<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Dominika Mazur on Medium]]></title>
        <description><![CDATA[Stories by Dominika Mazur on Medium]]></description>
        <link>https://medium.com/@domi_mazur?source=rss-99164cd33a1f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*_ZdVaj0__atzYvZ_.jpeg</url>
            <title>Stories by Dominika Mazur on Medium</title>
            <link>https://medium.com/@domi_mazur?source=rss-99164cd33a1f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 15:12:06 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@domi_mazur/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[AI vs. human. How my data duel with an LLM turned into a collaboration.]]></title>
            <link>https://medium.com/@domi_mazur/i-wanted-to-fight-llms-but-we-became-user-research-buddies-instead-heres-what-happened-efdc0ebb78ed?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/efdc0ebb78ed</guid>
            <category><![CDATA[qualitative-analysis]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[qualitative-data]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[user-research]]></category>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Mon, 13 Jan 2025 02:09:39 GMT</pubDate>
            <atom:updated>2025-01-21T15:37:16.496Z</atom:updated>
            <content:encoded><![CDATA[<h4>I took ChatGPT through a series of trials to see how well it can analyze qualitative data and it changed my view on how it can be used in user research.</h4><p>It’s 2029 and the machines have taken over the world and destroyed humankind. AI military vehicles are driving over human skulls in a desert, which used to be a California metropolis, shooting lasers at people running away in the dark.</p><p>This scene from the 1984 Terminator movie gave me nightmares when I was a kid. I thought that this movie showed what was really going to happen in the future.</p><p>Later, I understood that it was just a movie, but this image stuck with me, along with the existential fear that humans can invent AI which, one day, might turn against mankind. Over the years, my childhood fear transformed into curiosity about emerging technologies and the desire to master them.</p><p>But…</p><p>when Large Language Models (LLMs) were released to the public, this old anxiety came back and the question of how AI is going to transform the humanity and my own life kept nagging at me.</p><p>To get some answers, I took an LLM through a series of trials to see how intelligent it really was and to see how well it could do my job. On each trial, the LLM had to prove its accuracy and credibility when it came to analyzing user research data. I picked user research analysis, because it’s a set of skills that I’ve mastered over a decade. This allowed me to do the trial as a side-by-side comparison with my own work and, more importantly, it let me face my existential fear that it’s possible that AI will be so smart that it can replace me in the future.</p><p>Spoiler alert, that’s not what I found.</p><p>First, I analyzed user research participant interviews by myself. Next, I asked ChatGPT to replicate what I did and prove its credibility and accuracy. This exercise was humbling and eye opening. It made me stop thinking about LLMs as my nemesis. I started to see ChatGPT as my research assistant, who has limitations, but can speed up my work and be useful with proper oversight. Here’s how ChatGPT did on each trial. If you’re curious about what I did step-by-step or would like to replicate my process, keep reading.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zub7asILhcScv-ehJ1NCmw.png" /><figcaption>Table 1. I took ChatGPT through a series of trials to see how well it could analyze qualitative user research data. This table summarizes the learnings. These findings come from my own exploration and don’t necessarily generalize to other qualitative data or analysis types, or the experience that other user researchers have had with LLMs.</figcaption></figure><p>Below, I describe what I did step by step and include the prompts I used for ChatGPT along with a few more interesting learnings which are not in the table.</p><h3><strong>About the research question and the data which was used for the analysis</strong></h3><p>I used <a href="https://digital.wpi.edu/downloads/wm117p47m?locale=en">publicly available</a> user interview transcripts to avoid putting sensitive participant and proprietary information into an LLM. The authors of the study interviewed experts and people from the local community about the traffic congestion causes in a neighborhood in Bangkok. My analysis focused on finding all of the possible causes of traffic that were discussed. I chose this research topic, because it was straightforward. I wanted to see if an LLM can handle a simple, well-researched topic where the output was to give a summary of all the causes of traffic. The interviews were also short (only 18 transcript pages) which felt like a good start.</p><p>I describe the analysis procedure step by step in Figure 1 with some commentary below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*36pEOo6EAjSIMc1KXIP-uw.png" /><figcaption>Figure 1. Narrative analysis and synthesis of user interview transcripts. Step by step walkthrough. I followed this analysis process by myself and then asked ChatGPT to replicate it. Table 1 summarizes how it did.</figcaption></figure><h3><strong>Methodology</strong></h3><p>I chose the analysis method based on my research question (<em>What causes traffic congestion in Thung Khru, Bangkok?) </em>and the data format (interview transcripts). In this case, <a href="https://atlasti.com/guides/qualitative-research-guide-part-2/content-analysis">content analysis</a> was the right fit.</p><h4><strong>Step 1. Coding qualitative data</strong></h4><p>The first step in content analysis is to code the data. Coding is about breaking qualitative data into small chunks of meaning and giving these data chunks a label. It allowed me to classify interview transcripts into digestible units and later re-organize these units of meaning into higher level themes. It also gave me the option to see patterns across participants. For example, if a majority of participants mentioned that poor infrastructure contributed to traffic, then it might be a key learning, because it’s top of mind for the majority of the people in the study.</p><p>Here’s an example of coded data:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Al22_HNtSQe2Mnas0uQ-VQ.png" /><figcaption>Table 2. Qualitative data coding. An example of how I coded the user interview transcript from participant #1 (P1).</figcaption></figure><p>After I finished coding, I gave the same job to ChatGPT. This boils down to giving ChatGPT a set of prompts.</p><p>I chose ChatGPT because I had access to the free version and the input text limit was enough to do the analysis. ChatGPT can also produce summaries in tables, which was helpful for what I needed to do.</p><p>You can <a href="https://docs.google.com/document/d/1kJV7b2SB3rcgkNGfdkiXEdq7-zXChTPPWah35kydjss/edit?usp=sharing">download the anonymized transcripts</a> and use the prompts below to replicate my steps. I copy pasted the original transcripts from a pdf. into a Google doc to anonymize them and didn’t have the time to clean up the Google doc formatting, so bear with me. The good news is that ChatGPT does not need pretty formatting to do the job.</p><p>First, I asked ChatGPT to code the participant interview transcripts. I couldn’t upload all of the transcripts at once, so I fed ChatGPT several interviews at a time (P1-P3 transcripts first, and then the rest). For coding the transcripts, I used the following prompt, which I developed based on materials I found online (<a href="https://www.youtube.com/watch?v=1TiuUQf6dBo">here</a> and <a href="https://www.youtube.com/watch?v=8dTs7D42ge0">here</a>). The prompts could always be improved, but they did the job well enough. One of my friends who works with LLMs told me that I could ask ChatGPT for feedback on the prompts and improve them based on what it says. I’ll have to try that next time.</p><p><strong>ChatGPT prompt #1. Coding qualitative data:</strong><em> I want you to act as a user researcher specializing in qualitative analysis. Review the transcript to identify relevant excerpts that address the research question “What causes traffic congestion in Thung Khru, Bangkok?” Generate phrases, (i.e., codes) that best represent the excerpts identified. Each code should be between two to five words. Can you put this output in a table form? [copy paste anonymized transcripts into ChatGPT]</em></p><p>ChatGPT generated the codes with corresponding quotes for a chunk of the transcripts and listed them in a table. In addition to returning the table, it also offered a summary with something it called “insights” where it attempted to list what it found from the transcripts. These “insights” were surface-level observations at best and did not meet my criteria of what an insight is, which I will explain later on..</p><p>Then I followed up with another prompt, because I forgot to tell ChatGPT to list which interview participant the quotes belonged to:</p><p><strong>ChatGPT prompt #2. Coding qualitative data: </strong><em>Can you add a third column to the table that you generated and in it put the information on which person the excerpts come from?</em></p><p>I used participant numbers instead of actual names to anonymize the data. ChatGPT accurately assigned each code to a participant. There were no mistakes.</p><h4><strong>Step 2. Character mapping analysis</strong></h4><p>Traffic congestion is an issue which involved multiple actors, so I decided to do a <a href="https://methods.sagepub.com/book/mono/narrative-inquiry/chpt/6-character-mapping-time-analysis">character mapping analysis</a> hoping that it could lead me to more findings. In character mapping you identify all the actors mentioned in the interviews, their actions and psychological state (cognition, emotion and reported speech). Sometimes the actors can be objects or places. My character mapping analysis involved reading the transcripts for the second time and making a list of all the characters mentioned in the interviews.</p><p>Next, I prompted ChatGPT to replicate my character mapping analysis.</p><p><strong>ChatGPT prompt #3. Character mapping analysis:</strong> <em>Again, I want you to act as a user researcher specializing in qualitative analysis and use the methodology called character mapping. Review the same transcript and make a list of all the key characters such as people and objects that play a role in “What causes traffic congestion in Thung Khru, Bangkok?” Generate characters and for each one of them find a quote that supports the evidence that this character was mentioned in the transcript. Can you put this output in a table form?</em></p><p>ChatGPT found and listed more characters than I did after prompting it twice. There was a lot of consistency with the characters I listed and the ones ChatGPT generated. I double checked the quotes corresponding to each character and they were all from the transcript.</p><p>Next, I used another prompt to analyze the remaining interviews.</p><p><strong>ChatGPT prompt #4. Coding qualitative data for the remaining transcripts: </strong><em>Now, I will give you another transcript, which is from the same interviews. Again, I want you to act as a user researcher specializing in qualitative analysis. Review the transcript to identify relevant excerpts that address the research question “What causes traffic congestion in Thung Khru, Bangkok?” Generate phrases, (i.e., codes) that best represent the excerpts identified. Each code should be between two to five words. You can use the same codes that you used for the first transcript if that makes sense. The text is an interview and I do not want you to code the questions asked by the interviewer. Can you put this output in a table form? Here’s the transcript:</em></p><p>This was the only time where ChatGPT made something up. I said in the prompt: <em>You can use the same codes that you used for the first transcript if that makes sense </em>and it said it did. However, when I looked at the output, all of the codes it generated in this step were new. It didn’t reuse any of the previous codes, but it said it did. This was a powerful reminder for me to remain skeptical and double check everything ChatGPT outputs.</p><p>Finally, I reused Prompt #3 to continue with character mapping for the remaining transcripts.</p><h4><strong>Step 3. Developing high level themes</strong></h4><p>In this step, I organized the codes into clusters, grouping together the codes with similar meaning. Each code cluster created a high level theme. I ended up identifying several themes, which described the key causes of traffic in Thung Khru, Bangkok? Once I came up with the themes, I dove back into raw data. I revisited the codes, pulled quotes from the participants to enrich the understanding of the themes and searched for examples to make the themes that were too general more specific and descriptive. I also reviewed the list of actors from the character mapping analysis to see if I could get any additional insights.</p><p>Next, I asked ChatGPT to extract high level themes from the codes it generated:</p><p><strong>ChatGPT prompt #5. Developing high level themes.</strong> <em>Can you group these statements into distinct themes and output a table showing which statement belongs to which theme. The themes should be 2–5 sentences long and focus on answering the question “What causes traffic congestion in Thung Khru, Bangkok?” Here’s a list of statements: [input all the codes that ChatGPT generated in the previous steps].</em></p><p>I stopped here and compared my themes with the ones generated by ChatGPT. There were a lot of similarities. We independently agreed on major themes. In addition, I identified themes which ChatGPT didn’t and ChatGPT found some themes that I overlooked, a win-win scenario. There was one LLM identified theme, which made me realize that I came up with a learning that was biased, so I was able to correct my bias because I was working with ChatGPT.</p><h4><strong>Step 4. Generating insights</strong></h4><p>Insights are more than themes, observations or learnings, so what are they? I like a couple of definitions:</p><blockquote>Insights are “ carefully constructed interpretative statements grounded in the truth of someone else’s life that make an argument. They are (…) designed to influence the actions of someone else.” They come in the form of “descriptive, explanatory and predictive” storytelling and their purpose is exactly that “describe, explain and predict”. ~ Paul Hartley</blockquote><blockquote><a href="https://www.oreilly.com/library/view/101-design-methods/9781118083468/">Vijay Kumar’s definition</a> expands on it. “An insight is the act of “seeing into” a situation or understanding the inner nature of what we observe. It is our learning from an observation through our interpretation by asking the question why.”</blockquote><p>Insights are created after a researcher has their learnings and observations. They are meant to evoke an Aha-moment. They foster breakthroughs in understanding and ideally move people towards action or inspire a solution.</p><p>I noted a couple of insights which jumped out at me while I was synthesizing the codes into themes. One insight was about the systemic nature of traffic congestion with multiple factors acting together causing it. The implication is that there isn’t any silver bullet to solve traffic congestion, multiple factors have to be addressed simultaneously. Coming up with insights is equal parts art and science. Normally, I’d spend much more time on generating insights, but I wanted to focus on the analysis and synthesis. Generating insights and testing how well LLMs can do it deserves a separate blog post.</p><h3>Conclusion and limitations</h3><p>I was initially skeptical about experimenting with ChatGPT to analyze user research and I still have major questions. For example, is LLM- assisted analysis ethical and secure? What kinds of biases does it have? One way or another, this experiment encouraged me to keep testing. I learned that working side-by-side with LLMs is like working side-by-side with a human research assistant. They have their strengths in reliably coding and synthesizing (see Table 1), and can process huge amounts of qualitative data very fast. My strength is in generating non-obvious insights and we can complement each other when working on a simple research question. There are still some important limitations to using LLMs:</p><ul><li>It’s opaque how the data that I fed to an LLM is used by the model. I used anonymized, publicly available transcripts for this exercise. However, for conducting actual user research at work, I need to be super clear about what happens to the data, because protecting participant privacy and proprietary information is very important. So far I haven’t found any resources online that can confidently explain what happens to the data once it’s fed into the model.</li><li>There is always a possibility that the model will hallucinate, so I wouldn’t be comfortable to deploy ChatGPT to do the analysis on its own without my oversight.</li><li>ChatGPT was accurate for answering simple research questions, but I’m not sure how it would handle a larger data set and a more complex research topic.</li><li>LLMs can only assist with text data input and this is limiting. Typically, user researchers collect a mix of data to understand context and non-verbal reactions. They also triangulate quantitative data with qualitative findings, add knowledge from previous research, conduct thought experiments to reflect on opportunities, etc. What about data in other formats like video, imagery, numbers? Human researchers still have an advantage in analyzing other data types and connecting the dots between various data types.</li></ul><p>Despite these limitations, I encourage you to try using an LLM to assist you with qualitative analysis. Try it yourself and share your reflections in the comments.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=efdc0ebb78ed" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Exploring Local Music’s Place in Global Streaming]]></title>
            <link>https://medium.com/@domi_mazur/re-share-from-linkedin-exploring-local-musics-place-in-global-streaming-a23334e933a3?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/a23334e933a3</guid>
            <category><![CDATA[streaming-music]]></category>
            <category><![CDATA[spotify]]></category>
            <category><![CDATA[local-music]]></category>
            <category><![CDATA[recommender-systems]]></category>
            <category><![CDATA[user-experience]]></category>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Mon, 03 Jun 2024 23:00:27 GMT</pubDate>
            <atom:updated>2024-09-15T22:41:27.877Z</atom:updated>
            <content:encoded><![CDATA[<p>Creating personalized recommendation algorithms that are grounded in human needs requires extensive research and a collaboration from a cross-functional team. Here’s an example of how a handful of researchers and a machine learning engineer at Spotify worked together to develop algorithmic recommendations for local music which celebrated local musicians and meaningfully engaged Spotify listeners. This research is one of the most fun and rewarding achievements during my time at Spotify.</p><p>What is the secret sauce for building personalized recommendations which connect local musicians and listeners and drive engagement on Spotify? Here are a few ingredients which made a difference:</p><p>1) A clear and compelling research goal rooted in real user needs. <br>2) A multidisciplinary team including user researchers, research scientists, data — scientists and machine learning engineers. It would also be great to add design, which we did not have the luxury to have on this project. <br>3) Engaging with listeners and musicians in multiple places through cultural probes and in-depth user interviews to embrace their perspective and what makes music local. <br>4) Rigorous translation of qualitative insights into algorithmic recommendations. <br>5) Testing the recommendations in a randomized, large scale experiment to evaluate their impact.</p><p><a href="https://www.linkedin.com/in/samfway/">Sam Way</a> is presenting the local recommendations paper at Websci24 this week. This research was conducted by a stellar and passionate team including <a href="https://www.linkedin.com/in/samfway/">Sam Way</a>, <a href="https://www.linkedin.com/in/mazurdominika/">Dominika Mazur</a>, <a href="https://www.linkedin.com/in/rkupferman1/">Rebecca Kupferman</a>, <a href="https://www.linkedin.com/in/megan-walsh-data/">Megan Walsh</a>, <a href="https://www.linkedin.com/in/angelinatize/">Angelina Tizé</a> and <a href="https://www.linkedin.com/in/ben-lacker-b4992567/">Ben Lacker</a>. The authors would like to thank <a href="https://www.linkedin.com/in/lizzy-moroney-4a631a21/">Lizzy Moroney</a>, <a href="https://www.linkedin.com/in/efunwunmi-toni-coker/">Efun Toni Coker</a>, <a href="https://www.linkedin.com/in/emmafiguero/">Emma Figuero</a>, <a href="https://www.linkedin.com/in/hollie-carpenter-b59559a2/">Hollie Carpenter</a>, <a href="https://www.linkedin.com/in/kim-scofield-insights/">Kim Scofield</a>, and the team at <a href="https://www.linkedin.com/company/firefish-ltd/">FIREFISH</a> for their collaboration on this research.</p><p>Check out the blog post describing this research: <a href="https://lnkd.in/gDiycjNw">https://lnkd.in/gDiycjNw</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a23334e933a3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Numbers or wisdom? Exposing the trust in numbers cognitive bias]]></title>
            <link>https://medium.com/@domi_mazur/numbers-or-wisdom-exposing-the-trust-in-numbers-cognitive-bias-17cfd0481c07?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/17cfd0481c07</guid>
            <category><![CDATA[cognitive-bias]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[quantitative-research]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[spotify]]></category>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Mon, 03 Jun 2024 22:50:38 GMT</pubDate>
            <atom:updated>2024-06-03T22:50:38.498Z</atom:updated>
            <content:encoded><![CDATA[<p><em>In this blog post I’m introducing the cognitive trap which I named the ‘trust in numbers’ bias. It’s the tendency to lean heavily on quantitative data while ignoring other valuable sources of information when making decisions. I speculate about the neurological underpinnings of this bias, give some examples of how this bias shows up, including some examples from my time working as a user research manager at Spotify. Finally, I discuss how this bias can be mitigated by collecting diagnostic data from diverse sources, avoiding amassing non-diagnostic data and reflecting on how much you weigh numerical data in your decision making. It’s a longer read so set aside some time, sit back and dig in.</em></p><p>There is a scene from the movie The Big Short¹ that I really like. It features Mark Baum, a character played by Steve Carell. It’s the mid-2000s and Baum wants to understand the US housing market to make important investment decisions for his hedge fund. Baum ends up not finding clarity while sitting in his high-rise NYC office and analyzing the numbers that seem reassuring. He thinks that there’s more to the story than the numbers are revealing and he decides to do something different. He packs his suitcase and goes to Florida to take a look at the housing market on the ground. There, he visits housing developments, talks to real estate agents, mortgage brokers and homeowners. After having these conversations Baum realizes that the housing bubble is about to burst and he rushes back to New York to make bets against the market. This movie scene² attracts millions of views on Youtube, because people recognize Baum’s brilliance. If he had stuck to looking at what quantitative data was telling him, he would have missed the signs of trouble. By going to Florida and hearing stories directly from the people involved, Baum found out about the housing bubble early on and made important decisions which earned him a lot of money. The moral of the story is that Baum avoided what I call the “trust in numbers bias”. This bias is about relying too heavily on quantitative data and omitting other types of information when making important decisions. Baum avoided it because he recognized early on how important it was to do research on the ground and listen to the real stories from people who were involved.</p><blockquote>Trust in numbers bias is about relying too heavily on quantitative data and omitting other types of information when making important decisions.</blockquote><p>I’m adding the “trust in numbers bias” to a list of many cognitive biases, which are errors in judgment that often lead to poor quality decisions. These biases exist because the world is too complicated for us to be able to think methodically about every decision that we make. Instead, we try to preserve our cognitive resources and take all kinds of shortcuts in our thinking. We may oversimplify complex problems or ignore some vital information to ease the cognitive load and reach a conclusion faster. We may also over rely on certain types of data, because they are an “industry standard”, readily available and “worked well for us in the past ‘’ and we hope that they guide us to make the right decisions this time around.</p><p>I want to expose the “trust in numbers” bias because it leads to pitfalls such as:</p><ul><li>omitting diverse, diagnostic information in decision making</li><li>inability to pick up on early trends which do not yet show up in numbers</li><li>inability to have access to information that is not quantified, which can generate tons of rich insights</li><li>problem oversimplification, reductionism and commodification</li><li>relying on quick, simplistic conclusions.</li></ul><p>Let’s unpack some of these pitfalls with a quick example. Imagine that you want to study a magnificent 100-petal lotus.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*BKhTrkiP7BiIyOco" /></figure><p>Instead of looking at its intricate petal arrangement, touching it and understanding its ecosystem, you fixate on the number of petals. Suddenly the number 100 becomes the key to understanding this impressive flower. There is also an urge to put some value on it. Is a 100 petals a good outcome or not so impressive?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/479/1*L2hdC7e3tAS8iri-9EAwmw.png" /></figure><p>This kind of reasoning sounds ridiculous, right? And yet, it is fairly standard in modern research, in economics, in government and in tech (see for example Andrew Chen’s blog post discussing it³).</p><p>The “trust in numbers’’ bias isn’t just an innocent shortcut. It’s a symptom of a significant neurological shift happening in Western thinking towards more linear, abstracted, and disembodied information processing as described by Ian McGilchrist⁴. McGilchrist, who is a psychiatrist, links this phenomenon to the imbalance in how humans in Western society use their brains and over-rely on the left hemisphere:</p><blockquote>Philosophy in the West is essentially a left-hemisphere process. It is verbal and analytic, requiring abstracted, decontextualised, disembodied thinking, dealing in categories, concerning itself with the nature of the general rather than the particular, and adopting a sequential, linear approach to truth, building the edifice of knowledge from the parts, brick by brick. ~ McGilchrist</blockquote><p>He goes even further, citing Heidegger and calling this trend “forgetting of Being”:</p><blockquote>Heidegger saw the disasters of Western materialism as stemming from a ‘forgetting of Being’ ~ McGilchrist</blockquote><p>According to Ian McGilchrist’s years of research, the left hemisphere has a predisposition to abstract information and rely on language. Its mode of operation is fixed and concrete. It tends to skip context, it generalizes and does not use embodied data. The fixedness, concreteness and abstraction, which are associated with how the left hemisphere operates, are also the mode of operation used to quantify the world in economics, statistics, psychometrics and machine learning. On the other hand, the right hemisphere’s take on reality is interconnected, embodied and holistic. It’s sensitive to changes, highly contextual and steeped in lived experience. This way of thinking is associated with disciplines like anthropology, art and movement practices.</p><p>Both of our hemispheres think in numbers, but each hemisphere approaches them differently. McGilchrist argues that <em>numbers can either signify absolutes — a quantifiable amount, as in statistics — which would suggest an affinity with the left hemisphere, or signify relations, which would suggest an affinity with the right hemisphere.</em> He further explains that signifying relations is about <em>“regularity of proportion or relationship, rather than number in any absolute sense, that underpinned music and beauty — the music of the spheres, the natural harmony of the universe. </em>The way I understand it is as follows — if we go back to the example of studying the 100 petal lotus, left hemisphere would focus on the quantifiable amount 100, and the right hemisphere would focus on the relations between the petals, the proportionality, the harmony of the flower, which can be in fact numerical, but is not expressed in absolute numbers.</p><p>McGilchrist argues that the problem with current Western thinking is that the left hemisphere tends to dominate and heavily affect how we operate in the world.</p><blockquote>Because the left hemisphere likes to think about numbers as fixed and unchanging and it currently is dominant, it will seek out and give preference to numbers in absolutes like in statistics. It will also try to simplify and quantify everything even when it does not make sense to do that, a phenomenon cleverly named quantifauxation by a statistician Philip B. Stark⁵.</blockquote><p>This hasn’t been scientifically tested, but I hypothesize that “trust in numbers’’ bias comes from the left hemisphere dominance and the preferences it has to see the numbers as absolutes. This tendency gets stronger as quantitative data becomes more readily available for us to consume.</p><p>If you are a human, you are prone to the “trust in numbers bias”. I witnessed this bias often and fell into that cognitive trap myself a few times. For instance, during my time at Spotify, my team was working on increasing the awareness of our various subscription plans. We had previous research showing that listeners in small households were not aware of and did not understand the subscription plan that Spotify created specifically for their needs. This insight surfaced from user interviews and later some quantitative research was conducted to roughly understand the size of the problem (think surveys or looking at analytics for page views). What’s important about this example is that there was a blind spot and it was uncovered in conversations with our listeners. The problem did not surface first in analytics or quantitative data. Could it? Theoretically yes, if someone had a hunch to look into the data or if there was a direct ask, but, in my 15 years experience as a user researcher, it often happens that important questions first arise from lived experience in the real world, from conversations and stories and from being captivated by a real world phenomenon that you can’t explain. There is a good chance that this issue wouldn’t have been uncovered for a while if it wasn’t for user research, which would lead to some missed opportunities for Spotify. From this initial set of insights which were a mix of user stories, analytics and surveys, the product and design team generated hypotheses for how to address the lack of awareness and understanding of the small household plan. We used what’s called a “thoughtful execution” framework⁶. This was stellar product design, but when it came to testing these ideas, the team got fixated on running an A/B test to compare the design ideas against a control condition. The team came up with lots of reasons why an A/B test would be superior compared to other methodologies. It would be faster, it would be more generalizable, the recommendations would be simple and concrete and we could make a decision and move on. Other methods of testing weren’t really considered. However, to one user researcher on my team, this focus on running an A/B test suggested that the team may have a quantitative data bias. There was an underlying belief that quantitative data will answer all the questions in a superior, highly reliable way and clear any doubts. There was also pressure on the designers and product team to validate their ideas with quantitative data, because there was an even wider spread quantitative bias in the leadership team to whom they reported. However, there was one relentless user researcher on my team who intuitively understood that the people involved in this project had this bias and she pushed and pushed that team to run some qualitative interviews with the listeners in the A/B test. She pushed despite visible discouragement from the team and the leadership, who kept telling her — why run user research, we’ll know all the answers through an A/B test. She did it anyway and guess what happened? The user research completed way before the A/B test was over, so it was faster despite the common belief that you can generate quantitative data faster. More importantly, interviewing users, entering their world and seeing the reality from their perspective generated insights which wouldn’t show up in the A/B test, such as the fact that some designs were perceived as ads and quickly dismissed. There were also some designs that were subject to greater message blindness than others, which made the understanding of the small household offer harder. In addition, the placement of where these offers appeared had a significant influence on how likely they would be dismissed. The focus on only gathering quantitative data stopped the team from fully knowing and experiencing the topic of our study, but they got back on track. In the end, the they were grateful for running both the user research and an A/B test. They put the story and the recommendations together based on data sources coming from different methodologies. This mixed method approach led to enriched conversations about what to decide next. It took one stubborn user researcher to challenge the quantitative bias by bringing an additional source of data and evolving the story.</p><p>Overreliance on one type of data can be especially harmful when the decisions that you need to make are of high importance. Imagine that you have to decide on a strategy direction that your company should pursue next year. If you take a shortcut and rely too much on quantitative data and ignore other sources of information such as early trends that are not usually captured by numbers, it can lead to poor quality calls.</p><p>Now, I’m not saying that we shouldn’t use quantitative data and I agree that running just an A/B test is sometimes enough. I’m asking that we enrich our data sources and most importantly gather diagnostic data. I often hear the argument that it never hurts to have lots of quantitative data. Isn’t it always better to have more information especially when making a difficult decision? I’d say it’s better to have diagnostic information rather than a lot of information. When you focus on amassing data, you’ll end up with some of it being diagnostic and a lot of it being noise. Consequently, irrelevant data mixed in with diagnostic data may lead to what is called the dilution effect (see for example Zukier, 1982⁷) where consuming non-diagnostic information next to relevant data clouds judgment and may lead to wrong predictions. If you had a math teacher in grade school who gave you word math problems with irrelevant information to intentionally lead you astray, you know exactly what I’m talking about. These frustrating math problems were designed to teach kids how to slow down and parse useful information from noise rather than take mental shortcuts to quickly produce an inaccurate answer.</p><p>Next time you have to make an important decision, I challenge you to check if you collected truly diagnostic data from diverse sources and to reflect on how much weight you put on quantitative data. Here are some examples of diverse data that can be diagnostic and help with decision making:</p><ul><li>Quantitative</li><li>Qualitative</li><li>Embodied — how things feel, gut feeling</li><li>Real life stories and biographies</li><li>Metaphors, symbols, legends</li><li>Historical data</li><li>Data from thought experiments and visualizations</li><li>Data from observation</li><li>Imaging</li></ul><p>I hope that looking at the above list will show how limiting it can be to put too much weight on quantitative data. It stops you from more fully knowing and experiencing the matter in question, so break free from the quantitative world and see what else is out there. By considering multiple perspectives and sources of information, we can paint a fuller picture and make better-informed decisions. After all, numbers only come to life when they are connected to the lived experience.</p><p>What do you think? Please comment below! Thank you.</p><p>[1]: The Big Short (2015) <a href="https://www.imdb.com/title/tt1596363/">https://www.imdb.com/title/tt1596363/</a></p><p>[2]: The Big Short (2015) <em>FrontPoint Partners’ investigation in Florida &amp; first trade. </em><a href="https://youtu.be/MesrrYyuoa4?si=Z2OFpFiJX598IkWL">https://youtu.be/MesrrYyuoa4?si=Z2OFpFiJX598IkWL</a></p><p>[3]: Andrew Chen. (May 28, 2024) <em>Why data-driven product decisions are hard (sometimes impossible) </em><a href="https://open.substack.com/pub/andrewchen/p/why-its-so-hard-to-be-data-driven?utm_campaign=post&amp;utm_medium=web">https://open.substack.com/pub/andrewchen/p/why-its-so-hard-to-be-data-driven?utm_campaign=post&amp;utm_medium=web</a></p><p>[4]: Iain McGilchrist. (March 26, 2019) <em>The Master and His Emissary: The Divided Brain and the Making of the Western World </em><a href="https://yalebooks.yale.edu/book/9780300245929/the-master-and-his-emissary/">https://yalebooks.yale.edu/book/9780300245929/the-master-and-his-emissary/</a></p><p>[5]: Philip Bradford Stark. <em>Miscellaneous stuff about P.B. Stark </em><a href="https://www.stat.berkeley.edu/~stark/other.htm">https://www.stat.berkeley.edu/~stark/other.htm</a></p><p>[6]: Annina Koskinen. (June 2020) <em>From Gut to Plan: The Thoughtful Execution Framework </em><a href="https://spotify.design/article/from-gut-to-plan-the-thoughtful-execution-framework">https://spotify.design/article/from-gut-to-plan-the-thoughtful-execution-framework</a></p><p>[7] Henry Zukier. (1982) <em>The Dilution Effect: The Role of the Correlation and the Dispersion of Predictor Variables in the Use of Nondiagnostic Information </em><a href="https://drive.google.com/file/d/1NjBmhJwDNrtufT0IzaehB400WtDgY0NK/view?usp=sharing">https://drive.google.com/file/d/1NjBmhJwDNrtufT0IzaehB400WtDgY0NK/view?usp=sharing</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=17cfd0481c07" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Think Global, Act Local. Designing for different markets at Spotify.]]></title>
            <link>https://medium.com/@domi_mazur/think-global-act-local-designing-for-different-markets-at-spotify-d4b43f7854b8?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/d4b43f7854b8</guid>
            <category><![CDATA[design]]></category>
            <category><![CDATA[spotify]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[growth-strategy]]></category>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Sun, 01 Mar 2020 14:36:06 GMT</pubDate>
            <atom:updated>2020-03-01T14:36:06.818Z</atom:updated>
            <content:encoded><![CDATA[<p>This piece was originally published as a Case Study <a href="https://spotify.design/articles/2019-07-09/think-global-act-local/">here</a>. My co-author is <a href="https://www.linkedin.com/in/oddlyssarides/">Odd Lyssarides</a>.</p><p>As Spotify expands to more markets, there’s a shift that can be seen at every level of the company. Previously, the prevailing mindset was European and US-centric. Most A/B tests were run in the English language and most personas drawn from the US market. But this has changed significantly. And as part of the dedicated Global Insights Team, we’re on a mission to make Spotify feel local and relevant in every corner of the world.</p><p><strong>We start with similarities</strong></p><p>Our team brings together user researchers, market researchers, data scientists and experts in other disciplines to identify growth opportunities, and build a better understanding of the new markets we’re entering. Our approach is to start by <a href="https://spotify.design/articles/2019-03-06/bringing-spotify-to-india-and-beyond/">looking for similarities</a> across these markets, as our colleague Heli Rantavuo mentioned. After all, tastes transcend geographical boundaries.</p><p>Yet there are many nuances in music behaviors that stem from cultural variation. Take K-pop fans, for instance. In South Korea and the US, they’re loving the same bands, tunes and artists, and listening in order to feel a sense of belonging and shared connection with their peers. In the US, however, there’s an additional reason for listening — the enjoyment of discovering and immersing themselves in foreign pop culture.</p><p>What’s more, local geography and infrastructure have significant impact on people’s ability to stream music and enjoy a service like Spotify. There’s huge variation in the cost of data packages, the speed and reliability of internet connections and the quality of phones. So we work hard to overcome cross-regional barriers to streaming and help everyone access their music, wherever and whenever they want to.</p><p><strong>Different methods, same goal</strong></p><p>Since the Global Insights team is cross-disciplinary and method-agnostic, our researchers and data scientists work hand-in-hand to define the burning research questions. Then together, we figure out the best way to answer these questions — whether that’s with an A/B test, a daily diary study or a cultural probe, in which listeners submit artefacts that represent the meaning of music in their lives. We combine knowledge from disparate sources and come up with stronger conclusions and better product solutions as a result.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/853/1*b3h5f-hoIdiU5evo4uae8Q.jpeg" /><figcaption><em>Cultural probes with K-pop listeners.</em></figcaption></figure><p>To see how this works in practice, let’s start with some questions.</p><p><strong>Interrogating music culture worldwide</strong></p><p>How do people discover new music in different parts of the world? What purpose does music fulfill in their daily lives? And how does music discovery differ locally and globally?</p><p>When we approach these kinds of questions, we may focus on a specific region — a place where our market fit needs addressing. Alternatively, we may focus on an opportunity that will make an impact in many regions, so that our improvements can easily be rolled out to other parts of the world.</p><p>We often start by analyzing internal data to understand how our users currently discover music — perhaps through features like Charts, Discover Weekly or Release Radar. We consider how such features perform in different regions and which currently drive retention and conversion.</p><p>When looking at our internal data, there are a few guiding principles to keep in mind. Firstly, there is no such thing as an average user, or an average A/B test result. An experiment which shows positive results on a global level might need to be further segmented in order to apply to individual countries, cultures or communities. And the findings of more locally-based experiments may not scale up successfully to the rest of the world.</p><p>Secondly, we always maintain a healthy skepticism of what internal data is telling us. For example, the data gathered since our recent launches in India and the Middle East may not be robust and representative, because we’re only just beginning to grow in those markets.</p><p>And finally, we steer away from simply comparing one country to another. It’s always more revealing to take a broader view across multiple regions and consider like-minded groups of listeners, as well as geographically-specific markets.</p><p><strong>Tapping into local knowledge</strong></p><p>As well as looking at internal data, we collaborate with local Spotify teams to get the all-important context for our findings — for instance, what recent marketing campaigns have been run in the region. We often then conduct user research studies to go deeper into issues that can’t be understood through analysis of the internal user data alone.</p><p>Depending on what we want to discover, we’ll use different research methods — conducting surveys and diary studies, or even visiting listeners for conversations in their own homes. We’ll involve our designers, product managers, data scientists and engineers, so the whole team is fully invested in the project. And we’ll start to identify patterns and similarities with other regions, in case a product solution could have wider relevance and roll-out.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/820/1*nssQTcI4hIKWb06o5DsvKw.jpeg" /><figcaption><em>Visits with Spotify users in their homes in Jakarta.</em></figcaption></figure><p><strong>Delving beyond data</strong></p><p>One area we’ve been investigating recently is how to minimize the barriers to streaming resulting from high data cost, poor quality cellular networks or low-end devices. And once again, our research methods are diverse and multi-disciplinary. To understand and segment the audience affected by these constraints, we first look at the types and ages of phones owned by our listeners, the cost of cellular data in relation to their income and the quality of network connection available to them. We then consider demographics, geography and behavioral segments as secondary dimensions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/779/1*SEK981eRoeMaDFM_tnKoJg.jpeg" /><figcaption><em>We were surprised and delighted by the colorful phone booths on the Brazilian streets</em></figcaption></figure><p>Analysis of our internal data shows that retention and engagement with Spotify is lower for listeners on older and less advanced devices. We believe that the technical performance of the app plays a part here. Yet there’s also a rich cognitive and emotional side to the issue of access constraints and the story can’t be fully told with metrics only.</p><p>For instance, many listeners in our user research sessions tell us they’re frustrated with the limitations of Lite apps — smaller, faster versions of regular apps — even though they understand these limitations on a rational level. We also observe that curating your music collection on low-end phones with poor data plans is a fragmentary, manual process. Yet our music fans are willing to use complicated workarounds, because simple methods like streaming aren’t always reliable.</p><p>It’s thanks to our team’s diverse skills that we’re able to capture such subjective stories and consider them alongside metrics to get a full picture of what our users experience. And it’s thanks to our close collaboration with regional offices and local listeners that we’re able to set out findings in context — in a global marketplace, with an ever-changing culture and personality.</p><p><strong>Read more from our Global Team in two companion pieces, </strong><a href="https://spotify.design/articles/2019-03-06/bringing-spotify-to-india-and-beyond/"><strong>“Bringing Spotify to India — and Beyond”</strong></a><strong> and “</strong><a href="https://spotify.design/articles/2019-04-12/designing-on-the-road-finding-inspiration-in-new-markets/"><strong>Designing on the Road</strong></a><strong>”</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d4b43f7854b8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Chinese calligraphy can teach us about how to be mindful and deliberate]]></title>
            <link>https://medium.com/@domi_mazur/what-chinese-calligraphy-can-teach-us-about-how-to-be-mindful-and-deliberate-e22f347cd8a2?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/e22f347cd8a2</guid>
            <category><![CDATA[mindfulness]]></category>
            <category><![CDATA[life-lessons]]></category>
            <category><![CDATA[lifestyle-design]]></category>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Wed, 27 Jul 2016 21:36:15 GMT</pubDate>
            <atom:updated>2016-07-27T21:36:15.196Z</atom:updated>
            <content:encoded><![CDATA[<p>I’m trying to concentrate more and multitask less, because when I focus, I’m more productive and the quality of my work is better.</p><p>When searching for new ways to become more mindful, I stumbled upon this text by Chiang Yee, a Chinese artist, poet and calligrapher. It’s about the role your body posture plays in being deliberate and focused:</p><blockquote>“When an archer has thoroughly sighted his target, poised his body, grasped his bow firmly, and aimed accurately, the arrow will almost certainly hit the mark. So with the calligrapher: with the mind concentrated, the body upright and balanced, the brush vertical, the dot or stroke should fall exactly on the appointed place” (Chiang Yee, “Chinese Calligraphy” Harvard University Press.)</blockquote><p>This quote made me understand that concentration on every task has to be both mental and physical. This is true even for activities that seem not to engage your body as much, like thinking or typing on your computer. From now on, when I do research, sketch, write, drive, cook and garden, I will think about my body posture and make sure it’s right for the task. This will help me stay concentrated and be more productive.</p><p>Here are some other tricks that help me stay more focused and mindful:</p><p>1) I put my phone on “do not disturb” most of the day. My phone should not be in charge of me, I should be in charge of my phone.</p><p>2) Each time I get an urge to get distracted from a task at hand and do something else, I don’t act on it. Instead, I visualize that this distracting thought enters my head through my forehead and then exits through the back of my head and disappears. This trick works 70% of the time, which is pretty damn good.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e22f347cd8a2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Keep UX Real: How to empower and inspire with UX Research]]></title>
            <link>https://medium.com/@domi_mazur/keep-ux-real-how-to-empower-and-inspire-with-ux-research-823fe9ff9c25?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/823fe9ff9c25</guid>
            <category><![CDATA[design]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[ux]]></category>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Sun, 08 May 2016 00:25:12 GMT</pubDate>
            <atom:updated>2017-04-26T20:26:09.234Z</atom:updated>
            <content:encoded><![CDATA[<h4>Recap of my talk from UX Poland 2016</h4><p>I was asked to share my presentation from UX Poland 2016 and decided to write a blog post instead of sharing the slides. Here’s the recap.</p><p>When I started my UX Research career I felt the happiest at the end of each project when I delivered the research findings. The project was done and it was time to celebrate. However, very quickly I realized that the most rewarding part of my job was not about delivering the results, but seeing UX research in action. I was the happiest when I saw research being used to innovate and make improvements. I also realized that, as a UX Researcher, I was responsible for empowering teams to act on research findings. I started experimenting with various strategies for how to motivate teams to act on UX Research. Here’s a list of things that I try to use at every step in UX Research project to make my work more actionable.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/831/1*p0GmcndOzw-LP74FlaN-2A.png" /></figure><h4><strong>1) Strategies at the beginning of a user research project, when you get a UX research request:</strong></h4><p><strong>HAVE A PLAN FOR HOW THE RESEARCH WILL BE IMPLEMENTED AND BY WHEN.</strong> Together with the team that commissioned the research, write down a concrete plan for how the research will be implemented. For example, your team will use research findings to create sketches or prototypes based on research or have a brainstorming session to come up with solutions. In addition, set a realistic timeline stating by when the results will be implemented.</p><p><strong>DEPRIORITIZE RESEARCHING THINGS THAT CAN’T BE CHANGED. </strong>Talk with<strong> </strong>your team about limitations and constraints. There may be areas in a project that you’d love to explore, but spending time researching things that your team can’t change is of little use.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1002/1*mRGHOQeWcsZjJTAvIw2c1Q.png" /><figcaption>For large teams, you can organize a story mapping workshop to encourage your team members to voice their opinion and make them feel like they are being heard. Stakeholders get together and discuss what the user journey looks like. Outcome is a living document that teams can refer to during product design, development and lifecycle.</figcaption></figure><p><strong>LISTEN TO YOUR TEAM, NOT JUST USERS.</strong> UX Researchers focus on feedback from users, but teams are more eager to accept and act on UX Research when they feel like their expertise is also taken into consideration. If you have a project with a lot of stakeholders involved, try doing a story mapping workshop to give your team a chance to voice their opinion about the user journey. Check out these resources about story mapping:</p><ol><li>User Story Mapping by Jeff Patton (2014)</li><li>UX Podcast episode #102 Story mapping &amp; analytics with Adrian Howard &amp; Mike Beasley (2015)]</li></ol><h4>2) Strategies for when you’re in the field conducting research:</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/651/1*MjHFypMiZtKwL6DlC_X_7g.png" /><figcaption>Create a research grid that team members can fill out as they are observing the sessions. This grid is a summary of important observations about each user.</figcaption></figure><p><strong>PUT YOUR TEAM TO WORK</strong>. Ask team members to observe, take notes, write down insights and look for themes or even interview users when you conduct research sessions. They can take high level notes on a whiteboard grid or use post it notes and do affinity diagramming to see emerging themes. Active participation in UX research makes teams pay attention and remember findings. You increase the chance that they will act on this knowledge.</p><p><strong>SEND UPDATES FROM THE FIELD.</strong> At the end of each day, send out a note to the team telling them about insights from research sessions so that they can start working on strong, obvious findings even before the research is completed.</p><p>For example:</p><blockquote>“Today we observed 7 people attempting to sign in as new users. 6/7 took 3 minutes trying to create a password that would meet all of the security requirements. Many abandoned the site … Here’s the video”</blockquote><h4>3) Strategies for sharing results after user research is completed:</h4><p><strong>DELIVER RESULTS QUICKLY.</strong> After fieldwork is done, keep the momentum going and deliver results quickly, while teams are still excited about research. It usually takes weeks to analyze research results and prepare the report, so what should you do? I like to send out an email a couple of days after fieldwork is done summarizing strong trends and what I call “low hanging fruit” (things that are easy to implement). It’s easier and faster for teams to read an email rather than a long .pdf document or a PowerPoint. They can start immediately applying some insights while I’m working on the final report.</p><p><strong>SHARE RESULTS AND RECOMMENDATIONS IN SMALL CHUNKS</strong>. While you are working on the final report, you can send your team daily updates including one finding and recommendation. Make them visual, concise and easy to access. It’s easier to commit to reading a short paragraph or a visual than to read a 100-page report. Delivering information in bite-sized chunks is also better for short attention span. Look up Luke Wroblewski’s posts on Twitter @lukew and LinkedIn and you’ll understand what I’m talking about.</p><p><strong>EASY FIXES FIRST.</strong> Begin with sending recommendations that are easy to implement. Working on easy fixes brings immediate gratification and enthusiasm for tackling more complex challenges. However, you have to be careful and make sure your team doesn’t rest on laurels and end up implementing only the things that are easy to fix. Encourage them to tackle more difficult problems after they implement easy things.</p><p><strong>GIVE CREDIT FOR POSITIVES.</strong> UX Researchers are usually the ones bringing bad news. We tend to focus on negatives, because ultimately it’s our job to look for issues and challenges…BUT reminding your team about things they are doing right works like positive reinforcement. When you give credit for positives, your team will feel rewarded and motivated to get things right. On the flip side, when all you bring is bad news, people feel helpless and avoidant.</p><p><strong>CREATE EMPATHY WITH THE USER.</strong> Include user photos, share quotes, show user videos. Stories and visuals lead to inspiration and empathy. As empathy grows, teams become better advocates for users and more eager to apply UX research.</p><p><strong>UNDERSTAND HOW YOUR AUDIENCE LEARNS AND CATER TO THESE DIFFERENT LEARNING STYLES.</strong> People absorb information and get inspired in many different ways. Some people like details and data-heavy reports, others like visuals. Some people like to learn through play or by trying things out. Forget PowerPoint and try experimenting with different deliverable types. My UX research team so far has tried delivering results in a form of playing cards, workshops and spreadsheets to cater to different learning styles and time commitments.</p><p><strong>PILE UP EVIDENCE.</strong> Qualitative UX Research is often criticized for relying on very small samples sizes. If you create a UX Research data repository, you will be able to show that a certain opportunity or issue came up in several research projects, which increases your credibility. It’s hard to resist mounting evidence</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/810/1*e5TsS0N3jGnvB50yLrqr3A.png" /><figcaption>At my company, we display UX Research results on the wall in our kitchen area and the hallway so that everyone can see what we are working on.</figcaption></figure><p><strong>MAKE UX RESEARCH VISIBLE.</strong> Showcase research results and artifacts in visible places. This gives you an opportunity to start a conversation about your research with anyone who stumbles upon a visual of your recent project in a hallway or cafeteria. What a great opportunity to pitch research-based solutions and promote your work!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=823fe9ff9c25" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When gamification doesn’t work… and what to do about it?]]></title>
            <link>https://medium.com/@domi_mazur/when-gamification-doesn-t-work-and-what-to-do-about-it-eefbf2131d98?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/eefbf2131d98</guid>
            <category><![CDATA[gamification]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[psychology]]></category>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Sat, 19 Sep 2015 22:48:56 GMT</pubDate>
            <atom:updated>2017-02-20T02:59:34.698Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zaXaqtzs8hBc4ZcMHebpTw.jpeg" /></figure><p>Gamification is becoming widespread in UX Design and marketing because it’s believed to raise user engagement. However, there are so many badly designed gamification experiences out there (<a href="http://www.imediaconnection.com/content/32280.asp#multiview">see examples here</a>) that some people in design industry started to call this approach marketing <a href="http://www.theatlantic.com/technology/archive/2011/08/gamification-is-bullshit/243338/">BS</a>. What is the problem with gamification and why only a few UX Designers can implement it successfully?</p><p>Gamification fails when it’s used as a standard bag of tricks without taking time to fully understand why humans are drawn to game playing and rewards in the first place. Here are several misconceptions about how gamification works. Take a look at these and verify how you approach gamification in your UX work. Reflecting on these misconceptions can help you avoid common mistakes and use gamification to do what it’s supposed to do — increase user engagement:</p><h3>Six misconceptions about gamification:</h3><h3><strong>#1</strong> Gamification is the same as giving out loyalty points, gifts and incentives</h3><p>Believe it or not, but some designers and marketers still think that gamification is the same as giving users incentives or loyalty points in exchange for performing a desired action. Gamification can be engaging without providing tangible incentives when it is designed to reward users with other intangible rewards such as pure fun and pleasure or mastery and learning.</p><h3>#2 People engage in game-like experiences mainly to receive tangible rewards such as loyalty points</h3><p>Sure, people like tangible incentives (e.g. financial rewards), but they usually have other motivations to engage in gamified experiences that are stronger than wanting a tangible reward:</p><ul><li>for social reasons (e.g. to spend time with others, to gain respect, admiration among others, to compare themselves with others, to gain approval, recognition, to cooperate, to compete with others)</li><li>to learn</li><li>for fun and pleasure</li><li>to challenge themselves</li><li>to engage in an interesting activity</li><li>for satisfaction</li><li>for novelty</li><li>to gain mastery</li><li>to support a cause one cares about</li></ul><h3>#3 You need to include incentives (e.g. badges) to get users interested in your gamified experience</h3><p>You don’t always have to include incentives, but you do have to align your game strategy with users’ goals. You also have to provide a measure of progress and feedback to let users know how they’re doing throughout the experience. You can also replace incentives with feedback and encouragement (e.g. applause or animated transition indicating moving to the higher mastery level).</p><h3># 4 If you get users to engage with your product through gamification, they will remain engaged</h3><p>Let’s say you got initial engaged in your gamification experience by providing tangible rewards and incentives. This is not enough to maintain engagement. If the goals of the game do not align with users’ internal goals and needs, user engagement will stop as soon as the incentives stops.</p><h3># 5 Gamification should primarily support company’s goals, users’ goals are secondary</h3><p>Users are not going to fully engage in a gamified experience that does not fully support their goals. They may initially engage in your game only to take the tangible rewards you offered for participation and then they will leave.</p><h3># 6 It doesn’t hurt to include game-like elements in the experience. It’s not going to do any harm even when it does not work quite as you intended</h3><p>It’s dangerous to use gamification without carefully planning how you’re going to use it. Some badly designed gamification experiences can discourage and frustrate users. Here are a couple examples:</p><ul><li><strong>Experience is externally controlled.</strong> For example, you got dropped from level 3 to level 1 in a game-like experience and you do not know how and why it happened and do not know how to change it — users who feel that they do not have control over making progress, achieving goals or getting points in the experience will not participate in your game</li><li><strong>Experience has unattainable goals</strong> (e.g. extremely high leader board scores) will make users withdraw and become frustrated</li><li>Experience is too confusing — game rules and instructions are not clear, users who do not know what to do to make progress in the game will withdraw</li><li><strong>Game is too hard</strong> — leads to frustration</li><li><strong>Game is too easy</strong> — leads to boredom</li></ul><p>To sum up, think twice before you use gamification in your app or a website and do not just apply a standard bag of game tricks. Think about the people who you invited to play this game and what they want to accomplish when they engage in this game. Are you helping them fulfill their goals and desires or making them feel stupid or frustrated?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eefbf2131d98" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Life in California, USA]]></title>
            <link>https://medium.com/@domi_mazur/life-in-california-usa-c53418523430?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/c53418523430</guid>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Mon, 23 Mar 2015 02:24:38 GMT</pubDate>
            <atom:updated>2015-03-23T03:43:46.366Z</atom:updated>
            <content:encoded><![CDATA[<p>I moved to California about 11 months ago. The places were I lived before (Warsaw, Poland and Salt Lake City, Utah) had cold, snowy winters and distinct four seasons. California’s weather is different, with only subtle changes between seasons. This makes me feel confused about what time of year it is. It’s March now, but to me it feels like June…</p><p>I found a poem which helped me get in touch with my feelings about California’s strange weather. This poem was written by Czesław Miłosz, who moved from Poland to Berkley, CA in 1960. We share similar feelings of disbelief and confusion about California.</p><p>Here’s an excerpt from the poem written in Berkley in 1975 and translated from Polish to English by Miłosz:</p><p>Soon after our arrival, Budberg, gently pensive,</p><p>Said that in the beginning it is hard to get accustomed,</p><p>For here there is no spring or summer, no winter or fall.</p><p>“I kept dreaming of snow and birch forests.</p><p>Where so little changes you hardly notice how time goes by.</p><p>This is, you will see, a magic mountain.”</p><p>…</p><p>Sultry Octobers, cool Julys, trees blossom in February.</p><p>Here the nuptial flight of hummingbirds does not forecast spring.</p><p>Only the faithful maple sheds its leaves every year.</p><p>For no reason, its ancestors simply learned it that way…</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c53418523430" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mobile UX research]]></title>
            <link>https://medium.com/@domi_mazur/mobile-ux-research-comparison-of-different-data-collection-methods-and-tools-df3e8e8f7896?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/df3e8e8f7896</guid>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[mobile]]></category>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Tue, 27 Jan 2015 18:26:55 GMT</pubDate>
            <atom:updated>2015-10-16T19:03:35.043Z</atom:updated>
            <content:encoded><![CDATA[<h2><strong>Mobile UX research,Part 1</strong></h2><h3><strong>comparison of different data collection methods and tools</strong></h3><p>Over the last few months our team conducted several mobile UX research studies. We used different set ups and tools in each study to improve our data collection method. Each method and tool has advantages and disadvantages. Some set ups are easier to put together and use, some are not. Some may be better for projects in a more natural setting, outside of the lab, some work better in a more controlled lab setting. Here is my opinion about the mobile UX research rigs that I’ve used so far.</p><p><strong>Ipevo mobile sled and Ipevo software: </strong>First we tried the <a href="http://www.ipevo.com/">IPEVO sled and software</a>. This set up has a platform with a webcam attached to it on which the user places their mobile device and holds it during the usability session (see image 1). We also included an additional webcam mounted on a <a href="http://joby.com/gorillapod">GorillaPod</a> to record users’ faces. Video inputs from both webcams were streamed into our lab computer and captured using a screen capturing software (we currently use <a href="http://www.telestream.net/screenflow/overview.htm">ScreenFlow</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Xj4TpKA-6lzMwBv2hrHTAA.jpeg" /><figcaption>Image 1. Ipevo webcam for recording mobile UX research</figcaption></figure><p><strong>Advantages</strong></p><p>IPEVO sled and software is easy to use and intuitive and you can have it up and running in 5 min, which is nice when you have limited time to get ready before the session. Compared to other set ups, IPEVO is easier to use “on the go” or in a natural setting outside of the lab. For example, if you want to test how car passengers use their mobile devices while the car is being driven or when you’re meeting the user in a coffee shop.</p><p><strong>Disadvantages</strong></p><p>Holding the sled with the mobile device gets in a way of natural use. The webcam recording the mobile screen slightly blocks user’s mobile screen view and the view of what’s in front of them, which can make them feel uncomfortable. Especially at the beginning of the session, users feel like it’s about to poke them in the eye. This set up is fairly light, but for long sessions, holding the mobile device on a sled can wear users out. My advice is to encourage users to put the phone down and take a break whenever they are not performing the task. Lastly, the video output quality from IPEVO is far from great.</p><p><strong>Mr Tappy mobile sled and webcam software: </strong>We also ordered and played with <a href="http://www.mrtappy.com/">Mr Tappy kit</a>. With this set up, similar to IPEVO, mobile device is placed on Mr Tappy sled (see image 2) and the second camera mounted on a GorillaPod records user’s face. Mr Tappy does not come with a webcam, so you have to attach a webcam to it. Make sure you have a webcam that can be easily attached, because not all of them will work with this sled. You can capture feeds from both webcams using a screen capturing software such as <a href="http://www.telestream.net/screenflow/overview.htm">ScreenFlow</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZxBoAy_uAgwTtaMsenz2jg.jpeg" /><figcaption>Image 2. Mr Tappy rig for recording mobile UX research</figcaption></figure><p><strong>Advantages</strong></p><p>Mr. Tappy is light, sleek and elegant. You can choose which webcam you want to use with this set up, but you have to mount the webcam on Mr Tappy, so you need more time to prepare before the session. Similar to IPEVO, this set up is recommended for sessions outside of lab setting, “on the go”, when participant is moving, etc.</p><p><strong>Disadvantages</strong></p><p>Holding the sled gets in a way of natural use. The feeling is similar to holding the IPEVO rig. Mr Tappy is pretty light, but holding it during a long session can wear out the users.</p><p><strong>Two webcams on GorillaPods, webcam software and mobile screen mirroring software: </strong>One of my colleagues used this set up and I want to try it out when I run my next mobile UX study. You use two webcams mounted on GorillaPods with one webcam filming the mobile screen (users hold the phone in front of this camera just like they would normally do it and there are no mobile sleds involved). The other webcam is filming the user’s face. You indicate the area where users should hold their phone with a tape to prevent users from accidentally moving the phone out of the webcam view. Inputs from both cameras are fed into one computer through the webcam software. In addition to two webcams, you use a mobile screen mirroring software (e.g. AirServer) to get the enlarged view of the phone screen on your computer. You can record all these inputs using a screen recording software.</p><p><strong>Advantages</strong></p><p>It’s a more natural set up, because users hold just the phone, and not the phone and the sled. Adding mobile device mirroring feed via AirServer makes it easier for researchers, observers and notetakers to follow where the user is in the flow.</p><p><strong>Disadvantages</strong></p><p>This set up gets in the way of natural use because the webcam recording the mobile screen can obscure user’s view of the mobile screen. In addition, this set up requires a table where you put both webcams, so it can’t be easily used “on the go” and may be harder to put together outside of the lab. This set up may have more pros and cons, but I have to try it first to find out. I’ll let you know when I try it out.</p><p>In the meantime, you can find out more about mobile UX research from this article published in UX Magazine — <a href="http://uxmag.com/articles/eight-lessons-in-mobile-usability-testing">Eight Lessons in Mobile Usability Testing</a> and here’s more information about mobile <a href="https://medium.com/@seanmelchionda/mobile-application-ux-testing-on-the-iphone-a5d93af41c0e">device screen mirroring</a> posted by our fellow Medium blogger. As far as I understand Display Recorder described in this Medium post is not a screen mirroring app, but a recording app, so it’s not going to display what users are doing in real time. This app is only for iOS devices, which is an obvious disadvantage when you want to test Android or Windows devices.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=df3e8e8f7896" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Self-report in UX research]]></title>
            <link>https://medium.com/@domi_mazur/self-report-in-ux-research-d7ae10f892c4?source=rss-99164cd33a1f------2</link>
            <guid isPermaLink="false">https://medium.com/p/d7ae10f892c4</guid>
            <dc:creator><![CDATA[Dominika Mazur]]></dc:creator>
            <pubDate>Tue, 29 Apr 2014 15:53:04 GMT</pubDate>
            <atom:updated>2016-07-29T18:31:37.425Z</atom:updated>
            <content:encoded><![CDATA[<h4>Unobtrusive measures and mediation</h4><p>The use of self-report measures in UX research has been widely criticized because users often cannot or do not want to accurately report on their preferences, intentions and attitudes. In addition, these reports may not correspond with actual user behavior. For example, people can tell you that they sporadically shop on-line at work, but when you observe them, you notice an opposite pattern — they shop on-line at work for 1 hour every day. People may not respond accurately or truthfully because a) they do not have good insight into their own behavior (e.g., they themselves don’t notice that they spend 1 hour a day on-line shopping at work), b) want to conform to societal norms (e.g., you shouldn’t shop on-line at work) and c) want to present themselves in a positive way (e.g., I want to be seen as a hard working person).</p><p>However, self-report measures can be designed to minimize such biases. For instance, researchers can use measures that ask users about their feelings, thoughts and behaviors in an indirect way. When asked in an unobtrusive way, people are not fully aware what exactly is being measured and they can’t control and alter their responses. For example, Initial Preference Test (IPT) measures self-esteem by asking people to rate letters in the alphabet. The idea is that people who have high self-esteem will rate letters in their initials higher than other letters. Most people have no idea that this test measures their feelings about themselves, so they are less likely to misrepresent their views and beliefs in this questionnaire. Because IPT registers subconscious feelings and thoughts, it may also be more accurate for people who do not have good insight into their mental process. Another example of an unobtrusive measure is Implicit Association Test (IAT) or Affect Misattribution Procedure (AMP).</p><p>I would encourage UX researchers not to give up on self-report measures because these measures can also help to answer the ‘why’ question in UX research with the use of statistical techniques such as <a href="http://www.tandfonline.com/doi/full/10.1080/03637750903310360#.U1_CaOZdUVk">mediation</a>. For instance, a UX researcher wants to know why a newly designed app is not popular among republican users. They think it may have something to do with the color scheme or symbols used to design the user interface, so they ask users to rate various color configurations and symbols. Using mediation, they are able to determine that particular symbols used to build the app are disliked by republican users, which in turn causes them to delete the app. The researchers can also determine which symbols are liked by both democrats and republicans and use them to build a new version of the app and test it.</p><p>I’m not suggesting that self-reports measures should compete with observational research. However, when they are selected and administered correctly, they can compliment observational research and add interesting insights.</p><p>Here’s an example of how I used mediation in my Ph.D. dissertation to discover that stereotypes about women’s math abilities can harm their self-integrity. Low feelings of self-integrity in turn harmed working memory and caused poor math performance on an academic test.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/930/1*ZcRmnpHQc0fJ1J58cTWKKQ.jpeg" /><figcaption>An example of mediation from my Ph.D. dissertation. Stereotypes lead to low self-integrity which causes poor working memory performance and finally leads to poor math perforance.</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d7ae10f892c4" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>