<?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 Alma Roshani on Medium]]></title>
        <description><![CDATA[Stories by Alma Roshani on Medium]]></description>
        <link>https://medium.com/@almaroshani?source=rss-96961ece9439------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Ar16Nr26c-T6h7rOyHukeg.jpeg</url>
            <title>Stories by Alma Roshani on Medium</title>
            <link>https://medium.com/@almaroshani?source=rss-96961ece9439------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 20:14:25 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@almaroshani/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[Why are some users who previously purchased from our app churning?]]></title>
            <link>https://medium.com/design-bootcamp/why-are-some-users-who-previously-purchased-from-our-app-churning-a17db0c41653?source=rss-96961ece9439------2</link>
            <guid isPermaLink="false">https://medium.com/p/a17db0c41653</guid>
            <category><![CDATA[mixed-methods-research]]></category>
            <category><![CDATA[user-segmentation]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[churn-analysis]]></category>
            <dc:creator><![CDATA[Alma Roshani]]></dc:creator>
            <pubDate>Mon, 01 Jul 2024 16:59:23 GMT</pubDate>
            <atom:updated>2026-02-01T13:18:08.040Z</atom:updated>
            <content:encoded><![CDATA[<h3>Why are some users churning?</h3><figure><img alt="A brick wall with an exit sign installed on it." src="https://cdn-images-1.medium.com/max/1024/1*mceDRXoDEQHipUWUJVO5wg.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@bernardhermant">Bernard Hermant</a> on <a href="https://unsplash.com/">Unsplash</a></figcaption></figure><h4><strong>Background:</strong></h4><p>Tapsifood is a startup in Iran’s food delivery industry, operating for about six months and currently in the early stages of its product development. As a UX Researcher of the company, I conduct user research to inform and guide product decisions.</p><h4><strong>Goal:</strong></h4><p>We need to understand why some customers are leaving and churning from our product. Although they have used our products in the past, their last use was several days ago. Identifying the reasons for this churn is crucial, as it allows us to quickly address any product issues and respond effectively if they are starting to disengage.<br> So, it’s important to find out which users started to churn.</p><h4><strong>User Segmentation:</strong></h4><p>The user selection process was my biggest challenge in this project. As our product was in an early stage and had few users, we couldn’t find churn users by, for example, sending a message to all of them (actually spamming them) and asking them if they churn. So, we need to identify target users among the smaller groups who might churn. My approach to finding churn users involved comparing each person’s behavior to their own previous behavior, as their behavior varied greatly. In the case of someone who typically shops every day, churning might be considered if he or she doesn’t shop for a week. Nevertheless, not shopping for 10 days wouldn’t be considered churning for someone who usually shops every two weeks. To illustrate this problem better, I will use the numerical example below. Assume a person shopped on the following dates; thus, the purchase intervals should be:</p><figure><img alt="A numerical sample shows a person shopped on these dates: 28 Apr, 3 May, 9 May, 9 May, 10 May, 4 Jun, 4 Jun, and 8 Jun. So, the purchase intervals are 5 days, 6 days, 0 days, 1 day, 25 days, 0 days, 4 days." src="https://cdn-images-1.medium.com/max/1024/0*wRJz09yoYkMDJOrT" /><figcaption>A person’s shopping dates and intervals</figcaption></figure><p>Now the question is, does this person churn or not? To answer this, we need to determine if it has been a long time since the user’s last purchase. This means assessing if the interval since their last purchase is “far enough.” But how do we define “far enough”? We need to compare the time since the last purchase to an index. Is “purchase interval average” a suitable index for us?</p><p>Let’s look at a numerical example. Our application was used by this person on average every 5.9 days. As his last purchase was 4 days ago, it appears he isn’t churning. A closer look reveals a 25-day interval between May 10th and June 4th, indicating a period of unusual behavior. This “25” is an outlier and doesn’t reflect their typical behavior. When this outlier is excluded, the average interval is 2.7 days, meaning they usually shop every 2.7 days. Since he purchased his last item four days ago, it appears he may have churned.</p><p>We could make a more precise assessment of whether a user has churned by comparing the last interval with the 70th percentile (meaning 70% of the intervals are smaller). If the last purchase interval is greater than the 70th percentile (or even the 80th or 90th percentile), we could say the user has likely churned. However, since our product is in its early stages and we have fewer users, stricter conditions would limit our research pool. To ensure relevant users in our study, I included screener questions in the survey, confirming that the respondents had indeed churned.</p><h4><strong>Research execution:</strong></h4><p>Since these users likely churned, their motivation to answer my questions was low. It was crucial that they felt genuine empathy for our communication to provide insightful responses. As a result, I had to communicate deeply with them in order to find answers to my questions. To achieve this, I used the “deep interview” method in the first part. As a result of the first part (deep interviews), I discovered that all problems cannot be addressed simultaneously. Therefore, we should begin by identifying the most important ones. To determine their importance, I looked at which ones were repeated the most. Therefore, the second part is needed now. In the second part of the study, I prioritized problems based on “quantitative survey” methods. In the Results section, I will share significant findings that I discovered during this process.</p><h4><strong>Results:</strong></h4><p><em>First</em>, a few days had passed since I initially identified these users, so their behavior might have changed, and some may have made purchases in that time.</p><p><em>Second</em>, genuinely churned users are less likely to respond to a questionnaire about a product they no longer use or like, resulting in a higher response rate than non-churned users.</p><p><em>Third</em>, some users with average purchase intervals, such as one day, were categorized as churned if their last purchase was two days ago. However, it wasn’t certain if they had really churned. Tightening the criteria (e.g., considering only longer intervals) would have significantly reduced the number of users in the study. Accepting the risk that some might not be churned, I included them in the target group and sent the questionnaire.</p><blockquote><em>(This article has been proofread for grammar and clarity by AI.)</em></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a17db0c41653" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/why-are-some-users-who-previously-purchased-from-our-app-churning-a17db0c41653">Why are some users who previously purchased from our app churning?</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Solving the problem of low response rate in chat]]></title>
            <link>https://medium.com/design-bootcamp/solving-the-problem-of-low-response-rate-in-chat-1e43b701a442?source=rss-96961ece9439------2</link>
            <guid isPermaLink="false">https://medium.com/p/1e43b701a442</guid>
            <dc:creator><![CDATA[Alma Roshani]]></dc:creator>
            <pubDate>Thu, 25 Apr 2024 19:50:09 GMT</pubDate>
            <atom:updated>2024-07-01T16:33:56.991Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="a photo that shows two people chatting with each other via the phone app." src="https://cdn-images-1.medium.com/max/1000/1*dY5qaAEbF2S8VTvXmbhLng.jpeg" /><figcaption>Photo created by Anahita Aghayi</figcaption></figure><h3>Context:</h3><p>Divar is Iran’s leading general marketplace with +35 million users who post +350 thousand advertisements daily in various categories: goods, automobiles, real estate, jobs, and services.</p><p>The study and analysis of user behavior on our platform, Divar, provides us with valuable insights to improve it. For example, sometimes we need to identify user problems by analyzing their data. Sometimes, we identify problems and prioritize them with the help of data analysis. Occasionally, we use data analysis to separate a specific segment of our users. This time, we use our user data to calculate the cost of our solutions.</p><h3>Problem space:</h3><p>One of our duties on the Divar Communications team was to improve communication between buyers and sellers via Divar chat. We discovered that some buyers did not receive a response from sellers when they contacted them via chat or the seller was slow to respond, which led to buyer dissatisfaction and issues.</p><p>To understand the response rate of sellers, we should have defined the response rate metric precisely. So, if the seller does not respond to the buyer within 24 hours, we assume the buyer has not received a response. Because it’s likely that the buyer has given up, or their needs have been met by that point.” Then, we calculated daily what percentage of chats the seller responded to within 24 hours. This percentage represented the response rate for chats in each category on Divar. Therefore, I compared the response rates across different categories. Compared to other categories, the response rate in <strong>real estate</strong>, <strong>vehicles</strong>, and <strong>jobs</strong> was significantly lower. So the question was, why did this happen?</p><p>We were able to solve sellers’ problems in responding when we understood this issue. Finding the answer to this question would enable us to address buyer issues and boost the response rate.</p><h3>Research execution:</h3><p>First, I checked some low-responder sellers in the three categories mentioned above. To understand why sellers aren’t responding, I decided to conduct an in-depth interview with some users. I took this decision because I suspected that some of these sellers may not want to explain why they aren’t responding because of conflicts of interest. So, I had to speak to them in a safe environment.</p><p>Here are some of their statements:</p><figure><img alt="a photo shows some quotes from sellers to answer this question: “Why they don’t answer the buyer’s messages in the Divar app?”" src="https://cdn-images-1.medium.com/max/1024/0*Hdc0xweQAUFuzZkj" /></figure><p>It appeared that most sellers did not respond to buyers because they were reluctant to receive messages in the chat section of Divar. They can also turn off chat in the application settings if they don’t want to receive messages via chat! <strong>So, why did they keep the chat feature active for their ads?</strong></p><p>However, the individuals I interviewed seemed unaware that they could disable the chat settings for their ads if they didn’t wish to respond to messages. Although chat is enabled by default when registering an ad, it can be disabled in the settings. As a result of the <a href="https://www.nngroup.com/articles/the-power-of-defaults/">default option principle</a>, people don’t seem to remember to turn off the feature when they don’t want to chat.</p><p>In addition, we observed that many of these business-type sellers were owners who operated in these categories, and their main reason for not responding was a lack of time. Additionally, we noticed that business activity was higher in those mentioned categories.</p><p>As a result, we had a new hypothesis to test!</p><p>It appeared that <strong>“Some of the business-type sellers were unaware that they could disable the default option for chat ads.”</strong></p><p>Based on this hypothesis, we could disable the default chat option for these individuals while considering the factors that we will discuss further. However, we would not close the door to those who wanted their chat to remain active, we would simply change the default settings from “on” to “off.” In that case, if someone’s chat defaults change to “off” after a limited period of inactivity, they can turn it back on.</p><p>Certainly, the actual number of responses will not change with this solution. However, confirming the validity of that hypothesis and making that change proves beneficial because of the reasons explained below:</p><ol><li>Some of our sellers who didn’t want to receive any messages in our application no longer received messages. Thus, their issues with receiving numerous chat notifications were resolved.</li><li>Buyers contacting those sellers would expect a response due to the active chat status in the app. After this change, buyers no longer expected responses from those who were not responders.</li></ol><h3><strong>Considerations</strong></h3><p>Changing the default chat setting for business-type sellers incurred costs for us. This was due to a lack of infrastructure for such operations, requiring back-end changes. Additionally, it was essential to consider that not all of these sellers fell into the category of non-responders to chat. Therefore, first, we had to make our definition of “non-responder” persons more precise.</p><p>Individuals behaving like businesses in those three categories probably had 3* or more live ads on Divar. Naturally, not all of these people were “non-responders”. So, we infer that those of them who respond to less than 30%* of their messages might be categorized as “non-responders”.</p><p>*: To determine these figures, I collaborated with our team data analysts and saw the data of all users. By checking the number of active ads for all sellers and their response rates, we obtained an initial insight into these numbers. However, for confirmation, we eventually conducted these calculations with other assumed values. Ultimately, the resulting values represented the best and least risky possible outcomes.</p><p>Secondly, we must illustrate the percentage of issues that will be resolved for individuals if we implement this modification. Thus, team members could assess the cost and impact of this change and make informed decisions.</p><h3><strong>Hypothesis verification</strong></h3><p>We needed to examine how disabling the default chat setting for the mentioned sellers at the beginning of the month would impact the overall response rate by the end of the month. To understand the potential impact of our changes on future outcomes, we looked into the past. We observed how this change could potentially affect the response rate.</p><p>For this calculation, we simplified several assumptions, to have an approximate estimate of the difference in response rates between the beginning and end of the previous month:</p><ol><li>We assumed that at the beginning of the previous month, everyone with an active chat was considered a responder. Under this assumption, the response rate was 45.7%, acknowledging that the actual rate is certainly equal to or less than this value. This is because some individuals responded to their messages later than 24 hours, and they should not be counted as responders in such cases.</li><li>We assumed that by the end of the previous month, individuals with less than 30% response rates had their chat turned off, receiving no further messages and not activating their chat. In this scenario, the response rate reached 73.2%. However, in reality, some of these individuals (with very minimal response activity) might have had their chat turned on. Still, we assumed that their number was likely limited, as they responded to less than 30% of their messages over a month. Therefore, we did not make an overly risky assumption.</li></ol><p>The result was very intriguing! As mentioned, we observed that the response rate could change from approximately 45.7% to 73.2%, which is a notable increase. With these calculations, we demonstrated how addressing the technical costs of this change could have a significant impact on increasing the response rate and reducing issues for both buyers and sellers.</p><p>Certainly, a more precise approach would involve conducting an A/B test. However, the associated time and infrastructure costs make such experiments much more complex. Therefore, we opted for this path of estimation and calculation.</p><blockquote>(This article has been proofread for grammar and clarity by AI.)</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1e43b701a442" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/solving-the-problem-of-low-response-rate-in-chat-1e43b701a442">Solving the problem of low response rate in chat</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Case study: A generative research about the “Deposit” feature]]></title>
            <link>https://medium.com/design-bootcamp/case-study-a-generative-research-about-the-deposit-feature-e5276d6d24e5?source=rss-96961ece9439------2</link>
            <guid isPermaLink="false">https://medium.com/p/e5276d6d24e5</guid>
            <category><![CDATA[mixed-methods-research]]></category>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[generative-research]]></category>
            <dc:creator><![CDATA[Alma Roshani]]></dc:creator>
            <pubDate>Sat, 01 Oct 2022 15:48:48 GMT</pubDate>
            <atom:updated>2022-10-21T16:02:28.255Z</atom:updated>
            <content:encoded><![CDATA[<h3>Case study: a generative research about the “escrow payment” feature</h3><figure><img alt="A woman who is holding cash and trying to pay for something." src="https://cdn-images-1.medium.com/max/1024/1*IacQqpq4o1w7rntGNKpciw.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@sharonmccutcheon?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Alexander Grey</a> on <a href="https://unsplash.com/s/photos/paying?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><h3>Context</h3><p>Divar is Iran’s leading general marketplace with +35 million users who post +350 thousand advertisements daily in multiple areas: goods, automobiles, real estate, jobs and services.</p><p>This case study describes a project I worked on as a UX researcher with a goods team. Goods team provides a platform for advertising products ranging from electronic devices to home and kitchen gadgets. In this note, I detail an experience of collaborating with several different UX departments on the “escrow payment” issue. In the following I describe the problem statement and the details of the research and its results.</p><h3>Problem statement:</h3><p>Based on the data from Divar user reports since March 2021, the goods category was identified as one of Divar’s most fraudulent categories. And the most common type of fraud that happened in this category was related to “escrow payment” fraud.</p><p>An example of this happens when the fraudulent seller gets a escrow payment from the buyer but never send them any commodity, or even worse there are not any real commodities at all!</p><p>Meanwhile, in many places in the application, we warn buyers not to pay without receiving the goods. For example:</p><figure><img alt="In Divar’s app, there are two frames warning buyers not to pay until they receive the goods." src="https://cdn-images-1.medium.com/max/1024/1*Mc0wU10pOqNU2MWhmDGQcA.png" /></figure><p>However, we receive a lot of fraud reports in this category. Why do people still do that? How can they trust buyers to pay the escrow payment to them? Or how could fraudsters gain buyers’ trust?</p><p>In fact, buying goods is a buyer’s need. Since we did not provide them with any alternative solutions in the Divar application, they began looking for solutions to convince themselves of the legitimacy of the transaction. The buyer who has chosen the desired product and intends to buy it needs to make sure that the transaction is safe for him. This is so that he can complete the transaction without worrying about losing the product.</p><p><em>This research is one of my most effective studies. Because I saw how many people lost their money unfairly. People, that amount of loss was impressive to them. And now there was nothing they could do. I would be very happy if we could solve their problem so that the losses of this model will never occur again.</em></p><h3>Research Execution:</h3><p>This research had two phases: qualitative and quantitative. In the qualitative phase, I tried to find the main reasons for my question. For the quantitative part, I prioritized them.</p><p><strong>Qualitative phase:</strong></p><p>As we checked the reports of fraud ads, we knew some of them were fraudulent. So we had access to those who had reported them. We were able to identify which part of them had made an escrow payment but not received anything in return. So, during in-depth and semi-structured interviews with these people, we asked them about what happened and heard their voices. Our main questions were:</p><p>1- According to your report, you had been scammed recently. Is that true? (purpose: data screening + warm-up)</p><p>2- Can you describe what happened? (purpose: reminding of the event)</p><p>3- How did the seller earn your trust? (purpose: identifying how scammers build trust)</p><p>Also, based on the answers given by these people, we could ask more questions.</p><p>The next step was to group their answers and separate them. Using the Affinity Diagram method, related facts were grouped into two main groups.</p><ul><li>The features of the seller’s character cause buyers to trust them.</li><li>Other features that were not relevant to sellers caused buyers to trust them.</li></ul><p>There are also subcategories within each class. Here are a few of them:</p><figure><img alt="An affinity diagram that shows two main categories of reasons for paying a deposit. Each category consists of several subcategories." src="https://cdn-images-1.medium.com/max/1024/1*WzM94d22b2QTwtzeWoJaAQ.png" /><figcaption>(Details have been removed for confidentiality.)</figcaption></figure><p>We now know some of the reasons. But do they have the same priority? The answer is no. We need the quantitative phase to determine priorities. The importance of this work is that we needed to pay attention to our most frequent cases when designing and implementing a solution for this problem. This was so that we could have the maximum effect.</p><p><strong>Quantitative phase:</strong></p><p>We need a much larger sample size in this phase. Because we wanted to generalize our conclusions to the population of people affected by fraud. So I used a questionnaire to determine the priority of options in two separate questions.</p><p>Choosing an option does not affect others since each user can choose whatever they want. In order to avoid primacy bias, I randomized the choices. To reduce errors as much as possible, we checked the copies of the questionnaire with the Divar UX writer. As well, I eliminated spam responses by verifying people’s responses with a question. The results were:</p><figure><img alt="Two charts that rank reasons in each category." src="https://cdn-images-1.medium.com/max/1024/1*zPMg_8BIGQoWRbTY4rrO1A.png" /><figcaption>(Numbers and details have been removed for confidentiality.)</figcaption></figure><p>For the population who are willing to pay the escrow payment, the results are valid at a significance level of 95% and a margin of error of 6%.</p><h3>Conclusion</h3><p>At this stage, we know about how people trust sellers and what reasons are more critical than others. To find solutions, the outputs were delivered to the design department. “Escrow Payment Service” is the result of these findings. We hope this solution will reduce Divar’s fraud rate.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e5276d6d24e5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/case-study-a-generative-research-about-the-deposit-feature-e5276d6d24e5">Case study: A generative research about the “Deposit” feature</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>