A version of this article appeared in the Spring 2021 issue of strategy+business.
A company leader’s existence is a long series of decisions, yet many executives don’t have a systematic approach to making those decisions. They rely on intuition; they don’t have an accurate understanding of their track record over time; and they fall prey to biases, cherry-picking new information to reinforce existing beliefs rather than looking at data objectively. These are challenges that Annie Duke has experienced firsthand.
Duke studied cognitive science at the University of Pennsylvania and was a few months from completing her Ph.D. when she had to take time off for health reasons. She started playing poker, thinking initially that it would just be something with flexible hours that she could do until returning to her studies. But she turned out to be very, very good at the game. Duke played professionally for 18 years, winning a World Series of Poker championship (one of the game’s biggest prizes) in 2002. She never returned to academia, but she did merge her academic insights about decision-making with real-world applications at the poker table.
Since retiring from professional poker in 2012, Duke has consulted with organizations — many in the financial-services industry — and written books. In November, she spoke with strategy+business about her latest book, How to Decide, a rundown of the myriad misjudgments that people bring to the process.
S+B: How did you make the switch from playing poker to working with company leaders?
DUKE: In 2002, eight years into my poker career, I was asked by the founding partner of a hedge fund if I would come in and speak to their options traders about how poker might inform their thinking. It was supposed to be a discussion about risk, but I ended up talking mostly about the effects of emotion on learning — how your most immediate recent experience, whether you’re winning or losing — can distort the way you think about risk. That was the point where I started to merge this conversation between cognitive psychology, which was where I’d started, and poker, and how those might inform each other. Science gives you the theory, but in poker you have to make high-stakes, fast-paced decisions under extreme uncertainty.
S+B: A lot of your thinking about decision-making sets the concept of luck aside, even though luck has a huge impact on the outcome of a decision, both at the poker table and in life. Why not focus more on it?
DUKE: You definitely have to consider luck. But in thinking about decision-making, luck just isn’t all that interesting, because you can’t do anything about it. You should understand that luck has an influence on the way things turn out. But it’s far more important to work with what you can control. That’s where I think people should put their focus: on what they know when they’re making a decision, which beliefs are true, and where they can find corrective information.
S+B: Based on your experience talking to executives, what’s the overall quality of decision-making at most companies?
DUKE: People in general don’t do enough reflecting on their own decision processes. They just don’t think about it in an orderly way. They don’t think about the range of potential outcomes, and they definitely don’t look back on their past decisions objectively. It doesn’t matter if I’m talking to a CEO or your average person — we just don’t think about our decisions very often, and we don’t have a good process for how we make those decisions.
S+B: What specifically do they get wrong?
DUKE: It runs the gamut. A lot of people rely on their gut. That’s particularly dangerous for people who’ve been successful, because their past successes give them confidence. You see this with people who are very strong investors for a few years and then suddenly have a big setback. That’s usually a good clue that they were running off their gut and not approaching their decisions — their model of how the world works — with any kind of skepticism.
S+B: In the business context, many decisions happen through group discussions and team meetings. How does that usually go?
DUKE: People have a lot of faith in group discussions, thinking that if you bring a group of people together, those people will tell you their point of view in an honest way. But there’s substantial research showing that that isn’t what happens when a group comes together to discuss anything. Anyone who’s been in a meeting has seen this. Once consensus starts to form, it generates its own momentum. It’s like a snowball that turns into an avalanche of consensus. And at that point, people no longer offer up their true perspective.
If you’re junior, you don’t want to disagree with the boss. Even for more senior people, as you hear everyone else in the room agree, you start to question your own beliefs. Lots of evidence shows the decisions coming out of these group processes are pretty bad. But because they were decided based on a consensus, it’s almost like the decisions are hyper-certified. People think, “There were a lot of people involved in that decision, so it must be good.”
S+B: Can you be more specific about the challenges that groups face?
DUKE: Many times when people are discussing a problem, they don’t realize that they’re not actually talking about the same thing. Some of the issue has to do with language. There’s a great survey in which two researchers, Michael and Andrew Mauboussin, asked people to consider a set of descriptive terms about the likelihood of a future event happening — terms like might happen or serious possibility. The researchers then asked people to convert those terms into actual percentage odds, and the answers were all over the place. The responses for the term serious possibility ranged from a 20 percent likelihood of happening to an 80 percent possibility — a huge variation. People didn’t even agree on what words like never and always meant in terms of numerical odds.
It doesn’t matter if I’m talking to a CEO or your average person — we just don’t think about our decisions very often, and we don’t have a good process for how we make those decisions.”
And this issue doesn’t just apply to probabilities — it applies to anything that you’re making a decision about. You see the same thing happening with COVID-19. People will say, “I’m meeting up with a friend, but it’s OK because I know she’s being safe.” What does safe mean? People have wildly different interpretations of that word with regard to the pandemic. Some people let their mail sit untouched for 48 hours, while other people are flying to vacation spots and dining at indoor restaurants.
There’s a term for this: naive realism. We think that because we believe something, there’s a consensus among other people with that same understanding. But if you’re not defining the terms of a given situation in actual behaviors — concrete facts, rather than summary-level adjectives — you get a fundamental misalignment in how you understand a particular situation.
S+B: Can you give me a business-specific example?
DUKE: Let’s say you’re on a hiring committee, and one of the things you’re looking for in potential job candidates is mentorship. People on the hiring committee will likely have very different ideas about what it means to be a good mentor. So if they don’t take the time to define that term, they may rate the same candidate in very different ways. That clouds the group’s ability to decide who the best candidate is.
S+B: What’s the solution to that problem?
DUKE: You need to figure out how you can make those differences explicit. For mentorship, you can break it down into specific behaviors, so the assessment is less subjective and more pinned down to actual inputs. For example, you could define mentorship in terms of the amount of time each candidate spends in a week helping junior colleagues solve problems. Or the number of mentees each candidate currently works with. If I ask people on the hiring committee to assess candidates based on more specific information like that, I’m going to get a higher-quality decision about who to hire.
S+B: You talk in How to Decide about how systematic decision-making can be harder for smart people. Why is that?
DUKE: Smart people often have the ability to make sense of a lot of information quickly. And that really is a valuable skill. They have a model in their head about how the world works, and maybe that model has worked for a long time. Maybe it’s still pretty accurate. The challenge is that the model can become entrenched. And people are more likely to focus on the data that fits with and even reinforces their model, rather than adjusting their model to accommodate new data as it comes in.
Another issue is that smart people tend to rely on their intuition more. In some cases, that’s warranted. If I can look at a data table and quickly understand what the numbers are saying, my intuition might be correct. But intuition is risky if we’re talking about a strongly held belief that someone has. If we’re looking at a data table for something neutral, without much emotional valence — like the technical specs of different types of metals — it’s not an issue. But if I’m reading a data table about gun control, then I’m in big trouble. Because I have some strongly held beliefs about it. Incidentally, this is an issue in politics. Studies have shown that it’s often the most informed people who have the most extreme views.
S+B: For a business leader, it seems as though these issues can overlap and reinforce one another. Let’s assume the leader is smart, and that he or she will need to run group discussions that will result in decisions. What should that leader do differently in a group context?
DUKE: The whole purpose of discussing things as a group is to raise the overall knowledge level of the team. People bring different perspectives — that’s the value. So for a leader, the challenge is to get all those perspectives to collide constructively. In a way, you don’t care as much about the areas where everyone agrees; you may as well make the decision yourself with no input from anyone else. Instead, you want to identify and explore the areas where there’s disagreement and divergence.
How to do that? The key is asynchronous work — asking for individual input before the discussion starts. Let’s go back to the mentorship example. To make a better decision about job candidates based on their mentorship ability, you need to identify that ability with some precision, in terms of actual behaviors and quantifiable inputs, prior to the interviews taking place. Then, if I were leading that committee, I would email the group and ask everyone to score each candidate on a scale of zero to five for each attribute. And you should always let people add a few sentences of explanation if they want, in an open field. You get a surprising number of additional insights from qualitative responses like that.
Everyone does that scoring independently, without seeing anyone else’s responses. The leader collates the answers into areas of agreement and areas where the responses diverged. Once the meeting starts, the leader can quickly move through the consensus and spend most of the time talking about the differences, which is where the good stuff comes up. That’s the opposite of a typical meeting, where 80 percent of the time is spent talking about the consensus.
S+B: But the consensus has value, right?
DUKE: I’m not saying that there’s no valuable information in there whatsoever, which is why you allow people to see the points of agreement and discuss them briefly at the beginning of a meeting. But in the context of a group discussion, the consensus is uninteresting.
If I want to make a good decision, I want to identify people I respect who are well-informed and yet have a completely different opinion than mine. And I want to spend time exploring that and understanding their point of view. I might be wrong, and if I’m wrong, I’ll benefit from having someone correct me. But even if I’m not wrong, I’m going to benefit by explaining my beliefs to others — the underlying rationale and evidence to support them.