Leonardo was exceptional, but because complex work produces a predictable gap between a summary and the situation it summarizes. A performance report can register the lag; it cannot automatically tell you which handoff created it. A customer survey can reveal dissatisfaction; it cannot settle whether the relevant problem is expectation, price, reliability, language, access or a different need entirely. A model can detect an anomaly; it cannot decide whether the anomaly is an error, a valuable exception or evidence that the category is wrong.

Evidence has a shape. Aggregated numbers are strong for scale, comparison and trend. Records from a particular case are strong for sequence and context. Interviews can surface missing language, incentives and workarounds. Experiments can test a proposed cause, provided the measure and conditions are clear. None of these forms is self-interpreting. Their limits should influence the size of the action.

For example, imagine an operations team sees that a recurring task takes much longer on Fridays. That pattern deserves attention, but it does not yet justify a staffing change. The team might inspect several Friday cases, compare the inputs with other days and ask the people doing the work what changes in the queue. Perhaps a weekly approval arrives late. Perhaps an upstream system is slower. Perhaps the team batches a different type of request. The first metric did its job by identifying a question. It did not answer it.

The same discipline matters when people invoke tools that seem to create instant insight. A generative AI system can summarize messages, draft a report or classify a queue. Before asking whether it can automate a role, map the exact decision it would touch. What information is input? Which parts require judgment? Who will notice a harmful error? What exception must still reach a person? What evidence would justify expanding its use? Observation makes those questions concrete. It turns enthusiasm or anxiety about a tool into an inspectable design choice.

There is also a limit to observation itself. Watching people can be intrusive, and local evidence can be mistaken for the whole. When a decision concerns safety, health, legal rights, discrimination, significant money or a person’s livelihood, direct observation does not replace relevant specialist advice, formal evidence or appropriate safeguards. The goal is not to elevate the observer’s confidence. It is to make the next decision proportionate to what is actually known.

The distinction that changes the next move

The useful distinction is between recognition and observation. Recognition is fast: “I know what this is.” It is essential for functioning in a busy world. Observation begins when recognition is no longer enough: “What would I have to notice before I could responsibly act on this?”

Recognition makes categories efficient. Observation tests whether the case belongs in the category at all. A leader recognizes “a communication problem.” Observation asks whether people received the message, understood it, trusted it, had the authority to act on it, or were prevented by a conflicting priority. A leader recognizes “resistance to change.” Observation asks what loss, risk, workload, contradiction or missing capability the resistance is registering. Sometimes the original label survives inspection. Often it becomes more precise, and precision changes the remedy.

The distinction protects teams from a common form of false speed. Under pressure, naming a problem feels like progress because it makes a response available. But a fast response to a weak model is not necessarily faster overall. It can create rework, damage trust and make people who saw the problem early less likely to speak next time. A short pause for evidence is not hesitation when it prevents a large error.

It also changes the role of disagreement. When people argue at the level of labels—“this is a quality problem,” “no, it is a staffing problem”—they may be defending rival stories. Bring the sequence into view and the argument can become more useful. Where did the case diverge? What did the person know at that moment? Which constraint was active? What happened next? A shared map does not erase judgment, but it gives judgment something public to work on.

This is a better lesson to take from Leonardo than the romance of individual brilliance. A healthy organization does not wait for one gifted person to see through the confusion. It gives people close to the work a way to report what the model misses; it makes their evidence discussable; and it asks leaders to revise an explanation without treating revision as loss of status. Observation becomes a collective capability.

A practical way to test your interpretation

Use a small field study when a recurring issue has acquired a vague label. The aim is not to document everything. It is to test whether the label is good enough to guide the next action.

First, choose one recent, representative case. “Representative” does not mean average; it means a case that contains the decision or failure you need to understand. Gather only the records a participant could have seen at the time: the request, relevant messages, system state, handoffs and outcome. This protects the review from hindsight. It is easy to say what should have happened after the result is known.

Second, reconstruct the sequence in plain language. What entered the system? Who made each decision? What had to wait? Where did someone improvise? What changed? Keep uncertainty visible. A blank space on the map is often more useful than an invented explanation.

Third, invite correction from people who know different parts of the work. Ask not “Do you agree with this?” but “What does this leave out?” and “Where would this map mislead a new person?” Those questions are less likely to force local expertise into a senior person’s framing. They also reveal whether the official process and the working process have diverged.

Fourth, name one interpretation and one disconfirming sign. Suppose the team thinks a slow handoff is caused by unclear ownership. The disconfirming sign might be that ownership is clear in cases that still stall, pointing instead to missing information or a policy constraint. This does not demand scientific certainty from ordinary work. It demands the intellectual courtesy of saying what could change your mind.

Finally, select the smallest test the evidence supports. It may be a clarified decision owner, a revised instruction, a prototype, a different review route or a two-week sample. State what will be measured, who will inspect the result and when the team will decide whether to continue, adjust or stop. A good observation practice ends in a bounded experiment or an explicit decision to leave the system alone; it does not become permanent study.

The method is deliberately light. A beautiful process diagram that nobody uses is only a new artifact. The test of the work is whether it improves a question, protects a person from an avoidable burden or makes a subsequent choice more accurate.

Common mistakes that create false certainty

The first mistake is observing only the official process. Policy documents, project boards and dashboards describe what a system is supposed to do. They may be necessary, but they are not a substitute for seeing what people actually do when information is missing, a customer is anxious or a deadline changes. Informal workarounds are not always evidence of poor discipline. They may be the only thing keeping a flawed system usable.

The second is treating a vivid exception as proof of a general rule. One difficult case can expose a blind spot. It cannot, by itself, establish prevalence. Keep both questions alive: what does this case show that our model missed, and how often does the condition occur? The first question protects learning; the second protects proportion.

Third, do not borrow an analogy and then force reality to fit it. Leonardo’s range across art, anatomy and engineering can encourage cross-domain thinking, but analogy is an invitation to investigate, not an answer. A hospital triage model may suggest a question about priority, but customers are not patients. A production line may illuminate flow, but people are not inventory. Write down what is portable, what differs and what evidence would show that the transfer has failed.

Fourth, beware observation theatre. A leader spends a morning with a team, hears a few stories and returns to the original plan unchanged. This can be worse than not visiting because it teaches people that their direct knowledge is ceremonial. Before entering the work, say what decision is open and what information could alter it. Afterward, show what was learned, including where the original explanation held up. Closing that loop gives observation credibility.

Finally, do not make inquiry a way to avoid responsibility. Curiosity has a pleasant side: every new question postpones the exposure of choosing. Set a time boundary. Decide what level of confidence is necessary for a reversible test and what level is necessary for a consequential commitment. If the evidence will not support the desired decision, either reduce the decision’s scope or obtain better evidence. Do not keep collecting detail merely because certainty would feel safer.

Choose the smallest responsible next step

The point of looking closely is not to become permanently cautious. It is to act with less fiction. A team that has watched the work, separated observations from explanations and named what could disprove its interpretation can move faster with a small test than a team arguing from slogans.

Choose one issue this week that people describe with a broad phrase: low adoption, weak quality, slow delivery, too many meetings, AI readiness. Do not start by asking for solutions. Pick a real case. Reconstruct its sequence. Ask the people closest to it what the map gets wrong. Then make one decision that is small enough to revise and clear enough to assess.

The historical material about Leonardo supports admiration for patient inquiry; it does not certify a leadership technique. The useful conclusion is more ordinary and more demanding. Before a label hardens into a project, give reality a chance to interrupt it. Put the system where people can see it. Treat exceptions as questions, not inconveniences. Record what would change your mind. Then turn the better explanation into a proportionate action.

That practice does not eliminate uncertainty. It makes uncertainty usable. It replaces the performance of instant insight with a habit of noticing, testing and revising that more people can share. In work that affects customers, colleagues and scarce resources, that is often the most responsible form of leadership available.

Sources

• Royal Collection Trust: Leonardo da Vinci—A Life in Drawing
• Victoria and Albert Museum: Leonardo da Vinci’s notebooks

The management applications in this article are editorial synthesis. They are not claims that Leonardo da Vinci used or endorsed a modern leadership method.