The leadership challenge is increasingly not acquiring information but knowing which information matters and when there is enough evidence to act.
After more than three decades working in decision analytics, I have developed a healthy respect for data. I have also developed a healthy skepticism about what we sometimes expect data to do.
Across more than 1,800 research projects, I have worked with organizations trying to understand customers, employees, markets, economic impacts, investment opportunities, public opinion, organizational performance, and a wide range of other complex issues. I have also spent much of my academic career teaching quantitative analysis and strategy. Although the settings have been different, the fundamental objective has usually been the same. That is, to use evidence to make better decisions.
That experience has convinced me of something that may sound unusual coming from someone who has spent his career working with data. More data does not necessarily produce a better decision.
Sometimes additional information improves our understanding considerably. Other times it simply gives us more information to process. The distinction is becoming increasingly important because organizations now have access to volumes of data that would have been difficult to imagine even a generation ago, and artificial intelligence will expand that analytical capacity exponentially.
For many organizations, the central challenge is therefore changing. It is no longer simply, “How do we get enough information?” Increasingly, it is, “How do we determine which information actually matters?”
Start With the Decision, Not the Data
One of the easiest mistakes in analytics is to begin with the data that happen to be available. An organization has customer data, so it analyzes customers. It has employee survey results, so it examines employee attitudes. It has hundreds of operating measures, so it builds another dashboard.
There is nothing inherently wrong with any of those activities. But they can reverse the sequence that I believe is most useful. The better starting point is not the data. It is the decision the organization is trying to make.
A request for research may begin with a statement such as, “We need to understand our customers better.” That is a reasonable objective, but it is not yet a decision problem. The more important question is what the organization intends to do differently once it understands those customers better. Is management deciding where to invest? Which customers to target? Which aspects of the experience to improve? Whether to enter a new market? How to allocate marketing resources? How to keep certain customers from leaving?
Once the decision becomes clear, the nature of the analysis changes. So does the amount and type of data that is required for the decision. Over the years, I have seen organizations collect enormous amounts of information before adequately defining the problem they are trying to solve. The result can be technically impressive research that is only loosely connected to the actual management decision.
Before commissioning an analysis, I usually want one practical point clarified. I ask which management choice the evidence is meant to inform. If that cannot be stated clearly, the analytical work is probably beginning before the decision problem has been fully defined.
Not Everything We Can Measure Matters
Modern organizations have become remarkably good at measuring things, and that represents genuine progress. But measurement can also create an illusion of importance. A metric can be precise without being consequential, a relationship can be statistically significant without being important or relevant, and a dashboard can contain dozens of indicators while still failing to tell an executive what deserves attention.
The distinction between statistical significance and managerial significance is particularly important to me because I have spent much of my career working with quantitative models. Statistical methods help us determine whether observed relationships are likely to represent something meaningful rather than random variation. That is valuable, but executives have another question to answer. Is the relationship large enough or consequential enough to change what we do?
Those are different tests. An analysis might identify a relationship between two variables with a high degree of statistical confidence. The finding can be methodologically sound while having very little practical effect on organizational performance. Conversely, an emerging issue may not yet provide the volume of observations we would prefer statistically, but the potential consequences of overlooking it may be substantial.
That is one of the places where analytics and leadership judgment meet. Good analysts understand what the model says, why it says it, and where its limitations lie. Good decision-makers understand what those findings mean within the larger context of the organization.
Dashboards Are Useful. They Are Not Decisions.
I am an advocate of dashboards when they are designed well. A good dashboard can help leaders identify trends, recognize exceptions, track performance, and see emerging problems much faster than a traditional report.
Dashboards also create risks. One is that what can be measured most easily begins receiving the most managerial attention. Another is that descriptive information can gradually be confused with explanation.
Knowing that customer satisfaction declined tells us something important, but it does not tell us why satisfaction declined. Knowing that employee turnover increased identifies a problem, but it does not tell us what caused employees to leave. Knowing that sales weakened in a market describes an outcome, but it does not establish whether the cause was price, competition, changing customer preferences, execution, economic conditions, or some combination of those factors.
A dashboard is therefore often the beginning of inquiry rather than the end of it. It tells us where we may need to look more carefully, but it does not necessarily tell us what we will find once we begin looking.
That distinction will become even more important as organizations adopt more sophisticated analytics and AI-generated summaries. A polished visualization or confidently written AI explanation can create an impression of certainty that the underlying evidence may not warrant. Executives consequently need to become more curious about how conclusions were reached, not less.
What exactly are we measuring? What population does the analysis represent? What information might be missing? Are we observing correlation or something closer to causation? How meaningful is the effect? What assumptions are embedded in the model? Could another explanation plausibly produce the same result?
Those questions are not signs that we distrust data. They are signs that we respect what good analysis requires.
More Information Can Eventually Make a Decision Harder
Additional information helps until it doesn’t. Past a point, the next report adds explanations without adding clarity and gathering more information becomes a way to postpone the decision. Organizations that stay there long enough start to confuse motion with judgment, and the people responsible for the decision start believing that another study is safer than a choice.
Every additional variable creates another possible explanation. Every additional report introduces another perspective. Every additional scenario creates another version of the future. Eventually, a leadership team can reach a point where gathering more information feels productive but is actually postponing the decision.
I have watched this happen in my public-opinion work here in Mississippi. When people were given a clearer picture of what a proposed policy would mean in practice, support did not rise. It fell. The additional detail did not produce a more persuadable public. It produced a more decided one. A leadership team that treated that second wave of data as an education problem was reading the evidence backward. The information had done its job. The public had already made a judgment.
A 2019 McKinsey survey of more than 1,200 managers found that fewer than half considered their organizations’ decisions timely, while 61 percent said at least half of the time devoted to decision-making was ineffective. More revealing than either statistic is the organizational pattern behind them. Decision processes lose effectiveness when ownership is diffuse, deliberations become unnecessarily broad, and information accumulates faster than leaders can separate what matters from what does not. (De Smet, Jost, & Weiss, 2019).
Those findings are consistent with what I have observed in practice. Sometimes a request for one more analysis is completely justified because the additional evidence could materially change the decision. At other times, another analysis is less about improving the decision and more about avoiding the responsibility of making it. Leaders have to know the difference.
There Is No Dataset From the Future
Strategic decisions pose a particular analytical challenge because the information we would most like to have often does not yet exist. We can examine historical customer behavior, analyze competitors, estimate demand, build forecasting models, test assumptions, calculate probabilities, and construct scenarios. All of those activities can substantially improve a decision. But strategy is ultimately concerned with the future, and the future has not yet generated its dataset.
Consider the decisions that regularly reach senior leadership. Should we enter a new market? Make a substantial capital investment? Launch a new product or service? Reorganize? Adopt an emerging technology? Pursue an opportunity before competitors do? These are precisely the decisions for which complete information is least likely to be available.
That does not diminish the importance of analytics. It clarifies its role. Analytics earns its value when it narrows the range of plausible outcomes and helps leaders choose with greater discipline than intuition alone would allow. It cannot remove uncertainty from a future that has not happened.
That is a much more realistic standard. It also requires organizations to become better at recognizing when they have enough evidence to act, because leaders who require certainty before making consequential decisions will frequently discover that certainty arrived only after the opportunity had passed.
I have also seen the other version of the same mistake. A team kept commissioning another round of research it already had, and the room grew more confident while the choice in front of them did not move. They were not short of data. They were using more analysis as a way to postpone a decision they already had enough evidence to make.
Artificial Intelligence Will Make This Challenge More Important
Artificial intelligence will expand our analytical capabilities considerably, and that is one of the developments I find most promising about the technology. Researchers and executives can interrogate information faster, analyze unstructured data that previously sat largely unused, explore scenarios, compare alternatives, challenge assumptions, and undertake analyses that once required substantial time and specialized resources.
Greater analytical capacity, however, creates an interesting paradox. The easier analysis becomes to produce, the more analysis organizations are likely to produce. As that volume increases, distinguishing useful intelligence from analytical noise becomes more important.
It is not difficult to imagine an executive team asking an AI system to generate multiple market analyses, strategic scenarios, financial models, competitive assessments, and recommended courses of action within a single morning. That capability is impressive, but the existence of the analysis does not mean the decision has been made.
Someone still must evaluate the assumptions. Someone still has to determine which objectives matter most, which risks are acceptable, which tradeoffs the organization is willing to make, and whether the recommendation fits the realities of the organization. AI can substantially improve the evidence available to leaders, but it does not eliminate the need for leadership judgment.
From Data-Driven to Data-Informed
This is why I have long preferred the term data-informed decision-making to data-driven decision-making. The distinction may appear semantic, but I believe it reflects something important about how organizations should use evidence.
The phrase data-driven can suggest that sufficiently good data inevitably point toward one correct answer and that the leader’s responsibility is simply to follow where the evidence leads. Occasionally, a decision really is that straightforward. Most decisions requiring senior leadership are not.
They involve competing objectives, imperfect information, timing considerations, organizational capabilities, risk tolerance, stakeholder interests, resource constraints, and consequences that cannot always be reduced to a single measure. Data should influence those decisions aggressively. It should challenge intuition, expose faulty assumptions, reveal relationships we might otherwise miss, and force us to reconsider beliefs that are inconsistent with the evidence. But evidence should not create the illusion that judgment is no longer necessary.
Michael Luca and Amy Edmondson argue that leaders should test whether an analysis answers the decision question and whether findings from one setting apply in another. They recommend examining causal claims, the sample and context, the outcome being measured, and evidence that could challenge the initial conclusion before deciding how much weight to give a result (Luca & Edmondson, 2024).
The practical lesson is straightforward. Leaders should neither surrender judgment to solely analytics nor dismiss inconvenient evidence. They should test the evidence, then decide what it means for the choice in front of them.
What Better Decision-Making Looks Like
After three decades in this field, I have become less impressed by the sheer quantity of information available to an organization and more interested in the discipline with which leaders use it. Organizations that make consistently strong decisions tend to be clear about the choice they face, the evidence that could change it, and the limitations of the information available. They surface assumptions, ask what would have to be true for a preferred option to be wrong, and weigh the costs of both acting and waiting.
None of that requires the largest dataset or the most sophisticated dashboard. It requires clarity, intellectual discipline, and leaders who are comfortable acknowledging that evidence has limits.
I have spent most of my professional career advocating for more rigorous use of research and analytics in decision-making, and I remain convinced that organizations make better choices when they replace unsupported assumptions with evidence. Experience has also taught me, however, that the value of data does not reside in its volume.
Its value lies in whether it helps us understand something consequential that we did not understand before and whether that understanding improves the decision we ultimately make. That distinction will become increasingly important as our ability to generate analysis grows. The future will give leaders more data, more models, more forecasts, more dashboards, more scenarios, and more AI-generated recommendations than any generation of executives before them.
The advantage will not go to the organization with the largest stack of research reports. It will go to the one that can tell when the next study would only make the room more comfortable.
References
Luca, M., & Edmondson, A. C. (2024, September–October). Where Data-Driven Decision-Making Can Go Wrong. Harvard Business Review. https://hbr.org/2024/09/where-data-driven-decision-making-can-go-wrong
De Smet, A., Jost, G., & Weiss, L. (2019, May 1). Three Keys to Faster, Better Decisions. McKinsey Quarterly. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/three-keys-to-faster-better-decisions
The leadership challenge is increasingly not acquiring information but knowing which information matters and when there is enough evidence to act.
After more than three decades working in decision analytics, I have developed a healthy respect for data. I have also developed a healthy skepticism about what we sometimes expect data to do.
Across more than 1,800 research projects, I have worked with organizations trying to understand customers, employees, markets, economic impacts, investment opportunities, public opinion, organizational performance, and a wide range of other complex issues. I have also spent much of my academic career teaching quantitative analysis and strategy. Although the settings have been different, the fundamental objective has usually been the same. That is, to use evidence to make better decisions.
That experience has convinced me of something that may sound unusual coming from someone who has spent his career working with data. More data does not necessarily produce a better decision.
Sometimes additional information improves our understanding considerably. Other times it simply gives us more information to process. The distinction is becoming increasingly important because organizations now have access to volumes of data that would have been difficult to imagine even a generation ago, and artificial intelligence will expand that analytical capacity exponentially.
For many organizations, the central challenge is therefore changing. It is no longer simply, “How do we get enough information?” Increasingly, it is, “How do we determine which information actually matters?”
Start With the Decision, Not the Data
One of the easiest mistakes in analytics is to begin with the data that happen to be available. An organization has customer data, so it analyzes customers. It has employee survey results, so it examines employee attitudes. It has hundreds of operating measures, so it builds another dashboard.
There is nothing inherently wrong with any of those activities. But they can reverse the sequence that I believe is most useful. The better starting point is not the data. It is the decision the organization is trying to make.
A request for research may begin with a statement such as, “We need to understand our customers better.” That is a reasonable objective, but it is not yet a decision problem. The more important question is what the organization intends to do differently once it understands those customers better. Is management deciding where to invest? Which customers to target? Which aspects of the experience to improve? Whether to enter a new market? How to allocate marketing resources? How to keep certain customers from leaving?
Once the decision becomes clear, the nature of the analysis changes. So does the amount and type of data that is required for the decision. Over the years, I have seen organizations collect enormous amounts of information before adequately defining the problem they are trying to solve. The result can be technically impressive research that is only loosely connected to the actual management decision.
Before commissioning an analysis, I usually want one practical point clarified. I ask which management choice the evidence is meant to inform. If that cannot be stated clearly, the analytical work is probably beginning before the decision problem has been fully defined.
Not Everything We Can Measure Matters
Modern organizations have become remarkably good at measuring things, and that represents genuine progress. But measurement can also create an illusion of importance. A metric can be precise without being consequential, a relationship can be statistically significant without being important or relevant, and a dashboard can contain dozens of indicators while still failing to tell an executive what deserves attention.
The distinction between statistical significance and managerial significance is particularly important to me because I have spent much of my career working with quantitative models. Statistical methods help us determine whether observed relationships are likely to represent something meaningful rather than random variation. That is valuable, but executives have another question to answer. Is the relationship large enough or consequential enough to change what we do?
Those are different tests. An analysis might identify a relationship between two variables with a high degree of statistical confidence. The finding can be methodologically sound while having very little practical effect on organizational performance. Conversely, an emerging issue may not yet provide the volume of observations we would prefer statistically, but the potential consequences of overlooking it may be substantial.
That is one of the places where analytics and leadership judgment meet. Good analysts understand what the model says, why it says it, and where its limitations lie. Good decision-makers understand what those findings mean within the larger context of the organization.
Dashboards Are Useful. They Are Not Decisions.
I am an advocate of dashboards when they are designed well. A good dashboard can help leaders identify trends, recognize exceptions, track performance, and see emerging problems much faster than a traditional report.
Dashboards also create risks. One is that what can be measured most easily begins receiving the most managerial attention. Another is that descriptive information can gradually be confused with explanation.
Knowing that customer satisfaction declined tells us something important, but it does not tell us why satisfaction declined. Knowing that employee turnover increased identifies a problem, but it does not tell us what caused employees to leave. Knowing that sales weakened in a market describes an outcome, but it does not establish whether the cause was price, competition, changing customer preferences, execution, economic conditions, or some combination of those factors.
A dashboard is therefore often the beginning of inquiry rather than the end of it. It tells us where we may need to look more carefully, but it does not necessarily tell us what we will find once we begin looking.
That distinction will become even more important as organizations adopt more sophisticated analytics and AI-generated summaries. A polished visualization or confidently written AI explanation can create an impression of certainty that the underlying evidence may not warrant. Executives consequently need to become more curious about how conclusions were reached, not less.
What exactly are we measuring? What population does the analysis represent? What information might be missing? Are we observing correlation or something closer to causation? How meaningful is the effect? What assumptions are embedded in the model? Could another explanation plausibly produce the same result?
Those questions are not signs that we distrust data. They are signs that we respect what good analysis requires.
More Information Can Eventually Make a Decision Harder
Additional information helps until it doesn’t. Past a point, the next report adds explanations without adding clarity and gathering more information becomes a way to postpone the decision. Organizations that stay there long enough start to confuse motion with judgment, and the people responsible for the decision start believing that another study is safer than a choice.
Every additional variable creates another possible explanation. Every additional report introduces another perspective. Every additional scenario creates another version of the future. Eventually, a leadership team can reach a point where gathering more information feels productive but is actually postponing the decision.
I have watched this happen in my public-opinion work here in Mississippi. When people were given a clearer picture of what a proposed policy would mean in practice, support did not rise. It fell. The additional detail did not produce a more persuadable public. It produced a more decided one. A leadership team that treated that second wave of data as an education problem was reading the evidence backward. The information had done its job. The public had already made a judgment.
A 2019 McKinsey survey of more than 1,200 managers found that fewer than half considered their organizations’ decisions timely, while 61 percent said at least half of the time devoted to decision-making was ineffective. More revealing than either statistic is the organizational pattern behind them. Decision processes lose effectiveness when ownership is diffuse, deliberations become unnecessarily broad, and information accumulates faster than leaders can separate what matters from what does not. (De Smet, Jost, & Weiss, 2019).
Those findings are consistent with what I have observed in practice. Sometimes a request for one more analysis is completely justified because the additional evidence could materially change the decision. At other times, another analysis is less about improving the decision and more about avoiding the responsibility of making it. Leaders have to know the difference.
There Is No Dataset From the Future
Strategic decisions pose a particular analytical challenge because the information we would most like to have often does not yet exist. We can examine historical customer behavior, analyze competitors, estimate demand, build forecasting models, test assumptions, calculate probabilities, and construct scenarios. All of those activities can substantially improve a decision. But strategy is ultimately concerned with the future, and the future has not yet generated its dataset.
Consider the decisions that regularly reach senior leadership. Should we enter a new market? Make a substantial capital investment? Launch a new product or service? Reorganize? Adopt an emerging technology? Pursue an opportunity before competitors do? These are precisely the decisions for which complete information is least likely to be available.
That does not diminish the importance of analytics. It clarifies its role. Analytics earns its value when it narrows the range of plausible outcomes and helps leaders choose with greater discipline than intuition alone would allow. It cannot remove uncertainty from a future that has not happened.
That is a much more realistic standard. It also requires organizations to become better at recognizing when they have enough evidence to act, because leaders who require certainty before making consequential decisions will frequently discover that certainty arrived only after the opportunity had passed.
I have also seen the other version of the same mistake. A team kept commissioning another round of research it already had, and the room grew more confident while the choice in front of them did not move. They were not short of data. They were using more analysis as a way to postpone a decision they already had enough evidence to make.
Artificial Intelligence Will Make This Challenge More Important
Artificial intelligence will expand our analytical capabilities considerably, and that is one of the developments I find most promising about the technology. Researchers and executives can interrogate information faster, analyze unstructured data that previously sat largely unused, explore scenarios, compare alternatives, challenge assumptions, and undertake analyses that once required substantial time and specialized resources.
Greater analytical capacity, however, creates an interesting paradox. The easier analysis becomes to produce, the more analysis organizations are likely to produce. As that volume increases, distinguishing useful intelligence from analytical noise becomes more important.
It is not difficult to imagine an executive team asking an AI system to generate multiple market analyses, strategic scenarios, financial models, competitive assessments, and recommended courses of action within a single morning. That capability is impressive, but the existence of the analysis does not mean the decision has been made.
Someone still must evaluate the assumptions. Someone still has to determine which objectives matter most, which risks are acceptable, which tradeoffs the organization is willing to make, and whether the recommendation fits the realities of the organization. AI can substantially improve the evidence available to leaders, but it does not eliminate the need for leadership judgment.
From Data-Driven to Data-Informed
This is why I have long preferred the term data-informed decision-making to data-driven decision-making. The distinction may appear semantic, but I believe it reflects something important about how organizations should use evidence.
The phrase data-driven can suggest that sufficiently good data inevitably point toward one correct answer and that the leader’s responsibility is simply to follow where the evidence leads. Occasionally, a decision really is that straightforward. Most decisions requiring senior leadership are not.
They involve competing objectives, imperfect information, timing considerations, organizational capabilities, risk tolerance, stakeholder interests, resource constraints, and consequences that cannot always be reduced to a single measure. Data should influence those decisions aggressively. It should challenge intuition, expose faulty assumptions, reveal relationships we might otherwise miss, and force us to reconsider beliefs that are inconsistent with the evidence. But evidence should not create the illusion that judgment is no longer necessary.
Michael Luca and Amy Edmondson argue that leaders should test whether an analysis answers the decision question and whether findings from one setting apply in another. They recommend examining causal claims, the sample and context, the outcome being measured, and evidence that could challenge the initial conclusion before deciding how much weight to give a result (Luca & Edmondson, 2024).
The practical lesson is straightforward. Leaders should neither surrender judgment to solely analytics nor dismiss inconvenient evidence. They should test the evidence, then decide what it means for the choice in front of them.
What Better Decision-Making Looks Like
After three decades in this field, I have become less impressed by the sheer quantity of information available to an organization and more interested in the discipline with which leaders use it. Organizations that make consistently strong decisions tend to be clear about the choice they face, the evidence that could change it, and the limitations of the information available. They surface assumptions, ask what would have to be true for a preferred option to be wrong, and weigh the costs of both acting and waiting.
None of that requires the largest dataset or the most sophisticated dashboard. It requires clarity, intellectual discipline, and leaders who are comfortable acknowledging that evidence has limits.
I have spent most of my professional career advocating for more rigorous use of research and analytics in decision-making, and I remain convinced that organizations make better choices when they replace unsupported assumptions with evidence. Experience has also taught me, however, that the value of data does not reside in its volume.
Its value lies in whether it helps us understand something consequential that we did not understand before and whether that understanding improves the decision we ultimately make. That distinction will become increasingly important as our ability to generate analysis grows. The future will give leaders more data, more models, more forecasts, more dashboards, more scenarios, and more AI-generated recommendations than any generation of executives before them.
The advantage will not go to the organization with the largest stack of research reports. It will go to the one that can tell when the next study would only make the room more comfortable.
References
Luca, M., & Edmondson, A. C. (2024, September–October). Where Data-Driven Decision-Making Can Go Wrong. Harvard Business Review. https://hbr.org/2024/09/where-data-driven-decision-making-can-go-wrong
De Smet, A., Jost, G., & Weiss, L. (2019, May 1). Three Keys to Faster, Better Decisions. McKinsey Quarterly. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/three-keys-to-faster-better-decisions