<?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 Nastia Larkin on Medium]]></title>
        <description><![CDATA[Stories by Nastia Larkin on Medium]]></description>
        <link>https://medium.com/@nastialarkina?source=rss-28bc0334c30a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*hZS13kPULC6tmHJEIl9Hpw.jpeg</url>
            <title>Stories by Nastia Larkin on Medium</title>
            <link>https://medium.com/@nastialarkina?source=rss-28bc0334c30a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 04:30:38 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@nastialarkina/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[I led my first AI product. Here’s what nobody warned me about]]></title>
            <link>https://nastialarkina.medium.com/i-led-my-first-ai-product-heres-what-nobody-warned-me-about-d9c4e33d4262?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/d9c4e33d4262</guid>
            <category><![CDATA[ai-product-management]]></category>
            <category><![CDATA[ai-products]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Tue, 26 May 2026 15:31:01 GMT</pubDate>
            <atom:updated>2026-05-26T15:31:01.048Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Jj_ViCMSHTVybiLcL8mw1g.png" /></figure><p>Three months ago I led my first AI product from scratch. In this article I share everything I got wrong, got surprised by, and eventually figured out. If you’re at the beginning of this journey, I think it’s most valuable to hear from someone who just went through it.</p><p>Before this product I had no experience building AI products, though generative AI wasn’t new to me as I had worked with image and music generation, text and copy, and qualitative and quantitative data analysis. What I hadn’t done was get under the hood: design agents, write system prompts, or ship something where the LLM is the core of the product. I also hadn’t led a team through something this technically ambiguous before. So there were two learning curves running in parallel: understanding the technology, and figuring out how to lead a team when nobody, including me, could predict what would work.</p><p>I work at an engineering intelligence platform: dashboards, graphs, tables. The product I was starting three months back is an AI layer on top of an existing data product. The AI layer lets customers query all that data in natural language and get answers that go beyond opinionated dashboards: arbitrary intersections of data, root causes behind spikes, explanations for anomalies.</p><p>The technical solution was a custom agent using an external LLM to query our data. That meant designing a system prompt on our side. In practice, three things combine to produce an answer:</p><ul><li><strong>Customer prompt</strong> — what the user types in our UI</li><li><strong>System prompt</strong> — our instructions that tell the LLM how to query our data</li><li><strong>The LLM itself</strong></li></ul><p>Both prompts go to the LLM together. The LLM can loop internally several times before returning a final answer — on each iteration it calls a <strong>tool</strong> (a query against a specific dataset on our side), checks the result, and decides whether it has enough to answer or needs to dig further.</p><p><em>A note on scope: this article doesn’t cover user problems, discovery, or how we validated value. That’s a separate story. This is purely about the technical side, what I wish I’d understood before we started building.</em></p><h3>Observability and evaluations are prerequisites, not afterthoughts</h3><p>Before anything can improve, you need to see what’s happening. We used Langfuse to trace the agent’s behavior. Two terms worth knowing:</p><ul><li><strong>Trace</strong> — one full customer request end to end</li><li><strong>Span</strong> — a single step within that request: an LLM call, a tool call, a streaming chunk</li></ul><p>Without this, you’re flying blind. And as a PM leading the team, being blind is especially costly: you can’t make informed prioritization calls, and you can’t have honest conversations with stakeholders about what’s actually going on.</p><p>Alongside observability, you need <strong>evaluations,</strong> essentially automated tests for LLM output. Each evaluation prompt needs a ground-truth answer so you can actually measure correctness. Evaluations track two things: <strong>quality</strong> (is the answer right?) and <strong>efficiency</strong> (latency, number of tool calls, whether they’re parallelized). For correctness you have two options:</p><ul><li><strong>Human evaluation</strong> — reliable, but slow and doesn’t scale</li><li><strong>LLM-as-a-judge</strong> — faster, but less reliable</li></ul><p>In practice you end up using both.</p><h3>Latency was our first real problem</h3><p>Once we had the first prototype in place, the first thing internal users noticed is replies coming too slow. This was the first thing the team had to fix. We tackled it in several ways:</p><p><strong>Streaming.</strong> There are two places to apply it: between the LLM and the backend, and between the backend and the frontend. Getting both right makes the product feel much more responsive even before the full answer is ready, because the answer starts to show piece by piece.</p><p><strong>Loop limits.</strong> We set a cap on how many times the LLM could iterate internally. This also helped avoid timeouts in some cases.</p><p><strong>Sub-agents instead of one big system prompt.</strong> The main loop uses tools defined on our side (querying specific datasets). Sub-agents take this further: instead of one large system prompt trying to handle everything, you split the work across smaller, specialized agents each responsible for a specific type of query. This keeps individual system prompts lean, which directly reduces time to first token.</p><p>One thing my team experimented with was narrowing the context the LLM works with, which significantly reduces reply time. When the agent is called from a specific page, we pass that page’s context directly into the system prompt, so instead of reasoning across the entire dataset, the LLM works only within that scope. It’s not a universal solution, but for cases where there’s a clear context to anchor on, it made a noticeable difference.</p><p><strong>Notifications.</strong> For longer queries, we notified customers when their answer was ready rather than making them watch a spinner.</p><p><strong>Reducing reasoning level.</strong> This was the most impactful change. Switching reasoning to low by default cut latency dramatically. To make sure we weren’t sacrificing quality, we ran evaluation batches at different reasoning levels and compared correctness scores. This kind of decision, trading off one dimension against another with data to back it up, is where having evaluations in place made the team’s job much easier.</p><h3>Creating the impression of responsiveness while the agent works</h3><p>This was one of the less obvious things to build. To make the wait for a reply feel shorter, I added a step where the initial customer prompt gets rephrased and shown as a “thinking” summary. This runs through a cheap, low-reasoning model so it’s fast.</p><p>But it does more than just look good. While rephrasing, that model also tries to extract useful parameters from the query (team name, time frame, best guess at which tool to call) and passes those to the main agent under the hood. The wait feels shorter, and the agent starts with better context.</p><h3>PMs can and should own the system prompt</h3><p>I had direct access to edit the system prompt and test results myself. This turned out to be one of the most valuable things I could do, both for the product and for how I worked with the team.</p><p>I used it to set guardrails: blocking irrelevant questions (someone asking for a pie recipe), protecting cross-team data, restricting access to sensitive data. These are product decisions and having the ability to implement them directly removed a lot of back-and-forth. It also meant the engineers could focus on harder problems.</p><p>It also became a tool for our designer, who used it to control output formatting, something that’s often treated as an afterthought but turns out to be fully manageable through the prompt.</p><h3>You don’t always need RAG to feed internal knowledge to the agent</h3><p>The standard advice for getting an LLM to reason about internal knowledge is RAG (retrieval-augmented generation): vectorizing documents and retrieving relevant parts upon request. We didn’t do that. Instead, I distilled our internal knowledge base into a concise markdown document, and the team attached it directly to the system prompt. No vectorization, no retrieval pipeline. It works well for a focused, stable knowledge base. We haven’t hit the limits of this approach yet, but in theory, once the knowledge base grows large or changes frequently, a proper RAG setup would become necessary.</p><h3>You never really know if it will work</h3><p>This is the thing nobody warns you about clearly enough. With a traditional product, you can predict the outcome with reasonable confidence. With an AI product, that certainty disappears. The LLM is not deterministic. The same input can produce different outputs, and a tweak that fixes one class of queries can quietly break another. You’re not engineering a system so much as negotiating with one. This uncertainty doesn’t just affect the product, it affects how you lead. Timelines are harder to estimate, “done” is harder to define, and keeping a team motivated through that ambiguity is its own challenge. If I had understood this at the start, I would have set expectations differently from day one: with stakeholders, with the team, and with myself.</p><p>Both learning curves, the technical one and the leadership one, turned out to be the same curve, really. The better I understood what I was building, the better I could lead the people building it with me.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d9c4e33d4262" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Intuition, Data, Vision: Three Ways Product Culture Shaped My Career]]></title>
            <link>https://nastialarkina.medium.com/intuition-data-vision-three-ways-product-culture-shaped-my-career-55d943f42201?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/55d943f42201</guid>
            <category><![CDATA[research]]></category>
            <category><![CDATA[product-culture]]></category>
            <category><![CDATA[product-thinking]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Fri, 26 Sep 2025 05:51:48 GMT</pubDate>
            <atom:updated>2025-09-26T05:51:48.853Z</atom:updated>
            <content:encoded><![CDATA[<p>This article reflects on my career in big tech and startups, and on adapting to product cultures that don’t always fit.</p><h3>An Intuition-Driven Culture at Yandex</h3><p>In the first half of my career, I worked as a product designer at <a href="https://yandex.com/">Yandex</a>. Today, Yandex is a big tech company with about 37,000 employees. It operates in search, e-commerce, ride tech, and many other areas.</p><p>Yandex was founded in 1997 by two schoolmates and began as a search engine. When I joined the company in 2008, it already had 1,500 employees and its product culture was still evolving. At that time, there were no product managers at all. Eventually, it became clear that market thinking was needed, so the company started to hire marketing managers whose responsibility was to define the product experience. Development back then was mainly driven by intuition. Several years later, the term <em>usability</em> popped up. Following industry trends, the user experience research department was established. Its main function was to conduct usability testing of mockups and prototypes of future features. Problem research still wasn’t a thing at that time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ASsJRq83lGXW3psz77vbaQ.jpeg" /><figcaption>Office vibes at Yandex, 2013. Image by Stas.</figcaption></figure><p>I came across problem research at Yandex for the first time when I was part of the e-commerce department. A large series of in-depth interviews with several user segments was conducted to explore their shopping motivations and behaviors. The results of that research were not used directly to define big bets or prioritize features, but they contributed to the overall understanding of user needs.</p><p>After almost 10 years at Yandex, and considering myself a fairly mature designer, I began interviewing with international tech companies. It quickly became clear that I lacked product thinking. Most of my previous experience had been intuition-driven, with only sporadic use of data. Viewed through the lens of international standards, I was somewhat disappointed by the company’s product culture.</p><h3>Learning Data-Driven Product Work at Avito</h3><p>My next big career step was <a href="https://en.wikipedia.org/wiki/Avito.ru">Avito</a>, formerly part of the Naspers Group. Avito is a large classifieds platform with 64M MAU. By joining a new organization, I also entered a completely new product culture, the one strongly based on data. The art of combining qualitative and quantitative data to shape future product decisions reached its peak at Avito. The mentoring mindset of senior product managers helped others adopt this approach. It was at Avito that I first learned about the double diamond methodology. Only then did I clearly understand the difference and importance between <em>problem</em> and <em>solution</em> research. Data was king in that company, and it became king for me. To support my day-to-day work, I also completed two heavily data-driven courses for product managers at <a href="https://gopractice.io/course/pm/">GoPractice</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oM46lQVrEbqTwxQfjV9p5Q.jpeg" /><figcaption>A working session at Avito.</figcaption></figure><p>The downside of this data-driven approach was data paralysis. Research could take an entire quarter to complete, which significantly slowed development and time to market. There was very little risk in the solutions, but at the same time they reached the end user much later than they could have.</p><h3>Balancing Vision and Data at Yousician</h3><p>After three years at Avito, which was a fantastic school for implementing a data-driven approach, I joined <a href="https://yousician.com/">Yousician</a>, an international mid-sized music education company. By that time, the Yousician app had already existed for nine years. For most of that time, it had been vision-driven. With a very strong CEO deeply involved in defining product development, the company followed its mission and vision. At Yousician, however, I encountered a new approach to research: it is conducted continuously. Once a week, we invite a respondent to the office, and the scenario for each research session is based on current needs. We either test the builds or mockups under development, or walk through major app experiences. The insights from each session are not used directly or immediately in development, since it would be risky to act on a single signal. Instead, we log the findings and summarize them once we have accumulated enough. All this contributes to building user empathy and developing a deep, integral understanding of user needs and behavior. Based on this empathy, the vision is refined. We don’t research each user problem separately, but rather form an understanding through continuous communication with users.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZoJp_vZ1tUroW7KFex3Ddg.png" /><figcaption>Raw data from a workshop at Yousician.</figcaption></figure><p>The downside of this approach is that covering all user segments is very time-consuming.</p><p>Another way to build user empathy is by following social media on the subject. The app has many loyal subscribers, and our social media marketing department has managed to build a community around it, since music as a hobby and learning is a topic people love to discuss.</p><p>I am still a little uncomfortable with the fact that big product bets can be defined without validated, coherent research. On the positive side, skipping these research steps helps the team move quickly and avoids paralysis from excessive data.</p><p>This vision-driven approach works best with a charismatic leader who carries the vision forward. Such a leader can explain the thinking behind it to the team, so that everyone understands, internalizes, and believes in it as strongly as the leader does. As a company grows, the leader is no longer able to communicate directly with the whole team, and interim levels of communication appear. In this case, it becomes increasingly difficult to convey the vision and inspire people to follow it. And thus, rational agreement comes onto the stage. Gradually, research results begin to matter more in convincing and inspiring the team to move in one direction or another. It is interesting to be part of this change process. However, even in a vision-driven culture, it is possible to use a data-driven approach to validate solutions and mitigate risks.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=55d943f42201" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Pros and Cons of Automatically Creating Song Backing Tracks for a Music Education App]]></title>
            <link>https://nastialarkina.medium.com/pros-and-cons-of-automatically-creating-song-backing-tracks-for-a-music-education-app-d96805a1a8ed?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/d96805a1a8ed</guid>
            <category><![CDATA[case-study]]></category>
            <category><![CDATA[music-education]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Tue, 09 Sep 2025 12:36:05 GMT</pubDate>
            <atom:updated>2025-09-09T12:36:05.182Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZP0BYmyF0TyhmDk6fc_CwQ.png" /></figure><p>This is the story of how we expanded the Yousician library in an inspiring and safe way.</p><p><a href="https://yousician.com/">Yousician</a> is a music learning application. At its core is the academic path, which users follow to gain basic skills and master techniques. Previous <a href="https://medium.com/@nastialarkina/how-the-bite-sized-approach-to-learning-music-set-users-up-for-success-e87b8ff0b132">experiments</a> showed that at the start of the learning journey users don’t care much about which songs to play. However, over time, users tend to lose interest in structured learning and shift toward practicing newly acquired skills and playing songs for fun. That’s where the Yousician song library comes into play.</p><h3>‘I Can’t Find Songs I Love’ is a Key Churn Problem</h3><p>In 2024, Yousician had about 500 songs, but research showed that not finding favorite songs was a key reason users left.</p><p>This relatively small number of songs is due to the manual effort required to produce each one. Backing tracks are created for multiple instruments (guitar, piano, ukulele, bass, and voice ) across different techniques (melody and chord-based arrangements) and difficulty levels. As a result, each song has several versions to cater to different user segments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v3TnA5om3hqTRol9ES3QdA.png" /><figcaption><em>The numbers next to the song titles indicate the difficulty level of each version.</em></figcaption></figure><p>Overall, backing track production is time-consuming and requires specialized musical skills.</p><h3>Automated Backing Tracks as a Potential Solution</h3><p>To tackle this problem there was created the automated system that generated backing tracks based on the chord sequences we already had for each song. This expanded our library 15 times: from 500 to 7,500 songs. However, the tracks were very basic, including only drums and bass, with no voice, melody or recognizable riffs.</p><p>The team was excited about the idea, but the launch carried significant reputational risk due to the low quality of the content compared to manually created backing tracks. It might even hurt subscriptions if users saw these songs early on.</p><p>Therefore, it was crucial to define a careful go-to-market strategy that would allow us to deliver value through the expanded song collection while mitigating brand image risks. As an initial step, all automatically generated song versions were labeled so they could be easily recognized on the search results page.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V0GSCYX6b_lAX-tJTv1aXg.png" /><figcaption><em>The label sais ‘Made with AI’</em></figcaption></figure><h3>Changes Driven by User Research Insights</h3><p>First I conducted user research by showing the library to people and asking them to play songs. This revealed several technical issues and a major expectation problem.</p><p>The label <strong>“Made with AI”</strong> caused confusion and set the wrong expectations. We realized we needed a clearer way to communicate what these tracks were. The obvious option, <strong>“Drum and Bass,”</strong> was rejected because it could be mistaken for the music genre.</p><p>To address this while still communicating the essence of the product the label <strong>“Beat Track”</strong> was introduced.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qNOhsUWfV7dAJQnjjSWr4A.png" /><figcaption><em>We also added Beta in brackets in order to set realistic expectations.</em></figcaption></figure><p>Respondents were excited to see familiar song titles and had high expectations for quality. However, they were often disappointed, as the songs were only partially recognizable. These backing tracks worked best when users were already very familiar with the song, essentially when it “<em>played in their head</em>”.</p><p>Based on this insight, communication was added to set realistic expectations before users played these songs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ht1Q3iQNH900vJeqIPN_ZA.png" /><figcaption><em>Whenever a user selected a Beat Track from search results, a popup appeared.</em></figcaption></figure><p>Another finding from the research was that when users started playing a familiar song, they tended to mumble or sing along to help themselves get into the rhythm and feel of the song. To support this karaoke-style lyrics were added to the main play mode.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_5m2QJ0RM-YP1NIf5JjKvw.png" /></figure><h3>Collecting Feedback from the User Community</h3><p>Afterwards we conducted a soft launch with our user community. We have around 1K loyal users and a communication channel with them via social media. The goal was to collect feedback on the new functionality.</p><p>Based on community reviews we designed a feedback collection form to integrate into the app. Thus we could get both qualitative and quantitative data.</p><p>The feedback was designed to be actionable, so we could make decisions based on it. For example, removing songs with low ratings and prioritizing the most popular songs for manual backing track creation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SN0m_UhYeYf1nfYd_IfJEg.png" /></figure><h3>Further experimentation: Surfacing the Functionality</h3><p>Once we confirmed that the quality of the automatically generated backing tracks was acceptable, we conducted an A/B test on half of our user base. The goal was to track retention metrics and determine whether the size of the song library influenced users’ motivation to start a subscription. The hypothesis was that if users could find songs they wanted to play, they would be more likely to subscribe.</p><p>The first experiment did not show any statistically significant change in metrics. Based on these results, we formed the following assumptions:<br>— unclear value communication<br>— poor discoverability of the functionality</p><p>To give users more control over content discovery, we added a search filter that included Beat Tracks by default. However, the filter appeared as a white bubble with a close button at the top of the search results, which looked like a banner. As a result, 80% of users closed it without reading, so most never saw the Beat Tracks at all.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RJkL5WEDzuTAhiKZgMUduQ.png" /><figcaption><em>Previously, the “Include Beat Tracks” filter was on by default and at the top of search results (right). In the new version (left), it was changed to an exclusion filter and set to off, making Beat Tracks more visible to a larger audience.</em></figcaption></figure><p>To address this issue we redesigned the filter as an exclusion filter. Now it is labeled “Exclude Beat Tracks” and set to <em>OFF</em> by default. Users would need to go into the filters page to enable it, which adds friction and makes it less likely to be used.</p><p>This change led to an increase in search hit rate and a decrease in song exit rate, both positive signals. However, the effect was seen only for existing old subscribers and did not impact new users or monetization metrics. This suggests that the number of songs in the library does not drive subscription motivation for beginners. It confirms the broader insight that content variety is more valuable to advanced users, while beginners prioritize guidance over discovery. This supports the differences in how <a href="https://medium.com/@nastialarkina/early-vs-advanced-users-motivation-and-behavior-in-music-learning-f03e8a343427">users approach music learning depending on their level</a>. These insights provide data for the next steps in more carefully tailoring the functionality to those who truly need it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d96805a1a8ed" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Early vs. Advanced Users: Motivation and Behavior in Music Learning]]></title>
            <link>https://nastialarkina.medium.com/early-vs-advanced-users-motivation-and-behavior-in-music-learning-f03e8a343427?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/f03e8a343427</guid>
            <category><![CDATA[music-education]]></category>
            <category><![CDATA[case-study]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Mon, 11 Aug 2025 13:57:15 GMT</pubDate>
            <atom:updated>2025-08-11T13:57:15.014Z</atom:updated>
            <content:encoded><![CDATA[<p>While working as a product manager at <a href="https://yousician.com/">Yousician</a>, a SAAS music learning app, I learned a lot about how people approach acquiring new skills and their music hobby overall.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IPtF4FXZHaMfG_eu1ijKaw.png" /><figcaption>The Yousician app listens as you play and gives instant feedback.</figcaption></figure><p>Yousician’s target audience includes complete beginners and people who have played guitar in the past. These users may have never picked up a guitar or may have played a few chords years ago and now want to return to their hobby. They all need guidance at the start of their journey.</p><p>As users progress their goals and behaviour change, so the app experience should adapt too. In this article, I’ll share examples of hypotheses we tested that led to unexpected results. What worked for beginners often didn’t work for advanced players, and vice versa.</p><p><strong>Content Experiments:</strong></p><ul><li>Familiar song in first session → ❌ Didn’t help beginners</li><li>Genre-based personalization → ✅ Boosted beginner engagement</li><li>Letting users choose first song → ❌ Hurt beginner flow</li></ul><p><strong>Play Mode Experiments:</strong></p><ul><li>No-fail mode → ❌ Didn’t help beginners, ✅ Helped advanced users</li><li>Stars in Practice mode → ⚖️ Neutral for beginners, ✅ Motivated advanced users</li></ul><h3>Bringing familiar music to the first session</h3><p>When we asked users why they started learning guitar, the most common responses were:</p><ol><li>To play for fun</li><li>To play music I like</li><li>To improve the skills</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EGcZjuSx62t8qPMyB_2G5A.png" /><figcaption>Users’ goals in their music hobby.</figcaption></figure><p>However, people typically have multiple goals, not just one. We were particularly interested in how important playing their favourite music is and when this becomes a priority in their learning journey.</p><p>Historically, Yousician only used non-licensed content created in-house specifically for the app. However, the company later began establishing licensing agreements to include popular songs.</p><p>The first experiments focused on introducing well-known content early in the learning path. The hypothesis was that by offering recognisable songs early, we could motivate users to subscribe in order to keep exploring the app. For example, we tested Ed Sheeran’s song <em>I See Fire </em>(I explained that experiment in detail in <a href="https://medium.com/@nastialarkina/how-the-bite-sized-approach-to-learning-music-set-users-up-for-success-e87b8ff0b132">this article</a>), which appeared in the very first session. The target metrics were first level completion (since learning content is divided into levels) and subscription conversion.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yFKlHnGP8uN8LvaAaqcRyQ.png" /></figure><p>Surprisingly, the licensed songs didn’t have the desired effect, and the experiment results were neutral. We concluded that early in the learning process, guidance is more important than recognisability. Users care less about what they’re playing and more about enjoying the experience of playing itself.</p><h3>Personalizing early experience through genre preferences</h3><p>Personalization of the experience was a hypothesis that had been floating around. To test it easily we decided to tailor the learning path based on users’ favourite genres. Luckily the app already had an onboarding survey that asked users to choose their preferred genres.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-tk5Tiy-1PKm8qXlg27CnA.png" /><figcaption><em>Onboarding question regarding user’s preferred genres.</em></figcaption></figure><p>Using the survey data we decided to filter some of the learning content. The main goal was to improve Day 1 Play retention.</p><p>At the end of certain missions users see “<em>Now You Can Play</em>” collections. These are large groups of songs they’re now ready to play. To progress, they only need to play one song from the collection.</p><p>Our solution was to personalize these collections by showing songs that match the user’s genre preferences from onboarding.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4lxMT4XLkwVqCu4Rvuwc2w.png" /><figcaption><em>Now You Can Play collection as seen from the learning path and it’s content.</em></figcaption></figure><p>The target metric, Day 1 Play retention, increased by 10%. The biggest impact was at early levels, and the free trial to subscription rate (our key guardrail metric) rose by 11%.</p><p>Unlike adding licensed songs to the learning path, this experiment showed that some personalization works even early on, just not for the very first songs. It seems giving users a choice might matter more than using well-known songs.</p><h3>Choosing the first song to play</h3><p>Inspired by the results of the previous experiment, we wanted to give users even more choice early on. So we let them pick the first song they play in the app.</p><p>Our main metric was <strong>Signup to First Song Played Successfully</strong>, since playing a song is the “AHA” moment that leads to subscription.</p><p>To save time we used existing collection functionality, similar to “<em>Now You Can Play</em>.” The solution was a collection of four songs, where users only needed to play one to move forward.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t6QDTt5RBa0kvlxCl3RbWg.png" /><figcaption><em>The collection was called Pick and Practice.</em></figcaption></figure><p>The experiment had mixed results. While users played more songs overall, other completion rates dropped. Most importantly, fewer users went from signing up to playing their first song. This means that fewer users reached the app’s key value moment and fewer were motivated to subscribe.</p><p>It’s unclear if the drop was due to limited song choices or a less-than-ideal user experience. In the control group, the first song was autoplayed, while in the variant, the user had to go to a collection and choose a song.</p><p>In the end, we learned that offering choice is valuable even early on, but clear guidance and a smooth experience matter more at that stage. Choice becomes more important later in the journey.</p><h3>No-fail mode</h3><p>From churn research we know that one of the major reasons for users quitting is hitting the “difficulty wall.” The early levels are easy, but at some point the difficulty curve of the learning path becomes steep and users can’t progress.</p><p>After several unsuccessful experiments with substituting difficult songs by easier options (which only delayed the churn point without solving the problem), we decided to make the playing experience itself less challenging.</p><p>The main gameplay mode involves the moving fretboard, where users play notes or chords of a song. The app listens to the user’s notes, recognises them and gives immediate feedback. Correct notes turn green, while incorrect ones turn red. If a user plays 15 notes incorrectly in a row, they “die,” meaning they are kicked out from the game, and the song ends. However, they can always restart.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ApKK1uvNI_cm9gotFwkiEw.png" /><figcaption>If the chord is played wrong it turns red.</figcaption></figure><p>The hypothesis was that by removing the “dying” condition and letting users play through the whole song regardless of accuracy, we could improve the experience. We decided to measure the success by tracking song completion rates and retention.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FzElp6mWqMwQ%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DzElp6mWqMwQ&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FzElp6mWqMwQ%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/56859223b8634a8223096d8d66a3a838/href">https://medium.com/media/56859223b8634a8223096d8d66a3a838/href</a></iframe><p>It turned out that early-level users didn’t care about the “dying” condition. Either the songs were easy enough that users completed them correctly anyway, or they were just looking for a challenge.</p><p>The only group that showed a positive effect on song completion rates were users at advanced levels, those who had been using the app for several months. These power users are few in number, so the change wasn’t statistically significant.</p><p>Most Yousician users prefer the current recognition system with visual indicators for missed notes. Tolerating mistakes by removing the red “missed note” feedback takes away a valuable part of the user experience.</p><h3>Rewarding practicing</h3><p>We had a hypothesis about rewarding users for practicing.</p><p>In the app, there’s a <em>Practice mode</em> that lets users slow down songs and loop sections. Unlike <em>Perform mode</em>, users aren’t kicked out, and many who use it tend to engage more and stick around longer.</p><p>However, Practice mode doesn’t give any rewards, no scores or stars, unlike Perform mode, where users earn stars for each part of the song.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TESjOix5T4vFJR9-b2tv3g.png" /><figcaption><em>Stars appear on the song end screen after playing in Perform mode. Users earn a star for each part of the song.</em></figcaption></figure><p>From user research, we know there’s a group of users we call <em>“star hunters”. </em>They keep replaying songs until they earn all stars.</p><p><strong>Hypothesis:</strong> If users could earn stars while practicing at lower speeds, they’d be more motivated to use Practice mode and prepare better for performance.</p><p><strong>Primary metric:</strong> Number of songs played.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FHykDGcVAu10%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DHykDGcVAu10&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FHykDGcVAu10%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/3dc8a94c4c590a2d830148d79c3c942b/href">https://medium.com/media/3dc8a94c4c590a2d830148d79c3c942b/href</a></iframe><p>To test our idea, we added star rewards to Practice mode. Unlike Perform mode, where stars appear only after finishing a song, in Practice mode, stars were shown in real time. This added complexity, since users could now earn stars for each part separately. We also had to adjust the rules so users couldn’t earn full stars at lower speeds (e.g., 50%).</p><p>The change didn’t have the expected effect. Overall, songs played per user dropped by 1.7%. However, among long-time users, this metric slightly increased, though the sample size was small, so the change is statistically insignificant.</p><p>Despite mixed results we received a lot of positive feedback on social media from loyal users. It seems rewarding Practice mode makes more sense during advanced stages, when users <em>have</em> to practice to progress. At that point they value being recognized for their effort, not just the outcome.</p><h3>One size doesn’t fit all</h3><p>These experiments showed that beginners and advanced users need different experiences. Early on, clear guidance and a smooth flow matter most. Later, as users gain confidence, personalization, choice, and recognition become more valuable. To truly support learning, the product should grow with the user, not just motivate them at the start, but keep them engaged throughout their journey.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f03e8a343427" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How the Bite-Sized Approach to Learning Music Set Users Up for Success]]></title>
            <link>https://nastialarkina.medium.com/how-the-bite-sized-approach-to-learning-music-set-users-up-for-success-e87b8ff0b132?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/e87b8ff0b132</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[music-education]]></category>
            <category><![CDATA[yousician]]></category>
            <category><![CDATA[product-case-study]]></category>
            <category><![CDATA[apps]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Sat, 03 May 2025 11:12:24 GMT</pubDate>
            <atom:updated>2025-05-03T11:12:24.934Z</atom:updated>
            <content:encoded><![CDATA[<p>This story tells how <a href="https://yousician.com/">Yousician</a>, a music education app, adopted the bite-sized training approach to learning songs. Initially aimed at boosting subscription conversions, our efforts unexpectedly improved user retention which was another significant business interest. As a result, we shifted our focus and goals.</p><h3><strong>Are iconic songs a good enough motivation to start playing a guitar?</strong></h3><p>The Yousician app provides a structured learning path with songs and melodies that users can play along with a backing track. The app detects the sound of their playing and offers immediate feedback.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HrrpaI3ZPPLuzp0Dv-vFrg.png" /><figcaption><em>As the notes move along the fretboard, a jumping ball highlights the next one. When played correctly, the notes turn green.</em></figcaption></figure><p>As a SaaS app, one of our key monetization metrics is conversion to subscription. To boost this, we hypothesized about what might be a good selling point at the beginning of the learning journey. During in-depth interviews, many respondents mentioned they had a specific song in mind they wanted to learn before starting to play guitar, while others were inspired by seeing someone else play.</p><p>We proposed that including several well-known songs from famous artists in the free, basic level of the learning path could inspire users to play their first song. Playing a song is the core value of our product, and we know from experience that this is directly linked to subscription conversion.</p><p>We selected popular songs by AC/DC, Nirvana, Red Hot Chili Peppers, and Ed Sheeran, and adapted them so even beginners could play. However, being heavily simplified these songs became almost unrecognizable when played without a backing track.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yFKlHnGP8uN8LvaAaqcRyQ.png" /><figcaption><em>“I See Fire” by Ed Sheeran was added to the Basics level and could be played in the very first session.</em></figcaption></figure><p>The experiment results showed no change in our key metrics — neither first song plays nor conversion to subscription increased. However, users who played “I See Fire” by Ed Sheeran returned more often than others.</p><p>I also noticed that Simply Guitar, another music learning app, prominently features Ed Sheeran’s songs at the top of its catalog page. Could Ed Sheeran have a unique appeal?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mSwNFrFr_pEtqnXWe7zfjA.png" /><figcaption><em>Ed Sheeran’s “Perfect” in the Simply Guitar app.</em></figcaption></figure><p>From this experiment, we learned that the specific song doesn’t matter much at the start of the learning journey; people simply enjoy playing. But how can we make this playing more engaging? This question led us to our next experiment, where we put together songs for absolute beginners.</p><h3><strong>Setting Players Up for Success</strong></h3><p>We began experimenting with a bite-sized approach. Our hypothesis was that breaking songs into manageable sections would set users up for success and motivate them to continue using the app. Additionally, introducing a muscle memory trainer before a song would help users complete it.</p><p>To validate this hypothesis, we conducted a series of user tests. Our educator composed a new, fairly complicated melody for the experiment. The melody was divided into several parts, each presented as a static tab. Users were instructed to listen to each segment before attempting to play it at their own speed multiple times. Once they achieved proficiency without time constraints, they progressed to regular mode, playing at full speed. An introductory video explained the approach.</p><p>Internal user testing revealed that even very beginner users could master complex pieces using this method. We then conducted an A/B experiment, integrating the entire training method into Level 1 of the learning path. This experiment showed a higher conversion rate from free trial to subscription and an increased song completion rate.</p><h3><strong>Using bite-size Learning approach to Break the Difficulty Wall</strong></h3><p>But could this same approach benefit players not only at early stages of their learning journey but also at more advanced levels with much more complex songs? This could address one of Yousician’s major churn problems — hitting the difficulty wall. This happens when users progress along the learning path, but at some point find the content too challenging. This leads to repeated failures and quitting the app.</p><p>Simplifying the content won’t solve the problem, but rather postpone it. Thus the bite-sized approach came into play.</p><p>We examined the learning path and identified the points with the highest churn, identifying the specific songs after which users stopped using the app. The idea was to apply the bite-sized training approach to these challenging songs, helping players master them instead of quitting after a series of failures.</p><p>By knowing the areas of the learning path with the highest churn rates, we manually applied the training approach to the most difficult songs. Although initially a patch solution, we saw that systematically applying this method could transform the way users learn and perform songs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cYKqtaEnVJCHYJM_89rclw.png" /><figcaption><em>Shaolin Master is the toughest song in Level 4 of Yousician. The “Learn Shaolin Master” task introduces a step-by-step method for tackling it.</em></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*izRtcAEjZW7X2IeAbqC5lg.png" /><figcaption><em>In a video users see the song broken into smaller parts. They’re advised to practice each part slowly before moving on to the next one.</em></figcaption></figure><h3><strong>Applying training approach to short songs</strong></h3><p>Based on insights from previous experiments, we decided to fully automate our training approach. Short pieces, particularly iconic recognizable riffs, could be an ideal content type to begin with.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wuHCl6iNdjkp37_9J6i4Jw.png" /><figcaption><em>Our content team curated a collection of these iconic riffs.</em></figcaption></figure><p>We hypothesized that by applying our bite-sized training approach to these riffs, users would experience greater success and retention. The training approach consisted of the following modes:</p><ol><li><strong>Listening</strong>: Users were encouraged to listen to the song, enabling them to recognize familiar tunes or grasp the rhythm and structure if it was new to them.</li><li><strong>Playing at your own speed</strong>: Users were presented with a static tab, allowing them to play without feeling rushed.</li><li><strong>Building speed</strong>: Users were given the opportunity to first play at a slower pace. The initial speed was determined by their performance in the previous step. If users played notes correctly, the algorithm increased the speed, while errors forced the algorithm to slow down the backing track.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vQ6B6CyBTPA2Xb-5HXGDag.png" /><figcaption><em>Listening, playing at your own speed, building speed and performance modes.</em></figcaption></figure><p>Usability research and feedback from social media uncovered several UX issues.</p><p>For example, at very low speeds (below 30%), the backing track became unrecognizable. To address this, we substituted the backing track with a metronome. As users played faster, the backing track gradually appeared, serving as a reward for accurate performance.</p><p>Another issue was that users were so focused on the fretboard that they didn’t notice the UI elements guiding the experience: the steps (listening, playing without speed pressure, building speed, performing) were overlooked. To address this, we added voiceovers that briefly explained each step the first time a user went through it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*--OiMgUsx_WwKqsAxY1SMA.png" /><figcaption>The final design of the play mode that embraced the bite+sized learning approach.</figcaption></figure><p>As a low-level retention metric which is easy to track during a short time period we picked next-week play retention. In the experiment this metric increased 13% due to the new experience. Additionally, conversion to subscription, tracked as a counter metric, also grew.</p><p>Encouraged by these results, the team embarked on redesigning the training approach in order to apply it across all songs in the Yousician app. In this iteration, the focus shifted from converting users to subscription to retaining users within the app by helping to master all types of content.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e87b8ff0b132" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Creating a Museum Experience App for Kids: From Idea to Pivot]]></title>
            <link>https://nastialarkina.medium.com/creating-a-museum-experience-app-for-kids-from-idea-to-pivot-401f64e20396?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/401f64e20396</guid>
            <category><![CDATA[kids-activities]]></category>
            <category><![CDATA[pivot]]></category>
            <category><![CDATA[kids]]></category>
            <category><![CDATA[ideas-to-execution]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Thu, 24 Apr 2025 12:38:13 GMT</pubDate>
            <atom:updated>2025-04-24T12:38:13.640Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>How I came up with an idea of a new way of exploring museums with kids</strong></h3><p>A year ago, I came up with the idea for a museum app designed for kids. My enthusiasm led me to join a bootcamp, with plans to eventually enter an accelerator and build a startup around it. I wrote about it in <a href="https://medium.com/@nastialarkina/my-bootcamp-experience-insights-challenges-and-lessons-learned-46ce15acda78">this article</a>. Though I didn’t end up moving forward with this exact path, the research and insights I gathered on the way were invaluable. They helped me refine my idea into something more practical — something that will not only provide value to engaged parents but also be personally enjoyable to create.</p><p>As a mom of two with a deep love for museums, I started taking my kids to exhibitions from the time they were newborns. To make these visits engaging I created lists of objects, printed them out, and had my kids mark off items as they spotted them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZQUiD4ZvwH1aVNutTT6YiA.png" /><figcaption><em>My kids going on quests in Copenhagen, Helsinki, and Istanbul.</em></figcaption></figure><p>This habit led me to take countless photos of museum displays while traveling solo, hoping to gamify future visits with my children. While many museums now offer their own interactive experiences for kids, I realized it would be exciting to create something that could gamify not just one visit, but a series of museum experiences — perfect for families exploring multiple museums over school holidays.</p><p>My initial idea was to create an app to make museum visits more engaging for kids. The app would feature quests for museums worldwide, where kids receive lists of objects to find in exhibitions. As they discover each item, they can mark it as found. Gamification elements like badges would reward them for achievements — such as spotting 100 objects, visiting 10, 20, or 30 museums, or exploring museums with themes like art, nature, or science.</p><p>This app would make offline learning more exciting for kids, keeping them engaged and relieving parents from the need to entertain them. The reward system encourages kids to visit more museums, helping parents plan future weekends. There’s even potential for kids to take the experience further by creating their own digital museums with objects they’ve collected, which they can share and explore with friends.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HBp-PB0SMFRXWLeUxWrfqA.png" /><figcaption><em>Major user flow: choose a museum, then mark an object as found.</em></figcaption></figure><h3><strong>Social media survey</strong></h3><p>As a first step in validating my idea I conducted a survey on how families spend free time together, with a focus on museum visits. I shared the survey through my personal social media, so the results may be biased due to the sample being primarily friends and friends-of-friends. The survey received 40 responses, with answers varying significantly based on children’s ages. Most respondents had kids between 4 and 10 years old.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tw5Fqq4rHyaCZ6ZMDxaOxA.png" /></figure><p>Half of the respondents reported visiting museums regularly with their kids, making it the fourth most popular activity after outdoor playgrounds, home-based activities, and zoos. When traveling, 70% of families include at least one museum visit in their itinerary.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PVqSVSuIN6KJ04Zxjh5Mgg.png" /></figure><p>In the past year, only 12% of families reported not visiting a museum at all. Meanwhile, 30% visit museums regularly, and 50% make between 1 and 4 visits annually.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9yFT1ZfCew1KT7eVcGlklg.png" /></figure><p>Technical, science museums and castles are the most popular choices, while art and history museums are less commonly visited. However, the experience varies greatly depending on the museum itself. During visits kids often ask questions that parents answer; about a quarter read exhibit descriptions or use audioguides, while nearly half simply explore on their own.</p><p>More than half of the kids have their own gadgets, and another 24% use their parents’ devices. Most devices (85%) are completely controlled by parents — they either set time limits, install the apps, or both. Only 10% of kids install the apps on their own.</p><h3><strong>Market estimation and monetisation</strong></h3><p>Based on these responses, I was able to make preliminary market estimations. The population of Europe is 746.4 million, with approximately 15% of that being children aged 0–14 (according to theglobaleconomy.com). I estimate that children aged 4–12, the target audience, make up about 60% of this group, or roughly 9% of the total population — around 69.1 million kids in Europe.</p><p>Applying similar logic to the United States, where children make up 18% of the population, the market size would be 35.8 million kids. This brings the total addressable market (TAM) to <strong>104.9 million kids</strong> (69.1M in Europe and 35.8M in the US).</p><p>For monetization I considered charging users for individual quests as in-app purchases. Based on the survey, about 50% of families with kids visit museums, and about 10% of parents are likely to install the app. This results in an estimated 5.2 million users for the Kids Museum Hunt app.</p><p>If more than half of families visit museums over four times a year and each quest costs around €2, the total potential annual revenue could reach €40.5 million.</p><h3><strong>Competitors research</strong></h3><p>I also did some brief competitor research. The main competitors are:</p><ol><li>Hands-on objects in exhibitions.</li><li>Free printed quests available at the museum entrance, and <a href="https://kidsinmuseums.org.uk/2020/03/10-museum-activities-to-keep-children-entertained/">downloadable activity sheets</a> from museum websites.</li><li>Individual museum apps (e.g., British Museum) aimed mainly at adults, which include text articles or audio guides.</li></ol><p>I would mention separately Google Art and Culture app, which offers a versatile experience with art exploration, mini-games, and a rewards system.</p><p>There are also several museum-themed games with extended experiences. My favorite is MyMuseum, although it’s purely digital with no offline connection.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hXIU7AfN1KvoZqVRseepgw.png" /><figcaption><em>A few examples of the museum and art-related apps.</em></figcaption></figure><p>Based on the insights the advantages of the Kids Museum Hunt app:</p><ul><li>Connects offline and online experiences.</li><li>Offers collectible and shareable rewards.</li><li>Provides a familiar experience across multiple museums.</li><li>Uses gamification to maintain interest and motivate more visits.</li></ul><h3><strong>Key assumptions about museum experience challenges</strong></h3><p>My primary assumptions about potential customer challenges were as follows:</p><ul><li>Parents struggle to provide engaging educational experiences for their children, finding it difficult to keep them interested, which often makes educational leisure activities less appealing.</li><li>Kids often get bored during museum visits.</li><li>After their first visit, children may not want to return to museums.</li></ul><p>If these issues aren’t addressed, parents may opt for less educational and less healthy activities, such as watching TV or playing video games.</p><p>I assumed that early adopters of my product would be parents who are deeply invested in their children’s free time. These parents prioritize quality time and seek activities that bring both joy and discovery. They typically have children aged 4 to 12, are willing to buy gadgets for their kids, but maintain control over the apps and screen time.</p><h3><strong>Problem research. Major insights from the in-depth interviews.</strong></h3><p>As the next step in validating my idea, I conducted in-depth interviews with 12 parents of children aged 2 to 10. The goal was to explore their leisure activities, how they plan weekends, what their kids enjoy, and the challenges they face, including their experiences with museums. I also asked about the reflections on the last museum visit and the most positive case.</p><blockquote><em>The key insight was that museum visits come with many logistical challenges beyond entertainment and education.</em></blockquote><p>Parents must navigate issues like weather, crowds, and noise, which can spoil museum visits. Planning varies widely: some parents meticulously plan, others visit museums only while traveling, and some go spontaneously. However, no one visits more than one museum a day.</p><p>Most parents try to <strong>consider their children’s interests</strong> (e.g., space, dinosaurs) when choosing a museum. If a child isn’t engaged with the topic, it becomes difficult to keep their focus — 7 out of 13 respondents mentioned this.</p><p>Typically, parents <strong>don’t prepare</strong> in advance for the visit, they improvise storytelling as they go. They often rely on guides or annotations, which they then <strong>translate</strong> for their kids — this includes both <strong>simplifying complex scientific language</strong> and, in many cases, translating from a foreign language (when visiting museums abroad, which is common among the European audience). 6 out of 13 respondents highlighted this challenge.</p><p>Visiting museums <strong>with friends</strong> is beneficial, as it allows kids to entertain themselves while parents can focus on the exhibits.</p><p><strong>Lunch</strong> is a key thing for many parents, with 9 out of 13 respondents bringing food along.</p><p>A common issue is that kids sometimes try to <strong>damage exhibits</strong>, often unintentionally, so a lot of parental attention goes toward <strong>preventing</strong> this.</p><p>Some parents seek to <strong>enjoy the museum experience themselves</strong>, taking time to examine the exhibits properly. To do so, keeping the kids engaged and entertained is a priority.</p><p>On average, museum visits last between <strong>1.5 to 2 hours</strong>. Some parents take their kids to the same museum <strong>multiple times</strong>, children tend to find new things to explore on each visit, and their interests evolve as they grow.</p><p>During a museum visit, kids often need a <strong>break</strong>, and in such cases, watching a video installation or even a YouTube video can help.</p><p>There is generally <strong>no post-visit reflection</strong>, as very few kids talk about the experience unless asked to. However, some children do develop an interest in what they’ve seen and may later explore the topic further on their own.</p><h4>Here are the things that most kids find interesting in a museum:</h4><ol><li>Hands-on objects (6 out of 13)</li><li>Listening to audioguides</li></ol><p>3. Watching video installations</p><p>4. Drawing objects seen in the exhibits</p><p>5. Finding small objects and details</p><p>6. Identifying and counting animals (e.g., cats and dogs) in the exhibits (such as in paintings)</p><p>7. The variety of media: video, sound, interactive displays, and imagery</p><p>8. Making a discovery, especially by the end of the visit</p><p>9. Learning something unique that they can call their own</p><p>10. Examining the others on what they just have learnt</p><h4>Main challenges when visiting a museum with kids:</h4><p>1. Kids get <strong>tired</strong> while parents are still eager to explore the exhibits.</p><p>2. A <strong>lunch</strong> break is essential.</p><p>3. Kids may unintentionally <strong>damage</strong> objects in the exhibition.</p><p>4. In colder climates, the process of <strong>undressing</strong> before the visit can be exhausting for everyone. By the time the whole family is undressed everyone is already tired.</p><p>5. When kids are of different ages their <strong>interests may vary</strong>, and the museum may not engage both. In order to make a visit meaningful parents often need to familiarize themselves with the exhibit highlights in advance.</p><p>6. Museums often lack descriptions of objects in <strong>language</strong> that’s <strong>understandible</strong> for children.</p><p>7. Keeping the right balance when preparing for a visit can be difficult, as parents don’t want to be too pushy in terms of educational expectations.</p><h3><strong>Reflecting on the results and next steps</strong></h3><p>A great share of insight collected through research concerned not the museum experience itself, but rather the <strong>meta experience</strong>, like weather, logistics, snacks, kids age. However, there were unexpected insights, like <strong>translations challenges</strong> and answering questions on the exposition. I understood that when coming to educational experience in the museum parents are already exhausted by taking their kids to it (the trip, clothing, lunch etc.). It might be tricky to remember that you also have an app for the museum experience.</p><p>However, talking about the experience itself parents claimed that keeping kids entertained requires energy and some help here might be valuable. This proved my main assumption of the value of the app.</p><p>The hypothesis that kids are not motivated to visit museums did not prove because in fact it is a parent who makes the decision on how to spend a weekend. However, they considers kid’s interest too.</p><p>I decided not to proceed with building the idea of the app, however, it would be a shame to abandon it at all. This is why I pivoted it a little and established a <a href="https://larkinanastia.substack.com/">regular newsletter </a>with printable versions of museum quests. I still enjoy creating those out of pictures I make myself during my museum visits. The collection of quests covers my personal interests and may miss major world museums, however, the love I put in it still makes it valuable to those parents who would like to make their family museum visits more fun.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wIDsraBfmKOWwLaPadtFTQ.jpeg" /><figcaption><em>Quests around places of interest in Estonia, Denmark, Italy and Czech Republic.</em></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=401f64e20396" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Bootcamp Experience: Insights, Challenges, and Lessons Learned]]></title>
            <link>https://nastialarkina.medium.com/my-bootcamp-experience-insights-challenges-and-lessons-learned-46ce15acda78?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/46ce15acda78</guid>
            <category><![CDATA[bootcamp]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Wed, 26 Feb 2025 08:09:31 GMT</pubDate>
            <atom:updated>2025-02-26T08:09:31.553Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EgUIG7cxm_P4HcHuXeKIJw.png" /></figure><h3><strong>Unexpected Interest in Startup Life</strong></h3><p>A year ago, in the middle of my product manager career, something unexpected happened — I found myself interested in startup life. An idea sparked, and it seemed promising enough to explore it. I conducted a quick possible competitor research, estimated roughly the total addressable market, ran a survey through social media, and concluded there could be a solid product-market fit with potential.</p><p>I still don’t quite know where my energy and enthusiasm came from, but suddenly it occurred to me that I wanted to dive deep into the startup vibe, from pitches to funding rounds. In Finland Startup Sauna is the hub for startups, and the name alone is a fun reminder of where I am. Several accelerators are also connected to it, and I considered joining one could be a great way to expand my network and meet potential investors.</p><p>I’m writing this and still can’t quite believe it was real. Entrepreneurship was never on my radar — my usual sense of responsibility makes corporate work intense enough as it is. Leading a startup might even challenge my mental health, but there I was, excited to see where this journey takes me.</p><h3><strong>Bootcamp and Market Validation</strong></h3><p>Before joining the accelerator I was required to go through a bootcamp. I had a few weeks before the start date, so I spent that time filling out an extensive survey about my idea and the data supporting my confidence that it would work. This is where my market research really proved to be valuable. Based on my survey responses I was invited to the week-long bootcamp.</p><p>The purpose of the bootcamp is to ensure that wanna-be founders validate their ideas qualitatively and confirm a product-market fit. I assume the requirement was based on a lesson learned by the accelerator from past experiences where founders rushed into building solutions without verifying that there was user demand for them.</p><p>At the welcome meeting it was announced that the bootcamp would require participants to conduct as many interviews as possible to clarify the user problem. The suggested target was forty interviews. Every evening participants were to submit a form on the website summarizing the day’s insights. Based on these insights, the organizers would select a set number of startups (I don’t recall the exact number) to join the accelerator.</p><p>Additionally, the organizers created a Slack channel, though it proved to be less useful than expected. Unfortunately, there was no dedicated activity to help participants get acquainted at the start of the bootcamp. I did connect with a few others in the lobby, but that was the extent of it. The networking experience would have benefited greatly from some facilitated activities to help people get to know each other. Overall there were three offline events: the welcome meeting, the results announcement, and a final barbecue party.</p><h3><strong>Conducting Interviews and Managing Workload</strong></h3><p>Right after the welcome meeting I began recruiting participants for my research. Half of them came from a local women’s chat, which is a goldmine of insights on nearly any topic and also a supportive community. The other half were came through earlier social media survey where I had added an option for those interested to join further research and interviews.</p><p>Somehow over the following week I managed to conduct twelve interviews on my own. The findings were supposed to be processed the same day. Balancing this with my main job was not easy, it required more effort than I’d usually take on. Still, I learned a lot, and the insights I gathered deserve a separate article.</p><h3><strong>Reflecting on the Accelerator Experience</strong></h3><p>In the end my idea wasn’t selected for the accelerator program — though in hindsight, this may have been a good thing. By the end of that intense week I was so exhausted that even I started to question the brilliance of my own idea.</p><p>Although I didn’t commit to a startup based on this idea, I decided to still share its core value with those who might be interested. As a result I created a regular email newsletter featuring content that was initially intended to form the core of the app. Now I prepare content every two weeks as a personal project.</p><p>A total of 57 teams applied to the bootcamp, with 45 completing the research phase. However, I never received any direct feedback on my own submission, which leads me to suspect that the final decision was based more on the written pitch than on the research results.</p><p>One of the most popular ideas among the organizers was an “Airbnb for parking.” The concept allows users to rent out their parking space during work hours to those needing it in their neighborhood. This is a beautiful idea, however, it can be challenged in many ways. It will be interesting to see if a similar model emerges in Helsinki or Finland in the near future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=46ce15acda78" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Developing a Warm-Up Feature into Routine Session for Music Learning]]></title>
            <link>https://nastialarkina.medium.com/developing-a-warm-up-feature-into-routine-session-for-music-learning-04c4b2e62faf?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/04c4b2e62faf</guid>
            <category><![CDATA[music-education]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[music]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[online-music-education]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Thu, 28 Nov 2024 09:16:59 GMT</pubDate>
            <atom:updated>2024-11-28T09:16:59.262Z</atom:updated>
            <content:encoded><![CDATA[<p>This is a story of the approach to resolving one of major user churn problem in learning musical instruments using Yousician, a SaaS music education app. Yousician provides a structured learning path with songs and melodies to play along with a backing track. The app detects sound and gives instant feedback on the user’s performance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yu7dwWmL7JV3jv6_PqV4vg.png" /><figcaption><em>A bouncing ball highlights each upcoming chord. When played accurately, the chords turn green.</em></figcaption></figure><h3><strong>Churn Insights as an Initial Incentive to Start the Project</strong></h3><p>The extensive churn research conducted by the Yousician team last year revealed that what appears as churn in playing a musical instrument is more accurately a long break. People become so passionate about music that they rarely quit entirely. When they start a music hobby, they often become deeply engaged, playing for several hours a day. The data shows that in the first weeks of using the app, sessions are long and frequent. However, life events often cause people to pause their hobby, but they seldom abandon it permanently.</p><blockquote><em>When speaking with users who canceled their Yousician subscription, we discovered that most of them still hope to return to playing guitar someday.</em></blockquote><p>The primary reasons for churn are:</p><p>1. Life circumstances such as family or job changes, relocation, etc.</p><p>2. Size of the Yousician’s song library. Users, after acquiring basic skills, seek specific songs they want to play, however, these might not be available.</p><p>3. Difficulty spikes in the learning path. At some point, songs and tasks in the learning path become too challenging.</p><p>4. Challenges in resuming after a break. Once a user takes a break, it’s hard to reestablish routines and form a habit.</p><p>Surprisingly, about 90% of our users have taken a break at some point. So, we began considering ways to make returning to their hobby easier and more rewarding. One common fear among users returning after a break is the loss of skills and the inability to play anything. We hypothesized that starting with something easy and familiar would help set users up for success. This led us to develop the idea of a warm-up.</p><h3><strong>First Experiments with Warm-ups</strong></h3><p>The first iteration of our warm-up feature included a tuning experience, a chord trainer, and a song the user had previously played. We developed a complex logic for selecting a song from several sources, such as the user’s favorites, successfully played songs, familiar tunes, and songs suitable for their skill level, following a strict priority order.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iNv4Z3dobcxFp8RMurhrDA.png" /><figcaption><em>The warm-up pop up was the first thing a user saw when starting a session.</em></figcaption></figure><p>As warm-ups were designed to address the churn problem, our primary metric for success was next-week play retention. This metric indicates how many users who experienced a warm-up returned to play again the following week. In our initial experiment, we saw a 3.9% increase in this metric. Two-thirds of our users started a warm-up, and about 90% of them completed it successfully. However, we received numerous complaints both on social media and internally about users receiving the same song repeatedly for their warm-up.</p><p>Based on this feedback, we simplified the song selection logic by removing the priority order and introducing a randomization factor. The subsequent experiment proved more successful than the first, with a 2 percentage point increase in users starting a warm-up. Additionally, users were more likely to begin the second and third warm-ups, indicating that they were not getting bored with the content and were developing a habit of starting their learning sessions with a warm-up.</p><h3><strong>Extending Warm-ups to Daily sessions</strong></h3><p>Regarding success of warm-ups, we decided to double down on the idea of predefined content and extend it to the whole session, applying it not only to users returning after a break, but to all app users. Our previous research and usage data of legacy functionality indicated that new users appreciate guidance and prefer not having to make choices or navigate content on their own.</p><p>Therefore, we decided to add a learning segment following the warm-up. For this segment, we selected the next task in the learning path after the user’s last completed task.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dWZT3iHsMV_IwUyqQ4redQ.png" /><figcaption><em>The daily session inherits the tuner and the warming up experience from the initial experiment, while introducing the learning component as a fresh addition.</em></figcaption></figure><p>It is important to mention that the learning path includes various task types:</p><ol><li>Videos</li><li>Trainers</li><li>Songs</li></ol><p>The engineering team developed a smart system that examined the task type and searched for the song. If a song was preceded by a video, a trainer, or both, these tasks were linked and presented to the user consecutively.</p><p>This approach allowed users to smoothly progress in their learning journey.</p><p>Following the launch of this feature, we received considerable feedback on social media, highlighting two major concerns:</p><p>1. The logic did not take into consideration sudden jumps in the learning journey. For instance, if a user typically plays at a moderate difficulty level but occasionally views a video from an advanced level, the daily session logic presents them with tasks from that advanced level, which may be too challenging. Although initially considered to be a corner case, this issue turned out to be more frequent than anticipated, so we needed to allocate some technical capacity to address it.</p><p>2. Another significant feedback concerned the warm-up content. With the addition of the learning component, the entire session turned out to be actually two songs played one after another. To tackle the problem, we revisited the idea of warming up with a chord trainer rather than a song. However, this approach created a limitation: the daily session could only be displayed to users already familiar with playing chords, which are introduced at the second level of the learning path. Thus users at earlier levels would get a daily session lacking a proper warm-up. Catering this segment was left out of scope of the next iteration.</p><h3><strong>Enhancing User Experience Through Visibility and Control</strong></h3><p>User testing conducted after the MVP launch revealed more user need which was not covered. While users appreciated the predefined session, they felt a lack of visibility into the session’s content and structure. To address the problem, we introduced transitional overview screens between the major parts of the session.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dXAowQ82tK66Gue8AFnprQ.jpeg" /><figcaption><em>The overview screens were introduced before each meaningful part of the Daily session.</em></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XbT67QeOC7j7jJDUBh_EPg.png" /></figure><p>Another issue identified during user testing was the lack of control over the session. To address this, we we added more diversity to the content in the Learn section. In addition to selecting the next task in the learning path, we included easier and more challenging options. We also introduced a shuffle button in the user interface, allowing users to choose between these three options.</p><h3>Outcomes and Future Directions</h3><p>Despite receiving extensive feedback and identifying potential improvements, consecutive experiments with warm-ups and daily sessions demonstrated an increase in our target retention metric — next-week play retention. This positive trend gave us the confidence to further develop this framework and adapt it to other types of content.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=04c4b2e62faf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Navigating Language Challenges in My International Work Experience]]></title>
            <link>https://nastialarkina.medium.com/navigating-language-challenges-in-my-international-work-experience-e417ade214c3?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/e417ade214c3</guid>
            <category><![CDATA[language-learning]]></category>
            <category><![CDATA[language]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[english]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Wed, 02 Oct 2024 13:32:56 GMT</pubDate>
            <atom:updated>2024-10-02T13:32:56.054Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ze7BAh8ZQYozibJrUgr8NA.png" /></figure><p>I have been using English as my main professional language for two years since moving to Finland and joining Yousician. Back in the day I graduated from the school specializing in advanced English but had little real-life practice. I scored ridiculously high on the IELTS Academic exam, qualifying me to study British literature in the UK, but real-life experience turned out been different.</p><h3><strong>Major language challenges</strong></h3><p>I didn’t expect language to be a major challenge at my first international workplace, but I did came across some difficulties.</p><ol><li><strong>Limited vocabulary.<br></strong>At Yousician we have an incredibly international team with people from all over the world. Instead of adapting to one culture, I incorporate elements from many. We use a blend of international English for daily communication.<br>In the first months it seemed to me like everyone used only about 300 words. I closely followed native speakers and realized they had simplified their language to <em>international English</em> for work. Yet our discussions were still complex and I needed a wider range of synonyms to be both persuasive and polite. During that time I really struggled with vocabulary.</li><li><strong>Balancing persuasion and politeness.<br></strong>However, my biggest language challenge has been balancing pushiness and politeness. As a product manager, I often need to say “No” to some things and “Let’s speed it up” to others. Culturally, it’s been tricky for me to evaluate when to push harder and when to hold back.</li><li><strong>Switching languages.<br></strong>At Yousician about a quarter of my colleagues are Russian (like me). While English is the official language much one-on-one communication still happens in native language. This constant language switching is mentally exhausting. To make it easier I decided early on to keep all my work-related notes in English, which has helped a lot.</li></ol><h3><strong>New English words and phrases from my first two years at work</strong></h3><p>Browsing through my personal notes from the past two years, I found about 50 work-related words, phrases, and idioms that were new to me and used on a daily basis. Here’s a refined and shortened list of them.</p><p>🤷 <strong>Hit the wall</strong>. Face overwhelming difficulties. This applies to app users who are so frustrated they stop using it.</p><p>🙋‍♀️ <strong>Have a say</strong>. To be part of decision making process.</p><p><strong>Rule of thumb</strong>. A practical method based on experience rather than theory.</p><p>🙋🏻 <strong>To bring (something) to the table</strong>. Refers to a participant’s initial contribution</p><p>🦾 <strong>To leverage</strong>. To use something to its maximum potential.</p><p>🔮 <strong>Opportunity discovery.</strong> A fancy term for testing a brilliant idea without data.</p><p>✊ <strong>Confident</strong>. Being sure about something, e.g., “I am confident this idea will work.”</p><p>👣 <strong>Baby steps</strong>. Moving in small steps which are even smaller than <em>iterations</em>.</p><p>⚖️ <strong>Use your good judgment</strong>. When you need to decide with no or too much data.</p><p>🤷🏼‍♀️ <strong>Educated guess and guestimation</strong>. When you make a decision with limited or unreliable data.</p><p>🕵🏻 <strong>Look into something</strong>. To investigate or examine a matter more closely to understand it better or to find more information. It often implies checking the details or exploring the issue further.</p><p>🦹‍♀️ <strong>Talent</strong>. This word replaces <em>specialist</em> or <em>professional</em> often referring to musicians, actors or models, such as when hiring a model for marketing.</p><p>🌰 <strong>In a nutshell</strong>. Briefly, in a few words.</p><p>⚫️⚫️ <strong>Connect the dots</strong>. A good phrase for getting everyone to start collaborating.</p><p>🪨 <strong>Set in stone</strong>. In Russia we used to say<em> cast in bronze</em>. I was pleasantly surprised to find a nearly identical, commonly used phrase here.</p><p>🔎 <strong>Surface functionality</strong>. The phrase is used when a feature is buried in the interface and needs more visibility or onboarding.</p><p>⬆️ <strong>Bring [this] up</strong>. The phrase means to raise an issue, usually in a large meeting.</p><p>🧂 <strong>Take it with a grain of salt</strong>. Usually applies to research results, indicating they may be skewed by factors like sample size or conditions.</p><p>🏕 <strong>Work in silos</strong>. Working alone, not sharing (interim) results.</p><p>💪 <strong>Double down</strong>. To put in maximum effort, literally meaning to double the effort.</p><p>✔️ <strong>This works for me</strong>. It suits me. Previously I would blame myself with “I can’t make it,” but now I say “This doesn’t suit me” sharing the responsibility.</p><p>🌞 <strong>To sunset (a feature/product)</strong>. The term refers to phasing out a rarely used feature gradually to avoid upsetting users. In Yousician, jokes often arise about the endless Finnish summer sunset, where the sun sets briefly at midnight.</p><p>🧠 In conversations, “<strong>thinking</strong>” often means <em>reasoning</em> or <em>justification</em>. For example, “What is the thinking behind this proposal?”</p><p>🔨 <strong>Hit the nail</strong>. It means “reach the target,” used to say a feature met its metric. 🪡 Similarly <strong>move the needle</strong> is used the same way.</p><p>🚀 <strong>Take (the project) forward</strong>. Means to advance a project or task, unlike <em>drive</em> or <em>push</em> which we used as borrowed terms in Russia.</p><p>➡️ <strong>Nudge</strong>. To push softly. The word replaces <em>push</em> in my usage.</p><p>🧩 <strong>Be part of the project</strong>. To participate in a project. Instead of <em>take part</em>, <em>be part</em> feels more engaging.</p><p>💭 Instead of <strong>gut feeling</strong> which was common in Russia people here refer to <em>intuition</em>.</p><p>💩 Shit hit the fan. Finally, a direct translation. In Russia, we’d say someone is “throwing shit at the fan,” and <em>throwing</em> alone specified the situation.</p><h3><strong>Expressions I miss</strong></h3><p>I do miss some Russian phrases commonly used in software development. I regret I can’t translate them literally for daily use. One favorite is <em>to develop a feature from shit and sticks</em> meaning the team lacks time for proper architecture and must build quickly. The closest English idioms are <em>cut corners</em> or <em>hacky implementation</em> though they are less expressive.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e417ade214c3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Guidance At The Beginning of the Learning Journey]]></title>
            <link>https://nastialarkina.medium.com/guidance-at-the-beginning-of-the-learning-journey-540b9a384168?source=rss-28bc0334c30a------2</link>
            <guid isPermaLink="false">https://medium.com/p/540b9a384168</guid>
            <category><![CDATA[music]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[education]]></category>
            <dc:creator><![CDATA[Nastia Larkin]]></dc:creator>
            <pubDate>Thu, 24 Aug 2023 04:12:14 GMT</pubDate>
            <atom:updated>2023-08-24T04:12:14.348Z</atom:updated>
            <content:encoded><![CDATA[<p><a href="https://yousician.com/">Yousician</a> is a SAAS music education app. It has an exclusively designed learning path that covers basic playing skills and techniques. Besides there is a library of popular songs by various artists. When a user plays a song the app listens to the playing and gives immediate feedback on each note played.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F01VlR0JY6zU%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D01VlR0JY6zU&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F01VlR0JY6zU%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/7f988586eb8fec9393ca2d6f5b40d68c/href">https://medium.com/media/7f988586eb8fec9393ca2d6f5b40d68c/href</a></iframe><p>Being a product manager of the core experience of the app I learnt a lot about the guidance people need while making their first steps to musicality.</p><p>The key value of the app is playing a song. As a user plays a song successfully, he gets inspired and highly motivated to proceed with the learning journey and start a subscription. Conversion to subscription is one of our key monetisation metrics. First song play is a proxy product metric to it. Thus playing a song within first time user experience is crucial.</p><p>In this article I will describe three cases of how we tried to make the song play easier and more enjoyable for the first time.</p><h3><strong>What is the value of guidance in the learning journey?</strong></h3><p>Most of research conducted throughout the whole period of Yousician app existence shows that people value guidance a lot. Alternatives to Yousician app are Youtube teachers and chords googling. As both are free many users start their journey from those points. However very quickly they realise that in addition to watching videos they need to do meta work of systemising their knowledge and building a meaningful sequence of what should be learnt in order to start playing. This is exactly the job users hire Yousician app for.</p><p>But how directive should that guidance be? This was one of the main questions for Yousician first time user experience design.</p><p>The app itself consists of two main parts — the educational syllabus and the songs catalog. The syllabus is the sequence of videos, trainers and songs organised according to skill techniques and complexity — from easy ones to difficult. Is is designed by professional musicians and music educators. The songs catalog is the collection of licensed music prepared to be played: with a leading melody and a backing track.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6fonQI83hepfhwHOIfPepg.png" /><figcaption>Songs catalog and learning path.</figcaption></figure><h3><strong>Too Much Freedom Distracts Users From Core Value</strong></h3><p>According to the research big part of our users install the app having in mind some particular song they would like to learn to play. Based on this insight we suggested that songs might be a good selling point when converting a user from a free trial to subscription.</p><p>To prove this hypothesis we designed a new version of the fist time user experience.</p><p>In the current version after a sign up a user was exposed to the autoplayed introduction video which was followed by the tuning experience. Only after that he was landed in the learning path where he could play his first song.</p><p>In the new version we landed a user in the learning path right away after the sign up.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j-6HcRbHiZefI38lRWMUug.jpeg" /><figcaption>More freedom.</figcaption></figure><p>We hypothesised that after examining the app and the songs catalog in particular a user might be more motivated to start a trial. However, the experiment failed. A users opened the songs catalog where all the songs were locked until they start a free trial. After starting a free trial users started to play songs and failed as being beginners they lacked proper skills to succeed from the get-go. This is why trials poorly converted to subscriptions. Another thing is that users started wondering around the app, they opened the profile, the tuner, weekly challenges section and failed to find the starting point of the learning path where the main value was.</p><p>The conversion from sign up to playing a song dropped by 11%, but subscriptions remained the same. Having this in mind we decided to abandon the idea of giving more freedom to a user and thus distracting him from the core value of the first time experience.</p><h3><strong>Too Much Guidance Does Not Do Good Either</strong></h3><p>Taking into consideration the results of the previous failed experiment we designed a new more directive flow. This time after the sign up a user was exposed to the autoplayed introduction video. Then the app took a user directly to the tuner. Afterwards the first song was autoplayed. And finally a user was landed in the syllabus.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O3uU2rdF4U5TD7ZXlM4UoQ.jpeg" /><figcaption><em>This time we designed a more directive experience.</em></figcaption></figure><p>This experiment failed as well. Eventually more users played the first song, however, they did not tend to move forward along the learning path. Thus we needed to find a good balance between guidance and discovery freedom for the first time experience.</p><h3><strong>Reasonable Guidance Balance Leads to Successful Playing Experience</strong></h3><p>After two unsuccessful experiments we returned to out regular app experience where a user was landed in the learning path after an introduction video and tuning.</p><p>In order to exclude the situation where a user fails to play difficult songs without gaining necessary skills we blocked tabs with Songs catalog and Weekly Challenges. By tapping on either of these there appeared a tooltip asking to finish the first learning level (we call it The Basics) in order to unlock this content.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p6vJIAyaOBvqfFDQ-wA1sA.png" /><figcaption>A user is nudged to get essential skills before exploring difficult songs.</figcaption></figure><p>This experiment showed a clear increase in all low-level product metrics. 5% more users played the first song and did it successfully. 12% more users finished the Basics. Less users started free trials, however, the conversion from free trials to subscriptions grew, so the players we gained were more motivated.</p><p>Thus we found an acceptable balance of guiding users towards successful play.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=540b9a384168" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>