Good Product Decisions vs Cognitive Biases

Alexandra Lung, Romina Chitic at UXDX EMEA. Video: https://youtu.be/_wJwq8W36k4

Readable transcript: edited from the recording's captions for readability (fillers and false starts removed, punctuation and section headings added). Wording is the speaker's own. Timestamps are positions in the video. Names marked [?] could not be verified against the audio.

188 biases, and why not all of them are bad

[00:00:00] Alexandra: 188. What's up with this number? Why did we put it up there? This is the number of biases referred to in the Cognitive Bias Codex. That's a picture of it here. And as we keep learning more about these we can keep adding to this list. It's important to know, however, that not all biases are bad. Biases are shortcuts, they are ways that our brains have evolved in order to make quicker decisions. And what we're going to look at today is some stories surrounding our biases and how we can better balance them out and build better products by learning how to do that.

[00:00:38] Alexandra: As a disclaimer, this is really hard. As everything that has to do with us being more aware of ourselves and our reactions, thoughts and emotions, it's a really, really hard thing to do. But it's also a very valuable thing to do, especially when working in product and design.

The framing effect in user interviews

[00:01:00] Alexandra: So in order to really get to know our users really well, the best thing is of course to go talk to them and observe them and really get to know them. But it's also during these interactions, and a lot during user interviews as well, that we see a lot of the biases kick in. And one of the biases that we see happening a lot in user interviews is the framing effect. The framing effect is based on the fact that the framing of the inquiries can influence our responses. Basically, by the way we ask the questions we're going to lead the user into answering in a certain way. So simply put, as human beings we do not make choices in isolation. We are highly dependent on the way that the question is asked or the information is presented to us.

[00:01:54] Alexandra: And here's an example. I'm sure you've all heard these kinds of examples: do you like this product? It's one of these questions where you can pretty safely assume that the answer will be yes, but most of the time it will be a yes that is pretty light, without taking all the different aspects into account. For example, a couple of days ago I was passing by a perfume shop and someone made me smell a perfume and they were like, do you like it? And I was like, yes. And I was just passing by, I wasn't thinking of, do I really like it, would I really buy it, how much does it cost, is this something that I actually need? I wasn't thinking of all of that, I just threw the yes out there. And I know how they use that information.

[00:02:42] Alexandra: But anyhow, it's also very linked to the users' own biases. Most of the time as human beings we want to be liked and we do not want to disappoint or hurt the person's feelings. So if someone asks us if we like their product, it's very, very rare that someone really says no. Something that you can do in order to have a bit of a less biased response to this is to ask things like, can you describe the last time you've used this product? Or, can you describe the last time you've done a certain action? Or, how do you feel when you use this product? These are both questions that are going to yield really better, unbiased results.

Confirmation bias during research

[00:03:31] Alexandra: Another example of a bias that happens quite a lot during user research is confirmation bias. In confirmation bias we're going to basically jump to the first piece of information that confirms something that we already believe. A lot of times we can see people cherry picking information or ignoring certain bits of information in order to confirm a certain belief. It basically looks like this: well, you've heard something, you want to do your own research, but you literally just jump on the first thing that already agrees with you.

[00:04:12] Alexandra: So confirmation bias in user research, we can see it in different forms. Here I took an example of more like discussion testing[?], and here's one of the examples: what do you think of this simple menu? And the person would answer something like, easy to think[?], easy to use, I think it's as simple as you said. So most of the time — well, here I think we're actually making a mix of confirmation bias and framing bias — most of the time the person in front of us will answer with simple, as we already said, confirming what we actually want to hear.

[00:04:57] Alexandra: Another example could be, what do you expect when you click here? And the person in front of us says something like, I think I'll see more details. But, good answer, let's move on. So really, once we hear a first answer, and most of the time that also kind of confirms what we're looking for, we are just going to cut the person in front of us off. And a really, really important thing to do, especially when we're talking to our users, is to give them time to express themselves, put them at ease, but also really listen. And the moments where we see them hesitating[?] are also the good moments where we need to dig in, in order to understand better their thoughts and the job that they want to do.

How to avoid these biases in interviews

[00:05:52] Alexandra: How can we avoid these biases during the user interviews? One of the things you can do is to dry run the interview with someone from your team. This helps you with things like, the more practical aspect is that you're going to be able to see more or less how much time it takes to run the interview. And the second one is that if you do this with someone who you're familiar with, and you're in a dry run situation, you can take more of your brain space during the interview in order to observe yourself, the way you ask the questions, the way the other person answers the questions. Or you can even ask the other person, if she's pretty experienced with user research and biases, they can also give some feedback linked to this.

[00:06:41] Alexandra: Another thing that is really important to do is to list all the assumptions before you begin the research. This can seem very normal, logical, but this doesn't happen all the time, and I've seen a lot of situations where if you do not list your assumptions and really make sure that everyone is clear of and everyone agrees with the assumption, then you're going to have a lot of debate and a lot of biases that kick in. A lot of times you'll see certain results and you'll be like, yes, yes, this validates. But is it really validating, or is it just that now all the biases of you being more attached to a certain answer, and confirmation bias, kick in?

[00:07:28] Alexandra: Another important aspect is, stop talking and listen. Really make sure that you do not use all the space by, well, first presenting your brilliant idea and selling it to the user, because that's not really the point in interviews, but also that you give enough time to the person in front of you to really think it through and be able to answer, speak their mind, and really, really listen to what they have to say.

[00:08:00] Alexandra: Another really important thing is note takers. I know that sometimes we don't always have time to have note takers to help us out with this, but if you can, it can be very, very valuable. Of course you can brief your note takers on how to take notes, more of quotes, in a really less biased way. But what's really valuable as well is that you can afterwards debate and synthesize the findings with the note takers. The times I've seen teams do this were the times where actually the results were really the less biased, and then the knowledge was really, really valuable.

[00:08:37] Alexandra: And last but not least, you can also use methods that go beyond what we hear, and really think about also what the person is doing, how they're behaving, why they're doing certain things. Like the what-how-why method, or empathy maps, that also make you think about what are they hearing, seeing, what's the context around them and their job when they're talking to you.

Sampling bias and membership bias

[00:09:04] Romina: As we've seen so far, in product design there are so many biases that happen during user interviews and during the research phase really, but most of them don't start there. A lot of our biases can come in even before a user ever steps into one of our meetings, joins one of our calls, or sees any of our mock-ups. And this affects who we talk to, whose opinion we take into consideration, but also, in the end, how well we truly understand our audience.

[00:09:34] Romina: One example of this is the sampling bias. So in sampling bias you unintentionally leave out a piece of your research audience. You just leave out some people out of the research pool. This happens to us all of the time. I have an example of this that happened to us. We were developing a four language product, and of course that's difficult, because we were trying to test it and not everyone speaks four languages, and the people in our research team did not speak all of the four languages of our audience. So how do we go about testing for a broader audience?

[00:10:08] Romina: And our first idea was, we're just going to interview people in English, because that will get us past our language barriers. And I guess you know where compromises can sometimes lead to. We very quickly noticed that we were speaking to a particular segment of the audience. We were talking to either people who were a bit more highly educated and can speak English, or we were speaking to foreign nationals, but leaving out a big chunk of our national audience.

[00:10:33] Romina: So we regrouped in the middle of the process to fix it. We spread out the interviewing process to include many interviewers with different language abilities, so that everyone could speak in their natural language. And we had a lot of people who wouldn't normally interview users, like PMs and content managers and even in some cases the CPO, interview users. And for us it was actually an enlightening exercise, because people who don't normally get to interact with our users did so, and I think it gave them a new perspective on our product. You can fix it, of course, but it's important to be aware that you're doing it.

[00:11:13] Romina: Another one that I find fits really well into this category is the membership bias. What I mean by this is that a popular person isn't necessarily a representative test participant. And if you choose your participants from a selective group of people, or from a specific group of people, the results that you get then from your research might not apply to a larger general audience.

[00:11:37] Romina: So I have an example here as well. For a bank application that I was working on, I got a list of users recruited for the interviews before I even started to really think about the process, and I noticed quickly that they all had something in common. So the people on this list were what the bank called high net worth individuals, so basically affluent people. And in the bank's view it made sense: you talk to people who have a big stake in your success and you keep them happy, and also their opinion matters because they are affluent.

[00:12:09] Romina: But we as researchers and practitioners know that these individuals are maybe not the ones who are even interacting with our applications and our systems. They might have someone do it for them, they belong to a select group. So basically their experience and their needs and expectations are not the same as the ones of the general audience. And this is what I want to highlight, that recruiting from a specific group of people can sometimes make sense from a cost perspective, as a compromised solution, but you always have to consider, how does this affect or influence the result of your general research?

[00:12:47] Romina: And there are a few things that you can do to avoid these sampling biases and membership bias. First of all, you should have a clear definition of your audience. Know who you're addressing, who we are trying to get to, and come back to this definition later on in the process, just to make sure that you haven't deviated from your original goals. Secondly, have a look at the numbers. Try to see all the data that you have available to you and create an in-depth picture of your customer base and also possible clients.

[00:13:17] Romina: And thirdly, involve a diverse team to bring in more perspective. I like this one because it's human that we only have our own experiences. We can only contribute so much to a team, and at the end of the day the things that we have personally experienced will limit the way we approach a project. And if there's anything that you could do about it in your own team, try to get a more diverse group of people involved. You'll have a wider perspective on the project and you'll probably be able to discover and address more of the needs of your users.

The impact of our biases

[00:13:50] Alexandra: So we've talked about a few different biases. You might then ask, okay, and so what? We are human, we're all biased. So what if we are very biased? So we're going to talk a bit about what is the impact that our biases can have. And the first one that can come to mind is a financial one for all different businesses. The fact that we do not listen to our users truly, or interpret the information they give us incorrectly, can lead us to build products, or even design entire new products or businesses, that will not satisfy the needs of our users or will not solve their initial problems. So we will end up with products that either don't get product-market fit or they don't grow as expected. So the first impact of our assumptions can be a negative one, financial.

[00:14:43] Alexandra: But this also goes beyond that. It goes really to impact the lives of our users. And an example is Google Maps, that quite a few years back launched what we call the cupcake feature. Basically for going from point A to point B, they were not only showing the way to go there but also how many calories you would be burning, with little pink cupcakes, saying how many of those mini cupcakes you would be burning.

[00:15:21] Alexandra: So as soon as they rolled it out, a few of the responses were very, very loud around, no, shaming people for eating cupcakes, for having little pink cupcakes out there. But also the impact was really, really strong on people that had eating disorders. And this can be really interesting, these kinds of features out there. Food and calories were really not considered something that goes into a Google Maps feature, so Google took this out very, very quickly. But I think it's still a very interesting example of how to take into account the needs of the different users, and how this can affect certain parts of your target.

[00:16:16] Alexandra: Another example is Barbie. Same, a few years back, Barbie launched Hello Barbie. So for years now kids were saying, I want Barbie to talk to me. So they created Hello Barbie, which actually was talking to kids. They trained Barbie with AI, and at the moment where they actually launched Hello Barbie she had quite limited vocabulary and topics of conversation. So a lot of times the conversation could go something like this: oh Barbie, I want to talk about, let's talk about science for example. And then Barbie would answer something like, yeah, we can talk about science, but let's talk about fashion, that's more interesting.

[00:17:05] Alexandra: So I guess you can see where I'm going with this. And of course I simplified a bit the conversation that Barbie could have, but it really went towards these kinds of assumptions on the limited conversations that Barbie could have at the beginning, and the fashion one was one of the preferred topics, let's say, or chosen topics. And I think this speaks a lot on how a product can influence, like in this case, generations of girls to come, and how important, how big our responsibility is when we build this product.

Biases in science, and decisions that outlive us

[00:17:49] Romina: Talking about biases and science, there was this recent study that proposed that they could determine whether a person would become a criminal by evaluating a single photograph of their face. What do we sacrifice to science? So more than 2,000 researchers, students and scholars opposed the publication of this study. They argued that historic forms of discrimination only get amplified by machine learning systems.

[00:18:19] Romina: And we also have to wonder who would be impacted by integrating machine learning into existing institutions in our society, given our own constantly evolving understanding of fairness and justice, and everything that we know about how the justice system works in our current world. But do we even think it would be possible to train an AI system that would be unbiased with the information that we have so far? I think someone once fairly said that first we build the tools and then the tools build us.

[00:18:54] Romina: Another example that I'm actually quite attached to is a decision that started in the 1950s, when there was a research project that aimed to determine what were the causes of heart disease and stroke. So this study, you may have heard about it because it was quite famous in the US. There were 22 countries in the original study, but only seven of them seemed to confirm what the researchers were trying to say. And here we already hear of bias slipping in.

[00:19:26] Romina: What the researchers were trying to say is that fat and cholesterol are the causes of heart disease and stroke. And because they were trying to confirm that, they left 15 of the countries that didn't correspond to that information out of the study, and only based their conclusions on seven of the countries in the study. And as a result of this one study, diets changed all over the world, and for the last 70 years we've all been seeing the conclusions of that. It took decades to reverse those beliefs. Today we know better, we know that fat isn't the enemy, that the picture of nutrition is way more complex than that. But you can see easily how decisions made in one part of the world decades ago can over time grow to affect everyone.

Biases in decision making: the sunk cost fallacy

[00:20:16] Romina: And then we have biases in decision making. We looked at recruitment and asking the right questions, but even when you've done all of that right, I think there's one hurdle left, and that's the human mind. Because we all like to think that our decisions are rational, and yet study after study tells us that that's hardly the case. Being sometimes wrong is the price that we pay for being able to make quick decisions, for being able to make decisions on the spot, and evolutionarily speaking it's better to make a bad decision than no decision at all.

[00:20:52] Romina: One bias in decision making is the sunk cost fallacy. You might have heard about this one because it's quite universal, but it is still perplexing. So you see it in simple things like going to the movies. You're watching maybe a bad movie, but you'll stay for all of it because you paid for the ticket, and rationally speaking it's maybe not worth your time, but the sunk cost effect pushes you into staying and continuing with the whole event.

[00:21:21] Romina: Or you signed up for a paid event and the topic isn't exactly what you were hoping for, or you can't follow what's been said, but the smart thing at that point might be to leave. But you spent a hundred dollars and you end up wasting your two hours as well. Even if you quit the event, let's say, your money isn't coming back, so you've already lost it. But unless you learn something in the two hours that you're there, then you're not losing only the money, you're losing also your valuable time.

[00:21:52] Romina: This we can see a lot in projects, I think especially in product decisions, where a lot of money has already been invested into a certain product that maybe isn't working out the way we wanted it to, and we will spend more time and sometimes go even deeper into that project rather than have the original investment turn into a waste. And it can be hard to understand and accept in the moment where it's happening, and even lead us to further losses. We're likely to continue in an activity if we've invested money or effort into it, and we'll even go against evidence that shows us that this is not the best decision for us. We've spent the money, we've put in the effort and we're not going to get that back by doubling down, kind of decision. But the sunk cost fallacy works on us because it's based on more than the alternatives currently in front of us. It's an emotional decision at the end of the day.

Ownership bias, and keeping our blind spots in check

[00:22:48] Alexandra: And another bias that we encounter making decisions is the ownership bias. So what comes to mind about this is that we prefer our own ideas over others, and we really get attached to our ideas. I think one of the examples that comes to mind is workshops, for example, like brainstorming. Doing workshops, a lot of times participants get attached to their ideas, depending on how seasoned they are in brainstorming and how soft they are[?]. But a lot of times you would see participants that really cling on to their ideas and want to push their ideas to be chosen in the end.

[00:23:29] Alexandra: And I think this also happens with ideas but also with different things that you do at work, like all the effort that you put in, but also more generalized, when you're building a product or a feature or a company and you become more and more attached to it, to a point that it's not very easy for you to take that step back and really make a more cold-headed decision.

[00:23:57] Alexandra: So we're getting to the end of the presentation, and I think it's really interesting to see all the different facets and all the different biases that can kick in, and we've only talked about a few of them here. But the truth is that we are also all human, so it's normal to have these biases, and we cannot get rid of these biases completely, and especially we cannot get rid of these biases overnight. What we can all do is to be conscious that we have blind spots, and to change our behavior in order to keep them in check. Keeping them in check by avoiding assumptions, listening to others and being really, really mindful. It really helps you strike that balance, and in the end really helps you become a better professional and even a better person in the end. Thank you very much.