Search Results

Everyone wants to build with AI but it's much harder to find an article from the perspective of an AI researcher, who experiences it from the other side. So, I invited Joy to write this article. He's also my brother :) Joy's experience spans both large-scale initiatives and resourceful projects on tight budgets. He's done a few cool things: Worked with the Government of India's Meteorological as well as the Agricultural department to build models for them. Was awarded a research grant of $2.8M from Google to invest in his AI research. He's also worked with me at Streamline. If you're looking to build with AI for your SaaS/tech org, reach out to him on Linkedin or via this email joylunkad@gmail.com . This blog is written to help anyone who wants to build something using AI but cannot figure out where exactly. It helps you build a strong intuition about what AI can do, especially for you, so you know how to find problems that can be solved and have a preliminary idea about how expensive it might be to build it. So let's dive right into it. What can AI do? Generally, I want you to think it can AT LEAST do whatever humans can do, given ENOUGH data and compute. Any process that has the following structure and satisfies the following constraints has an extremely high chance of being solvable by AI - Input → Some Intelligent Processing → Output Where, 1. There is readily available input and output data, or it can be procured. 2. The input contains all the necessary information for a human to generate an accurate output. There is no need to include the person's skilled prior knowledge or general knowledge in the input. And, If prior knowledge is needed to solve the problem, then you should be able to formulate it as another solvable problem, ie, follows that structure and also satisfies the constraints. Are there any sharp bits that I should keep in mind? These being the only constraints just goes to show how many problems in the world are just waiting to be solved. But just because a problem is solvable, doesn’t mean it is feasible to build a solution. Gathering enough data and compute can be extremely expensive. In general, the lesser the prior knowledge that needs to be built in, the cheaper and easier it gets. And if a problem requires a ton of prior knowledge but someone else has already incorporated it into another model, the costs to train an AI become significantly cheaper. More data and more compute almost always translate into better results and similarly, if insufficient, the AI will be brittle, and make unpredictable and hilariously stupid mistakes. You can also try breaking down a big unsolvable problem into multiple small solvable problems, chaining their inputs and outputs in such a way that their ends look like the ends of the original big problem. You can do this in cases where you can't find enough data for the big problem. The benefit of solving a big problem without breaking it apart is improved performance on the final task. This is usually a very big boost in performance comparatively. The benefit of chaining smaller problems when you could solve a big problem is that it usually significantly reduces the expenses. What do I do after identifying a problem that fits in that structure and satisfies the constraints? I recommend that you have a simple discussion about it with an AI expert. There are many datasets and models out there, either open-sourced or behind an API, that can be used to bypass the expensive steps. This talk will help you get a crude idea about how feasible it actually is, and more often than not, it would be much cheaper and faster to build than you originally imagined and at the same time, more novel/revolutionary. I could go into detail about patterns and nuances to look out for that might help you identify and differentiate between ideas that are genuinely hard/expensive and low-hanging fruits but that might make this a very long and exhausting read with diminishing returns, and even then I wouldn’t be able to make an exhaustive list. You now know everything you need to know to identify solvable problems in AI. The blog ends here but if you want to develop some pattern recognition and clear up any misconceptions, I wrote examples of some problems being solved by AI. I highly recommend going through them. Example 1: Art Generation (Eg: Midjourney) I am using this as an example as it's not exactly obvious how it fits into the structure and satisfies the constraints. So, this example shows how far those rules can be stretched, helping you find opportunities where it looks like there aren’t any. If we rewrite the problem, we might get a clue. How about “creating realistic images and art from a description”? This allows us to put the problem into the structure → 1st constraint - For the input & output data, we could use the billions of images scattered across the internet, and we could use their captions as their descriptions. 2nd constraint - For an artist, a text-based description is enough to create a related piece of art. We can now be sure that it is possible to build such an AI tool. These captions are not always the description of the image, and also a single image can be described in many ways. Thus, our data is not perfect. Also, our model needs to learn a lot of prior knowledge about people, animals, objects, etc as well as some understanding about the world. We now know that these issues will make it extremely expensive to build it. Example 2: Automatically Detecting Fractures from X-Rays This is a pretty simple example, but it might clear up any misconceptions. Prior knowledge is only required if it is completely absent from the input and output data completely. If it is implicitly present, ie, it can be learnt from the relationship between the input and the output data, then we don't need additional steps. X-rays (input) → A doctor examining the X-ray → Report (output) 1st constraint - Hospitals should have huge and high-quality databases containing X-ray images and their reports. 2nd constraint - X-rays, maybe MRIs in really hard cases, are all a doctor needs. As for prior knowledge, the doctor only needs the knowledge of the skeletal structure which is available to the AI in the input data. Now that we know it is possible to solve it, and it will be relatively easy. Example 3: Self-Driving Cars Vision from the car's cameras (input) → A driver → some driving decision (output) 1st constraint - For the input, we require extensive sensory data from car cameras. As for the output, correct driving decisions can be collected from the steering wheel, accelerator, brake, etc. This data doesn't exist naturally, necessitating the preemptive installation of cameras and sensors on cars. Additionally, we must persuade customers to purchase these equipped cars to gather the necessary data. This poses a significant challenge, but it is achievable. 2nd constraint - For a human, it's just decent motor skills and some practice, hence it is self-contained, ie, just cameras are enough. Now that we know it is possible to solve it. Let's figure out how expensive it might be. This looks easy, but a sharp bit here is that it is very difficult for AI bots to do things in the real world. The input data must be all-encompassing, capturing every aspect of driving, from road conditions and weather effects to the unpredictability of human behaviour. The greater the amount of implicit prior knowledge hidden within the input data, the more extensive and varied the data required. Although the problem remains solvable, it's harder than it looks. Example 4: Facial Recognition for Security Current and verified images of the target’s face (input) → A person checks if it's the same person → Approves Entry (output) 1st constraint - Companies like Google, Meta, and Apple can ask users to tag themselves and others in images. Using this, you can create pairs of current and verified images. 2nd constraint - Self-contained, and no prior knowledge is needed. Now that we know it is possible to solve it, and it will be relatively easy. Thanks for reading! Joy Lunkad Linkedin | Email: joylunkad@gmail.com

Hi, I'm Khushi! And this is how I use Reforge AI for work. Warming up the AI leads to better results If you wish to ask questions like: "What are the types of monetization strategies a company like Streamline can use?" Try asking something like this first: "What are some monetization challenges Streamline could face?" If your goal is to find solutions, have Reforge AI align on problems first. Ask for tactical advice to move beyond the strategy AI can end up being too strategic at times. You might want to probe it for more tactical tips. "Streamline already offers a generous free tier. How can I prevent it from cannibalizing sales for the premium product? List some ideas and pros and cons of each. If you can pull examples from Reforge's curriculum or the live case study videos, that'd be great!" Another prompt that I use at times, is this: "Can you give me some tactical ideas of what I can try and test out? Less strategy. More actionable inspiration." Other prompts ideas Every prompt must go through a warmup period. Otherwise, the responses are too strategic and obvious. For OKRs "I'm tasked to write OKRs to own the marketing and PLG department for a sales-led business......some more back and forth..... What should they be?" For Benchmarks: I prefer to use benchmarks from Reforge vs use Google. "What are the benchmarks for conversion rate for an ungated product with a free trial and a free product?" "What factors influence conversion rates most?" For Roadmap Planning: "What are some projects growth engineers on my team should work on next? I'm trying to build our roadmap? Please share some ideas for projects." "Great, write more." Dealing with strange results If Reforge doesn't have relevant content for that topic, then it isn't very helpful. For stuff you can google, I find it helpful to use ChatGPT/Claude more than Reforge. Generic questions I asked lead to average responses. What is buyer journey mapping and why is it important? What are some questions to ask during a buyer journey process? Full coverage content For content that Reforge has in its curriculum, the replies are very thorough. You'll start seeing the result of it in your questions. I find talking about adjacent topics helpful. Example 1: How to design cancellation flows? How to analyze cancellation metrics? Are there any metrics for cancellation flows? List the metrics and the formulas . Example 2: "I have an infrequent product where users set it once and forget about it while it continues to offer ROI in the background. What are some challenges products like this can face?" " How can I add more frequent use cases for a product like Churnkey? Churnkey is a retention automation platform for high-volume subscription businesses. On average our customers retain 20-to-40% of subscription revenue they would have otherwise lost to churn. It reduce voluntary customer cancellations as well as involuntary churn due to credit card processing failures." Results Client 1: "We’ve been very impressed with what you accomplished in just one month!" -> Offer extended Made an 80 slide long PLG deck, which was in depth with zero fluff. And extremely tailored to their use case. Client 2: "You've really nailed our product" Client 3: Is pretty happy too! Got a promotion! .. and so on! Final thoughts Previously, I left jobs to take up the courses. Whatever Reforge said, it took me 10x longer to apply. I had anxiety and fomo for not being able to complete as many courses as I wanted to. I honestly like the feature to bridge gaps. I would bother the EIRs and email them for help. Now I don’t do that, thank god. Thanks, Khushi

The cheapest way to learn in SaaS is probably to buy the product and see what they're doing. Dropbox Revenue From a numbers standpoint, Dropbox has 700M users, of which are 18.16M are paying  customers (around 2.5% of all users). 2.5% is a high enough number for a PLG company. They grew revenue by 7% (~180M) in 2023. The first quarter of 2024 saw slower growth in terms of revenue though. To learn from Dropbox, all I needed to do was pay $30 and see everything I could as without having to join the team. What a steal. I gathered 130+ screenshots across their entire app, and labelled a lot of stuff 🥵 The company I work for has a paid plan with Dropbox so I know that creatives are their most important target audience. Creatives have a few challenges. They work with very heavy files and must keep them organized. They also work with clients and have an approval process. Every product that Dropbox built was about virality. Choosing creatives as an ICP also has virality baked in if you think about it. More on that to come in a different blog post. Homepage The homepage is focused on conversions. You see ‘Find your Plan” at the top. The secondary CTA is sign up for free. The fifth fold prioritizes work use case over personal in the visual hierarchy. Forces people in the Systems 1 thinking by giving two options instead of just one. The footer also focuses on the "Get your plan" CTA Idea💡 Direct people to the pricing page from the homepage where they can get a free trial Activation The first modal window I see in the onboarding asks me to upload a file and share it. When I try to share it, I see a paywall. This is brilliant! First, the feature is optimized for shareability. Dropbox stands to gain from this. Second, they show me a nice feature (password to a file). The share button is disabled. The feature is highlighted in blue to draw attention. If I didn't want to pay I could click on the "Create and copy link" text, and it would still work. Idea💡 The first feature you introduce to can lead them to a shareable feature. When you introduce a paywall, you might want to let people opt-in for a trial. Buying a plan When I click on "Buy essentials", I'm not taken to the pricing page but rather led to their plan directly. Yearly is selected by default for obvious reasons. The one odd thing is that I can't start a trial. It was available on the public facing pricing page. I signed up with multiple accounts. So, here's another thing that Dropbox is testing. Dropbox will also ask you in the onboarding itself which plan would you like to choose. You can see that the free plan is almost hidden below the fold. They also offer to remind me when I have a few days left before my trial (but I didn't really read that copy). Different layouts for the same plan I installed the desktop app and haven't upgraded yet. On the final screen of the onboarding, it tries to sell me into the "Plus" plan. Again, it doesn't mention I could start a free trial on the CTA. Anyway, the imagery is great! It compares the free vs the paid plan. Although, 2 GB on the free plan is 0.1% the size of the paid plan limit. Not 33% that this visual demonstrates. Tip 💡 Compare the free plan vs paid plan visually. Might need an art director. Now, this is interesting. You saw earlier with the Essentials plan, I never saw Dropbox Plus. But now I do. Plus comes with no free trial, since it's a cheaper plan. I also got a little confused with their naming. How was Essentials more expensive than Plus? All prices uses 1 and 9 so it got even more confusing. There are a host of clever things Dropbox does Showing how close am I to the plan limit (Using 39KB of 2TB) with a progress bar No free trial vs free trial Show both monthly and yearly plans at once, and calculated in both frequencies Look at the differences, even in the features shown on the side. Dropbox lets you bypass their free trial. It's a good anchor to the free trial. You can request an invoice too. Manages expectation that it needs to be a larger order (team of 15+) They offer multiple payment providers (although some may not work well). Dropbox doesn't let me abandon the purchase flow. Once I clicked on the upgrade option via the onboarding flow in the desktop app, it doesn't let me simply abandon it. Lifecycle Emails Dropbox emails are very inspiring. They have the perfect email previews, use lots of liquid and HTML, is super on-brand, ties every email to their core action metric, and use proper sub-domains for sending so it never hurts deliverability. There's a lot to like! Email 1 Email 2 Upsells Dropbox does a great job at upselling me into a plan. In the UI, you'll see "Invite Team Members" plastered at nearly three different places. If you were to click on it, you'd see this modal that asks you to upgrade into Dropbox Business. So, essentially I was upsold while I was in the middle of a trial. And I can't purchase just one more seat, so I need to purchase a minimum of three seats. Once you upgrade to business use cases, the invite flow is also incredible. Your team members are assigned tasks by you. Two are selected by default. One of the tasks is to upload their personal folder. The more team members upload their personal folder, the more likely will you hit those storage limits if any. Another cool thing Dropbox does, it it will ask all your team members to invite more team members in the onboarding itself. They customize the reasons for inviting differently for team members vs admin. Admin might care about security but team members might care about other things. I'll talk more about their invite flow in another blog post. It's extensive! If I were to get a trial on the business tier, it wouldn't give the bare minimum of 3 licenses but rather give me five. Feature Discoverability A big reason why people pay is how deeply you are integrated with their workflow. Many users might churn stating that they left because a feature wasn't available. When, you probably had that feature and they couldn't discover it. The beautiful thing about software is that people pay when they receive value and stop paying when they don't. Feature discoverability plays a strong role in making the first part of it happen. The coolest thing that Dropbox does is that each product has their own tutorial/onboarding. Dropbox's Replay had the best onboarding of all times. My mind was blown. You can watch it here . Can't embed it unfortunately, no idea why not. They also added a PDF file because one of the features was to "Send and Track" files. This PDF would come in handy as I discovered another feature of theirs, hidden under their most used button. My first thought when I saw this was, "Huh!? You can edit PDFs these days? Wow". This is probably so useful. The thing about feature discoverability is that people might not even know what they need until you show it to them. And Dropbox introduced me to 10s of features in a matter of a few minutes without being intrusive. Such as this one: "you can convert file formats with Dropbox!?" Sometimes, it would trigger the good old modal window when I opened up a new product. Other times, it would highlight the button enticing you to click on it. When I uploaded a file, it automatically opens up in the sidebar and the first thing I saw was this share button. And it doesn't stop there. Dropdown gives you reasons why you might want to share. In-product onboarding is more natural. They open up a feature I haven't yet explored. If they detect that I have explored a feature, it won't open up. This button on the main dashboard keeps changing based on which feature I might have not explored yet. When to monetize They monetize the number of templates, and it is not just 1-2 but a healthy limit of 15. Perhaps, it triggers a sales conversation. A niche little feature that lets you compare versions, is an add-on. Btw, the upsell you see in a feature disappears after the first click. So, they don't continuously spam the user. And also open up room for few features. Contact sales is not visible on the cheapest tier. I saw a popup only when I hit the business tier. Failed Payments When you login, this is the modal window it’ll show to you. You see this banner with a countdown at all times. Plus a "Add billing details" as a CTA that replaces the upgrade button. Thanks for reading, I'm going to improve this article in the next couple of months. There's a lot to learn!

There are three reasons why people share 1. Financial: Robinhood, Paypal (give get $10) 2. Personal: Whatsapp, Slack, Linkedin 3. Social/Word of Mouth: Tesla, Tinder, Stripe And, our new product wasn't relevant to our existing audience We were launching a great product. It was free and open-source. However, the product itself wasn't something our existing audience could use. This means, I couldn't piggyback off of our existing audience to get that first spike. Slapping money on the wall Given the circumstances, it would be impossible to get people to share the product organically. They had no skin in the game to share. So, the only option left to us was a financial incentive. Pay people to share. I'd say it was a weak financial incentive. A giveaway. Only 1 winner. So people know they might not win anything even after putting in the effort. A give and get program is much stronger in comparison because there's an assured return. Ours was worth $300 which was what many people in our audience earned in an hour. I was worried it'll fail So a couple of hours before the launch, I pitched the team to pivot away from the financial incentive. 🥲 The solution? The alternative I suggested looked something like this 👇 Instead of asking people to share with friends, I wanted to encourage people to share with influencers instead. This was the exact email I drafted: Without tracking referral links If I were cooler, I'd say that attribution is a myth. That's why I did away with unique URLs allowing people to track their referrals. But if I were honest, I just didn't have the time to set up something more polished. Plus, the email tool we use doesn't integrate with Sparkloop which is what I would've wanted to use. The campaign was a total success To be honest, I was impressed how well it performed. We've been featured in languages I don't speak. And the quality of posts are much better than had we paid for it. Here's why: The influencer gets a warm intro from an existing fan. The mention looks more authentic than a sponsored post. And the user doesn't have to create content on their own to share. All they have to do is forward an email to one person. Interestingly enough, we got featured by influencers that never do sponsorships. I know this because I did reach out to them right after they posted. This was the exact moment when I realized the campaign was a success. Influencers that took no sponsors promoted us for free. And I didn't even have to reach out to them. It worked at scale. Across different languages. Targeting both macro and micro influencers. Did it convert? It absolutely led to bottom line revenue. I have a form on the post-purchase flow sequence that asks people where they discovered us from. And a few people referenced these influencers. How virality works? I read an article on Substack that explained how ideas go viral. And you'd be surprised to learn that ideas don't go viral when one person shares with another. In fact, they go viral when someone shares with a lot of people and it explodes. So, if a tweet goes viral, more often than not there's someone famous that retweeted it. I used that as an insight to convert a referral marketing campaign to an influencer marketing campaign. Because even if 10 people shared it with an influencer, we would've gone viral. Final notes I'd recommend you try this out! It was a bit creative and I haven't seen anyone else do it. So, it's not overly used yet. Plus, I can't take full credit for this. I have a wonderful team. Supportive founder. Great product to market. Good luck and god's support. So, thank you so much for reading this long letter.

Tbh, this is an unsponsored appreciation post for Mixpanel. Insights that stitch marketing with product data were historically impossible to do so for us, and I now have data that I only dreamt of having. Goal: Un-silo marketing and product analytics Traditional growth marketing courses recommend using tools like GA and GTM. However, these tools keep data siloed. They lack detailed insights into user engagement, activation, monetization, and retention. The solution is to unify marketing and product data in one place. It'll save time and resources from spending money on the wrong channels. However, historically, this was challenging for small teams to accomplish. The starting point: Mixpanel vs GA4 When I first joined Streamline, our analytics were almost non-existent, resembling those of any new SaaS company. We tried multiple software tools, including Google Analytics, Matomo, and Plausible. Despite spending a significant amount on consultants, the data we got was inadequate. Our developers were particularly hesitant about diving into Google Tag Manager and marketing tracking, a common sentiment because of their unintuitive product. The most perplexing issue was that most conversions were showing up as 'direct' traffic . The complexities of setting up Google Analytics were far greater than we had anticipated. We couldn't even track across our entire stack. At one point, we even considered relying solely on intuition and post-purchase surveys for insights because analytics tools were failing us. The turnaround: embracing Mixpanel The game-changer came when we started using Mixpanel for product analytics as well as marketing analytics . With the help of our engineering team and a consultant, we implemented simple product analytics. We used server-side tracking to bypass adblockers, a crucial step since 30-45% of our tech-savvy audience uses them. Unlike GA, Mixpanel could track these users. The depth of our exploration The more we used Mixpanel, the more we realized its capabilities. We were able to achieve full-funnel analytics across all our tech stack , and answer critical questions like: User Behavior and Engagement Where are users coming from? Track UTMs and referral domains. What is their behavior in the product? How many activate and take core product actions? How many people drop off at each stage in the funnel (e.g., from a pageview on the marketing site to copy/download to search asset)? Product Features and Metrics Which features are popular with different segments of the audience? Should we pivot to a different ICP? What happened to core metrics after we launched the redesign? Where exactly are people getting stuck in the product? Marketing and Outreach We could send cohort data to lifecycle marketing tools (e.g., send email to users that start a trial but never take core action event) Which partner is sending how many users? How many users sign up / start a trial / convert? To which plans? And when? If we run an ad on Twitter, how many users copy/download at least a single asset? How many convert after a period of X days? For people that use the free product, do they retain? Are we getting most of our new traffic from extensions and plugins? SEO and Content Performance Which SEO blogs perform best in terms of activation, engagement, and conversion? Which user converted from which blog? (Something GA won't tell you). How long does a user retain from each blog? Data Analytics and Modeling Map attribution models based on last touch and first touch data. Backport product signals to performance marketing and optimize for other goals besides conversions. What's the word of mouth coefficient for the app? Implementation team: who did what The setup was achieved by just a full-stack engineer and myself . For the initial product analytics setup, I brought on a consultant. This took a few weeks of work + additional few weeks of testing data. The work with the consultant wasn't very successful so we eventually took it on internally. For the marketing analytics this also took a month long implementation process. Our tech stack includes a react app, framer websites, a blog hosted on Ghost, a shopify store, and multiple plugins and extensions. We use a mix of server-side and client-side tracking to get the most amount of data. We ran through the analytics setup ourselves and made some mistakes. That's ok and was expected! Dev-details: For UTMs, we used Mixpanel's out of the box setup. To stitch prelogin and post login data, we used a combination of mixpanel.alias (backend) and mixpanel.indentify (frontend). If you are using only frontend library, then mixpanel.identify will do all the work for you. The UTM params are stored in the cookie but are only being sent on certain events and not all of them. I use the Initial UTM params on the profiles most of the time so it's not that big of a deal. But it can be a problem and you could use formulas to bypass this. Google Analytics vs Mixpanel We still have code lying around for GA4 but I honestly cannot stand their UX anymore. I don't recommend using it. Plus, Mixpanel integrates with Hotjar so you can see a lot more visual data alongside metrics. But this is on a business plan. Mixpanel's also free up until 20M events, which should be pretty sufficient to cover all pageviews and MAUs. Earlier, they had MAU based pricing and it was tougher to pay because all site visitors would get counted agaist an MAU based pricing. So, as far as Google Analytics vs Mixpanel goes, I'd pick Mixpanel all day. The Success The insights we get out of Mixpanel was historically impossible to do so and we couldn't be happier with the time invested. What we ended up with was a robust analytics system that I once only dreamt of having. This allows me to focus on channels that work and remove ones that don't. It also helps identify friction points in the product. The limitations (As I see today) 1) Filters display only 3,000 rows at a time. To view more, you can filter, segment, or export in CSV format with up to 10,000 results. However, for larger sites, especially in programmatic SEO, you'll need to analyse outside of Mixpanel. 2) There is no native integration with Search Engine Console. However, GSC data is sampled and not entirely accurate for low volume keywords in terms of click data. So Mixpanel is reliable in terms of click data. 3) Unlike Google Analytics, there's isn't a no-code integration with Looker or Data Studio. 4) Tag Manager integration is now available. I implemented it quickly on my site ( toption.org ) in 10 minutes, but it doesn't track 'time spent'. This might be an error on my part but could also not be. But the team is shipping fast. A month ago, they didn't have GTM but now they do (after I had a feedback session with their PM). So, tech is evolving in the right direction. [SOLVED: it works with "Session Duration", not "Duration"] 5) If you have a landing page that also acts as a navigational site item, it's much harder (or impossible to get data) on how many users land because you need to exclude users that just find the page via navigational search. [ SOLVED ] 6) You don't see metrics like bounce rate right off the top of Mixpanel but you could create a funnel report. [ SOLVED ] Thanks for reading! - Khushi Lunkad

Hey, we lost our homepage 👋 If you were to visit streamlinehq.com today, you'll go straight to the app. No marketing site. No same old "get started CTA". And you can export as many icons as you want without ever signing up. For first time users, we display a welcome modal. Take a look here: It's a bit uncommon in SaaS to drop the homepage for the app, but not all that uncommon for several B2C companies. Reddit lets users interact right away. So does Instagram. Have you ever seen Instagram's marketing site? Nope, because people don't enter Instagram via the marketing site. They land on a profile page, an individual reel, a hashtag etc. So, what should you do? Make all pages that people enter via ungated? Not so fast. Let's take a look at how LinkedIn personalizes. 1️⃣ Direct visits to a profile page have a compulsory sign up. 2️⃣ But if a user lands on a profile after searching on Google, they soften the ask a bit. People landing via organic search can bounce off. So, they have the ability to 'close the popup'. 3️⃣ And ungate it altogether when users land on a singular post or a blog article written on LinkedIn. Linkedin matches the user's intent & motivation with the level of friction they can add. And that's the ideal goal. The larger a company, the better it can segment. Everything will always be a tradeoff regardless of whether you're a large company or a small company. You have to answer a few questions: 1) Do I gate or ungate? 2) Do I make it compulsory or not? 3) Do I use a soft nudge or a much stronger one? I'd like to share how we arrived at this decision. And what was the impact (good and bad) of doing so. After that, I'll share some more examples from other companies. The evolution of our homepage This was one of the versions we had 2 years ago. There was a navbar that I can't seem to add in the screenshot so please use your imagination! And by all means, this was a pretty good homepage. In fact, I later found out that it was featured in Jeremy Moser's guide calling it the "perhaps the best landing page hero on the internet". It's nice to know that a homepage performs well when you look at the data and reports. However, it's even more special when you discover that other people reference it too! Having external positive validation makes it more challenging to iterate because you don't want to break what's working :) But we did iterate. Our homepage evolution: Second version At the time, I thought it was pretty good. People understood our value proposition so well that they repeated it verbatim in user survey interviews. We also had buttons for people to install a Figma plugin. So, people could build a habit of using Streamline in their daily workflow. Our homepage evolution: Third version We had a rebrand. And here's what it looked like. In the previous versions, I had a very strong influence on the marketing site. I dropped screenshots and built out the wireframes. But in the third version, my primary goal was to let the art shine. My brief was "It shouldn't look like anything marketing has designed." I think for marketers, we tend to own the marketing site and logged-out website experience. But there may be times when you want to give up that control. More so, when you market to technical audiences like devs or designers. Our design team is incredible so I wanted to let their craft lead the process. In my case, I shared as much context as I could and supported the design process. I focused on CTAs, pain points, and value props. But beyond that, I didn't want to offer any direction on design. I only copy-edited. And once we shipped, we were pretty happy with it. The data also confirmed it was a good move. Our homepage evolution: Fourth version 🟢 After this, we dropped the homepage and moved it to a separate page. Now, when people land on Streamline, they are greeted with a welcome modal which they can close and continue with their experience. It took 2 designers, 1 developer, and me around a week to get this live. SEO constraints: It represented a step back in terms of SEO because a lot of the on-page SEO would be lost. As long as we delivered on the UX, our rankings wouldn't tank. You may also have duplicate pages so you'll have to handle some redirections. I'd recommend visiting Search Engine Console > Links to see which pages have more backlinks before you make the 301 redirection. Emojipedia/FSymbols handle this pretty well since their traffic comes via programmatic SEO. Test, if you can Quantitative : I had also A/B tested it over the years and we had data to confirm it would play out well. Qualitative : People enter Streamline in more ways than the homepage. It's just like tools like Wikipedia, ShutterStock, and Emojipedia. People don't follow a traditional path of the usual homepage ⮕ app pattern. With or without a modal window We wondered sending them directly to the app but having that first-time landing experience would help address some of the value props and hesitations even though some users may drop off. With or without an onboarding Rows has an onboarding whereas we don't. Excalidraw is a diagramming tool without an onboarding. It would depend on how intuitive your product is. And what are the current priorities for the team. If people fail without an onboarding, then you probably need one. Do I trust my gut or data? I don't think you can truly fully be data-driven. A lot of it will also come down to what your gut says. Handing analytics We've migrated to Mixpanel for marketing and product analytics. I wrote about our implementation here . Originally, we were using Mixpanel with localStorage and had to opt for cookies instead to make the subdomain tracking work. Our inspiration We were inspired by Typefully . They have a modal window and bento designed cards. Another inspiration was rows.com . They dropped their homepage and saw a massive improvement. Their Head of Growth wrote about it here . Sanity checklist Do users get a value out of the marketing site or do they only click on the first CTA to enter the app? Do you expect activation, retention and other key metrics to go up? Or are you only going to feed the app low-quality, de-motivated traffic. Marketing site can help raise the motivation levels. Are there more ways to enter the app and is the app ungated? What would be the simplest possible way for you to validate? Maybe via a paid ad split test? Change link in bios? Email CTA? Do you have a horizontal product with lots of use cases? Do you have enough traffic to make it worth it? Will the unit economics make sense? If people reset cache in their browser, could they misuse the platform? Are you ok with that? Is it reversible or do you have a lot of problems that can arise? Why didn't I share metrics? Lying with numbers is pretty easy and I could do that if I wanted to. But honestly, I don't trust when I see claims like "boosted revenue by 30%", or "bounce rate decreased by 24%" because more often than not, it only shows a part of the picture. I often question these claims. What if your bounce rate decreased but you only had 1000 visits a month? What if the activation rate increased because you found product-channel fit this month? If I did want to share numbers, I'd have to share everything with you. - User count - Activation metrics - Retention data - Vanity metrics like bounce rates - Qualitative feedback - Before vs after - With modal vs without modal - Impact on enterprise sales and so much more. Only then, should you trust my data. But since that might be too much information, I guess the only data point that you'd need from me is, "Hey this worked. It may or may not work for you but I hope you're inspired!". But you can still verify by some simple math: Say you have 100k visitors to the homepage. Of these, only 20% click through while the rest drop off. You could get that click through rate to a 100%. Maybe users are slightly less motivated and activate at 15% instead of 25%. Control = 100k*20%*25% = 5k users activated Test = 100k*15% = 15k users activated Thanks, Khushi

There are a bunch of education programs out there where software companies give access to students and/or teachers. This might mean: a free product a lite version of the product a custom product a discounted product I've set these up a few times now, and wanted to share some learnings. Always set up a loop It's easy to just discount the product and allow access. But, if you're not the incumbent, I'd encourage you to go a step further. Can you encourage educators to bring all of their students along? And vice-versa? For example, this form asks students to refer their educator. (Tip: Let the student opt-in to an introduction before you ask this question.) I've found that sharing educators handbook improves adoption rates. These typically cover do's and don'ts. BJ Fogg Model explains users takes action based on how easy something is to do and how motivated they are But what you can only learn in his paid course , is that it's far easier to move people along the X axis than Y axis. Essentially, it is very very hard to motivate people. So just make the task simple. And then make it even simpler. In this case, if you can make it simple for educators/students to opt-in, it helps far more than you would realize. You could get away with an average benefit if you made it easy to opt-in. The contrary is true as well. In the past, I've offered each educator a custom URL that they can share with their students. This it was quite successful and helped with retention on both sides of the market (student + educator). This custom URL gets embedded in the course forever. Add it where the friction appears I love adding on the pricing page as an FAQ. Pricing is typically a friction point for students. So, I've found it to be an ideal place to add it. Figma also places the education program in the pricing page. You can add it in the onboarding flow. This is what Notion does, and it's typically more generous (and aggressive!). It's proactive vs reactive. In-app pricing modals are another good place to add them. These show up when you click on an upgrade button in-app and aren't navigated to the /pricing page URL. 3. Automate when you hit scale The third tip is to automate. Once you hit scale, you want to monitor for fraud and unauthorized use. There are a few tools to help with that. The simplest setup is just to use some Zapier/Relay app logic. And connect that to your email marketing system. For example, for one company, we wanted students to use the product and share their learnings publicly on Reddit, Medium, Instagram etc. That was the growth loop that we ideated. It was designed to help generate UGC. For students, learn-in-public was a good strategy to build a portfolio of work. We wanted to automate the follow ups. If a student had 3 months worth of access, we wanted to send a 'collection' email at the end of each month. It was easy to orchestrate with a simple email marketing tool. Just look at their start date and trigger every 30 days after that. There are a few other tools too. SheerID can help verify educators and students. This is useful if you're giving free access to educators but monetizing students. Harvard uses them. Another option is to use Github's Student Developer pack. Useful if you want to reach a dev-first audience base. This takes out the need to verify students yourself but will restrict people who don't have a verified student ID. See MongoDB, Heroku, and datacamp are all on this list. UNiDays is another one for a more mass-market product that can handle verification for you. There are many permutations on how to offer access. It really depends on your tech stack and ability to invest more into it. You can even localize to each country and each region! The biggest challenge I've seen is that you may not want to restrict to only those people with a .edu email address. People often learn on-the-job or in bootcamps. This happens with a more niche product where your market isn't as horizontal as a Notion. But being the default tool of choice is far more important. In that case, you may have more flexibility to give them access with a slighly more manual process. You can allow students/educators to upload an ID instead of using a .edu domain. And there are more creative ways to consider ;) Thanks for reading! P.S Other alternatives to consider: Student Beans and ID.me . Best, Khushi

If I told you that... Watermarks don't have to be visual That they could still be added to users who've paid you money in order to... remove watermarks. Or that they could even be added to your enterprise customers in a very sales-led org Then, you'd probably say that I'm making things up... Well, watermarks don't need to have a singular edge case. They don't have to look like a brand vomit, making your product impossible to use in production. There are more ways than one that benefit free users too. Let's look at some examples. 1. Streamline uses suffix in the name When users download free icons/illustrations from Streamline, a suffix is added to the icon name. The suffix is a combination of the brand name (Streamline) and the icon family (Ultimate). It also shows up when users copy/paste the code: Alert Bell Notification 2 Streamline Icon: https://streamlinehq.com

I have both Windows and Apple devices. Asus Zenbook, the one with touchscreen Macbook Pro, the one with 48GB RAM I want to test our product on both Apple and Windows. And more importantly, each operating system has it's own marketing channels to piggyback on. So, I can't feel confident working in tech, without having access to both ecosystems. For example, Apple has a menu bar at the top where apps can live. This doesn't exist on Windows. So, if I only had a Windows machine, I would never know! Apple has some limitations too. For example, Windows has a very powerful clipboard and emoji system. I can copy images, links, text. And it gets stored. I can pin copied items like my email address, calendly link, and GST number. On Apple, life isn't always easy. Pasting emojis requires so many clicks that people buy a streamdeck device. On Apple, you can't open two full screen apps on the same desktop. CMD+Tab is terribly inefficient, especially if you have more than 1 chrome window. So if your app is an overlay, things get harder. The natural user behavior and what people are accustomed to on both devices is incredibly different. Trying to change consumer behavior is hard. It would make sense to lean into whatever it is that they're trying to do. For example, if your product isn't successful with Windows users, it might be good to question why? Arc, the browser company, was Apple-first. I didn't understand the hype when I was Windows-only. And it made sense why. Many of Arc's features were on Windows already. Some even better. Pinning chrome profiles to the dock is possible in Windows, but not on a Mac So, it didn't make sense for Arc to offer it to Windows users. Instead, they kept it invite-only for Mac users. For Loom, I can almost bet that majority of their customers are Apple users. That's because Slack's video product on Apple laptops is unstable. And Loom on Windows is unstable. Windows also has their own screen recorder software that is a mix of Veed and Loom. Apple only has Quicktime. It's primitive and can't even mirror a recording. The audience expectations will be different because the context is different. One will expect more settings and shortcut keys than the other. If your product is used by knowledge workers that work on multiple screens, then there are more caveats. Apple often doesn't allow certain apps to be moved to a second screen. For example, I can't move Todoist to my second monitor. Logitech MX Keyboard/Mouse has an app that can't work well on Apple because Apple doesn't have all shortcut keys that Windows has. Apple has something Windows doesn't too. They have more indie-developed products like ScreenStudio, Bettertouchtool etc. This is stuff I haven't often seen on Windows. I honestly bought a Mac for ScreenStudio :) You also get realistic colors on a Mac than you'd get on Windows. The camera quality and speaker volume is better. Mac doesn't heat up or has fans. So there's a certain aesthetic people are used to, that your product has to fit into. I could go on and on.... but the bottom line is that if you're working in tech, maybe get both devices. Why Apple Wins, Even Though... For me, a Windows machine is more powerful. It helps me be more productive. Everything takes an extra click on Apple to achieve the same thing. And many times, it's just not possible to do things on a Mac. So objectively, Windows is better across all things I care about. But Apple still wins on one thing. Their marketing. And how simple it is to buy their product. Apple doesn't confuse you. See Apple's webpage. Naming is easy - it's Air or Pro. Very few options - at most, there are a couple of options to choose from. Now, let's look at Asus: Instead of letting me filter, they categorize laptops into home vs work. Why is this necessary? Can't a laptop be used for work and home, both? They have complex naming, like Zenbook vs Vivobook. How can a consumer understand what the difference is between those two at first glance? Even if I filter to a specific type like Zenbook, I'm flooded with so many options to choose from. What exactly is the difference between Zenbook 14 vs S14, if pricing is the same? What exactly is NPUs used for?... Is Dell any better? I see many popups. Some that I can't close easily. My first decision is to choose between Dell Pro vs Dell Plus vs Dell Pro Plus vs Dell Pro Premium vs Pro Max.. I'm serious. That's the 1st choice 🥲 There's something like "Alienware", which I don't immediately understand. Even when I filter for 'Professional Grade Productivity', I get flooded with 13 options to choose from in a UI that is confusing. Thought it couldn't get any worse. But it does. This is HP. It never loaded. Reloading never fixed it. The first choice to make was "Laptops on Sale" vs "Laptops". That's an odd choice to make, because everyone would want a cheaper price. But with a cheaper price, what am I losing? I see banners plastered all over the screen. Discounts thrown too early devalue the product. Lenovo personalizes, but at what cost? Yes, I live in Pune. But I'm not ready to buy the second I land. They say to order before 5pm. I don't understand why customers need to be told that on the first session. Especially for something that's a thoughtfully purchased item and not an impulse purchase. If Apple wins, maybe they deserve it. Asus has a much better product and a far superior service. They send people home to fix things for free, if all I do is write an email. I don't see Apple shipping features as fast as Windows does, or even to the level of quality that Windows does. I've had Macbook Air while I was in college, and I don't see many new features in the past 5 years. But the buying process with Windows is complex. You'd need an engineering degree to understand what product you need to buy. The other thing that Apple does is that they have a robust student program. It's the only ad they're running in India (or the only one I see). Switching costs are so high after a certain age, that it makes sense to bring people early in their journey. Two lessons for me to take forward: Make buying simple (Apple wins) Make people productive and be innovative (Microsoft/Asus wins) Thanks for reading!

Zombie users are paying subscribers who pay you but don't use your product. They get counted as active subscribers, when in reality they should be considered as churned subscribers. How to measure zombie users in your product? Step 1 - Identify your engagement metrics This could be anything like pageviews, sending a newsletter (eg: Mailchimp), creating a draft (Jasper), sending a message (Slack), or creating a board (Mixpanel). Step 2 - Create a filtered view of your users I like to keep my annual and monthly subscribers separated but you can cluster them together. Here's a rough idea of how that could look like. Then sort the report by value. Customizing your subscription health 1. Using % trends You can add an event with all active subscribers. Then use a formula to find the percentage of healthy active users or zombie users (eg: zombies / total subscribers%) 2. Segmenting further You can also create three groups: healthy, at-risk, and zombie. So, you can try to win-back at-risk users. This is important to do because Zombie users might come back later and request for a refund. They might also not be happy with the brand. What should you do about zombie users? I really like what Slack does with Zombie users . As soon as they track that a user has not been active in the past 30 days, they don't bill the workspace owner. Why exactly is Slack doing and isn't doing? 1. Always charge for 1 user even if the user becomes inactive. This is probably because Slack has a cost to serve in terms of data storage costs for the workspace. 2. Don't refund the charge but instead add a credit They aren't refunding for the month. Instead, they're saving a little bit of cashflow plus ensuring that the customer's bank statements remain clean. I don't like using a lot of payment apps like Google Pay because they offer cashbacks. Cashbacks ruin my account statements. Why is this a good billing strategy for zombie users? People often hesitate to start a subscription because they're worried they will forget to cancel it. This pricing decision reduces friction to conversion. It's also good for brand and word of mouth. Clearly, I'm writing about them :) And it's also good for the customer. They aren't shocked or feel like an idiot when they see their statements and realize they forgot to cancel. But I acknowledge that this pricing decision can come from a place of privilege. Some companies can't do it if they need the cash. One of the companies I worked for signed an annual contract with Sendbird (a messaging service). Now, development time took longer but Sendbird wouldn't extend or cancel the contract. This was a tiny startup paying thousands of dollars for a service they weren't using. We wrote to them a few times but they just stopped responding. In a world of Sendbirds, be a Slack :) Thanks for reading! Khushi

I have a serious problem. It's when copy is full of feelings like "this is the best software for" or "you'll double revenue by x". Everything is disputable. That's cotton candy copy. This is my plea to.... stop writing such copy? To describe Streamline's Flex icon sets, we've said something like this. Go ahead, try and read it!! "Flex redefines icons with smooth, flowing curves, moving beyond rigid structures for a more natural, dynamic aesthetic. Designed on a 14px grid, it brings fluidity and elegance to any design. With adjustable size, stroke, and color, these icons adapt seamlessly to any project. How enjoyable was the reading experience for you? It wasn't for me. Every set in Streamline comes with a color editor so you can change colors. It’s cotton candy copy because it tries to say too many things at once using vague adjectives. Every icon is supposedly seamless to integrate (atleast in Streamline). 14px is redundant. It exists as a separate line item on the page anyway. Flex is a wonderful set, and the description doesn't do it justice. [Please know that Streamline was ok for me to share WIP projects and share my learnings with the world. Companies may differ so check your legal contracts.] Now, read something like this, "Bobbie Organic Whole Milk Infant Formula is the first and only American-manufactured, USDA Organic whole milk infant formula and our closest to breast milk yet." This copy is probably easier to read. It does a few things: 'first and only' explains your reason to exist. 'American-manufactured' is another USP 'closest to breast milk' is a visual metaphor that people can immediately understand Let's look at another example along the lines of metaphors. "AI-first UCaaS for team collaboration lets you work together without friction using Meetings, Chat, Docs, and more, all built into Zoom Workplace." First of all, do you see the amount of text on this page? It's not meant for skim-reading (and most people skim read). Zoom also uses jargon like 'UCaaS'. Do you know what is that? See, I don't have any context on what this kind of copy is doing for Zoom. But, as someone with an opinion, I feel compelled to say, that I wouldn't want to write Streamline's copy like this. It's an anti-inspiration of where we'd want to be as a product . We don't want to overestimate what the customer knows, nor understimate their knowledge. Let's take another example. "Being naked is the #1 most sustainable option. We’re #2." It's visual. It's concise. It's extreme. It tells everything the customer needs to know, in the fewest words possible. Here's another one. "Romance for one. Shop the Maven Dress from Reformation, a cap sleeve maxi dress with a mock neckline, drop waist, and pleating at the skirt." It's visual. It's a metaphor. It's factual (cap sleeve maxi dress), with all details you can't immediately see. It is not subjective. There is nothing on the copy that you can dispute. Let's compare with another brand that is opinion-driven and is subjective. "Loud, comfy & so much fun! Made from the softest fabric, it's like wearing a hug that's also chic. This is all you need to make a statement without saying a word. The statement? Being chic should always be this comfortable!" There are too many adjectives. “Loud,” “comfy,” “fun,” “softest,” “chic,” “comfortable.” These are all subjective and mean nothing concrete. Who decided it's comfortable? Tries too hard to sound fun. You don’t learn anything about the product itself (style, silhouette, fabric type, structure). Redundant and wordy. Repeats “chic” twice, uses clichés (“make a statement without saying a word”). Exaggeration is ok. Subjective/opinionated isn't. The reason why I'm sticking to fashion and beauty brands is because there's not a lot of differentiation. It's an over-saturated space. And yet, they have found a way to communicate clearly. Software has so much more to offer. So, if copywriters get to know the product a bit deeper, we can write a lot more meaningful copy. Let's take an example from rhode. They have a moisturizer, named as "glazing milk". The name alone is very visual! The description is better! The essential prep step for your skincare routine. Glazing Milk is a potent, nutrient-rich complex with a milky texture that leaves skin feeling hydrated and glowy while boosting the skin barrier over time. Notice the words used: essential prep tells you when to use it skincare routine tells you to add it to your routine potent, nutrient rich, leaves skin feeling hydrated, boost skin barrier... all do things. They're active, factual and sensory. Not vague. texture 'milky' is described visually no repetition or any filler Unlike Streamline, this is not wordy. Every sentence adds some meaning. On the other hand, some brands use “perfect texture” or “incredible formula” or use adjectives that are subjective rather than objective. Metaphors go a long way. Songwriters often use them. This is one: “You should think of your energy as if it’s expensive. As if it’s a luxury item. Not everyone can afford it. Not everyone has invested in you in order to be able to have the capital for you to care about this.” Compares energy as expensive. As a physical item. I looked at the top Billboard songs and got ChatGPT to analyze it for metaphors. Here's what it found. Song Excerpt Metaphor Explained Daisies  – Justin Bieber “We grew from the dirt, and I still see your roots in me.” Concrete imagery (soil, roots) Lose Control  – Teddy Swims “I feel like a ghost without your breath.” “Ghost without your breath” conveys dependence without adjectives. Soda Pop  – Saja Boys “We fizz till we’re flat / taste the rush then fade to black.” Everyday object (soda) turned metaphor. Soda as a stand-in for youth, excitement, and burnout. Simple nouns (“fizz,” “flat”) do the emotional lifting; no subjective adjectives needed. I Got Better  – Morgan Wallen “Turned my heartbreak into horsepower.” Uses a physical measure (“horsepower”) to depict inner change. Love Me Not  – Ravyn Lenae “I’m the echo in the canyon / you forgot to call back.” Sound and distance depict rejection. Your Idol  – Saja Boys “Polished my pain till it gleamed like your chain.” Turned something abstract (pain) into something visual (chain) If you have a favorite artist, take a look at their lyrics :) Typically, I want to avoid jargon as best as possible. But sometimes you can still have jargon, if it has a novelty effect. For example, Loro Piana describes their trousers as: "Crafted for refined ease, the Coste trousers are knitted from pure vicuña – a precious fibre sourced from the Andean camelid, revered for its softness and natural warmth. The half-fisherman’s rib construction accentuates the sleek, tapered-leg silhouette, anchored by a flexible waistband. Considered details, like ribbed edges and subtle drop stitches, sign off the pair with the Maison’s characteristic savoir-faire." It's long. But it's readable. Each word is unique. You may not know what 'Andean camelid' is but it's precise. It uses a real, specific term instead of a generic word like “rare animal” or “soft wool.” “Andean” instantly paints a picture of mountains and altitude. You can almost feel the terrain. It implies exclusivity through specificity. The reader senses luxury because few people know what a vicuña or Andean camelid is. The expensive diction is still clear and literal. There is no fluff like “exquisite,” “beautiful,” or “perfect.” You can only write non-subjective copy if you know the product well enough. When you don't know the product, you gravitate towards writing subjective copy. We invent feelings when they we don't have the facts. Knowledge makes copy concrete. I've created a CustomGPT for myself with all examples. If you have a style you like, save the ideas somewhere on a Notion doc, and paste them into ChatGPT as a pdf to work on top off. You'll be surprised at the results. ChatGPT gave me this copy for Flex, "Flex is a fluid icon family inspired by nature’s curves. It replaces rigid geometry with smooth, flowing forms. It’s more organic and expressive than Core (neutral) but subtler than Plump (chunky, playful)." It needs a few more tweaks before it's customer-facing. But I'm pretty happy with the bot! If you feel like, drop a comment below or say hi !

The first step to managing churn is to track your voluntary and involuntary churn rates separately. I cannot stress how important it is to break it down, especially as you begin to hit scale. I’ll share a few ways you can calculate your involuntary churn rate, starting from the simplest to a more comprehensive one. Method 1 : Easy, less accurate Method 2 : Complex, more insights Method 1: Use Stripe ...or any billing provider. Step 1: Export the Recoveries Report Navigate to: Stripe Dashboard > Revenue Recovery > Failed Payments Set your date range (e.g., last 3 months, or custom range). For custom range, you may need to use Sigma. Open to view code. You can customize the date window. select reporting_currency, initial_payment_decline_reason as decline_reason, sum(initial_failed_amount) as initial_failed_amount, count(*) as initial_failed_count from recoveries where coalesce(initial_payment_failed_at, paid_at) between cast('2025-11-01 00:00:00' as timestamp) and cast('2025-12-01 00:00:00' as timestamp) and initial_payment_decline_reason is not null and initial_payment_failed_at is not null group by 1, 2 order by 3, 4    Click "Export" and download the recoveries data as CSV This gives you the Recoveries_[date_range].csv file Step 2: Get Your Starting Subscriber Count Navigate to: Stripe Dashboard > Billing > Subscribers Set the date to the start of your analysis period (e.g., Oct 1) Note the total number of active subscribers This is your denominator. Step 3: Calculate Involuntary Churns from CSV Open the CSV and filter/count rows where: retries_exhausted = "true" AND amount_paid = "0.00" AND recovered_at is empty This count is your involuntary churn. Step 4: Calculate the Rate Involuntary Churn Rate = (Involuntary Churns / Starting Subscribers) × 100 Example using your data: Involuntary churn = 40 (from Step 3) Starting subscribers = 1000 (from Step 2) Involuntary churn rate = 40/1000*100 = 4% This is a rough, back of the napkin method to calculate your involuntary churn rate. ❌ Don't use this Stripe's "Failed Payments" chart because it includes: Payments that eventually recovered (through retries/dunning) Temporary failures that customers fixed You can extend this calculation to the voluntary churn piece as well. Step 5: Extract the total churned subscribers. Make sure you set the exact time period as your exported CSV. It might look something like (Oct 1st to Dec 23rd) if you're exporting mid-month. Let's assume this was 200 subscribers when exported for ~2.5 months. So subscribers lost due to voluntary churn is 200 minus 40. = 160 Voluntary churn rate is 160/1000 = 16% 80% of your churn is voluntary (customers actively choosing to cancel), while 20% is involuntary (payment failures). The biggest downside of this method is it only captures permanent involuntary churn (customers you lost), not total involuntary churn risk (all customers who experienced payment failures, including those you successfully recovered). Method 2: Use Proper Tracking Larger companies track churn at the moment of cancellation by capturing the cancellation trigger, not retrospectively analyzing payment data. Step 1: Extract reason code Every subscription cancellation gets tagged with a reason code: This is how Stripe treats it: "cancellation_requested" → Voluntary "payment_failed" → Involuntary "payment_disputed" → Involuntary (chargeback) Voluntary churn would encompass when the user clicks on the cancel button, user doesn't renew, user downgrades to the free tier. Involuntary churn would encompass if subscriptions are cancelled due to a payment failure and when dunning retries are exhausted. Step 2: Use an internal system Beyond Stripe, we would also want to use an internal system. You should track all the payment methods you use, the locations where users come from, reasons why they churn, their card network, and so on. Your churn rate calculation should also be a rolling churn rate instead of a point-in-time snapshot. The simple method won't work if you're growing fast. You'll want a lot more granularity in the data. And you need help from data scientists. Step 3: Draw conclusions and hypothesis You'd want to start cleaning up the data for all your important regions (e.g., if the US is very important for subscriptions) so you can measure the idiosyncrasies of each market. You might find that Brazilian users tend to wait quite a while before subscribing, even way past their 30-day free trial, up to 45 days for conversion. If you discover that you have higher involuntary churn, paired with customers returning about 60 to 90 days later, it might point at payment friction. You can switch payment providers in particular regions as an experiment. Through data mining work, you'd discover insights like price sensitivity by region, preferred payment methods by country, and behavioral patterns that differ across markets. Realization of churn One accounting question that decides churn rate is when should a failed payment be realized? And when should a voluntary cancellation be realized? I'll leave you with this note from Netflix to make a decision, "A membership is canceled and ceases to be reflected in the above metrics as of the effective cancellation date. Voluntary cancellations generally become effective at the end of the prepaid membership period. Involuntary cancellations, as a result of a failed method of payment, become effective immediately." Stripe allows you to customize this in their platform. Benchmarks Involuntary churn benchmarks across billions in subscriptions look somewhat like this: Industry Total Annual Churn Annual Involuntary Churn % of Total from Involuntary Digital goods 38% 11% 29% Business services 40% 10% 26% Personal services 36% 9% 25% Education 40% 10% 24% Merchandise 39% 9% 23% SaaS 38% 8% 22% Travel & lodging 43% 8% 18% Leisure 36% 6% 17% Insurance 37% 3% 9% Business Model Annual Churn Annual Involuntary Churn % from Involuntary B2C 39% 9% 24% B2B 38% 6% 16% Price Tier Annual Churn Involuntary Churn % from Involuntary <$10 40% 14% 35% $10–$30 37% 9% 23% $30–$100 34% 9% 26% $100–$1,000 30% 6% 19% $1,000–$10,000 24% 4% 15% >$10,000 15% 4% 24% Hope this helps :) ~ Khushi Lunkad