The Reality of Building an AI Learning App: 4Months, Zero Revenue

Hi, I’m Iaroslav, aka @m_0_r_g_a_n_. Today I want to share my story of spending almost 4 months building an AI English Learning web app and getting ZERO paid users. It’s been an incredible journey… 😅

I think it could be interesting to see another story, unlike those super successful ones like “I got 50K users in 2 weeks” or “I got $10K MRR in 1 month”, because I think it’s what reality looks like in 95% of the cases.  Real success is super-hard to achieve.

For context, over the last 10 years, I have worked in different startups and had different roles, such as Software Engineer, Product Owner, Product Manager, Engineering Manager, and Tech Lead. During my university years, I also had some design and QA experience. Now I have a 9-to-5 job as a Tech Lead in a great UK-based startup.

Around 7 years ago, I burned out working on my SaaS app for Amazon sellers (which was shut down at the end). Over the years, I tried Unity, YouTube, Amazon KDP, and Swift. After all that, I decided to stick to the stack that I know best (Ruby & Ruby on Rails) and learn some frontend frameworks to become a full-stack developer and deliver an app independently.

To quickly learn the NEXT JS and keep up with the latest tech vibes simultaneously, I decided to work on a small pet project called Grammar AI, which generates English grammar tests based on any grammar rule. It seemed useful and not so hard to start.

For the MVP, I decided to start with the following features:

1. Test Generator itself as the home page. The initial idea was that all tests are generated on the fly with the OPEN AI API. I was too naive 😆.
2. AI chatbot, which acts as an English tutor (GPT API based).
3. As I wanted to have some AI magic for the users to happen, I thought that it would be cool to have the ability to get an explanation of why your answer to a test question is wrong or correct.
4. Recurring payments to get payment from the user.
5. Basic user account logic.
6. And all the other stuff that you usually need (Terms of Use, Privacy Policy, Cookie Policy, cookie consent pop-up, Analityc)

THE TECH STACK

If you are interested in what components and services I choose to build the app, here is the full list:

Hosting

• Backend: AWS EC2 (Server), AWS RDS (database), AWS SES (emails)
• Frontend: Vercel (Server)

Frontend Stack

• NEXT JS
• CHAKRA UI

Backend Stack

• Ruby on Rails as API. I used a lot of gems for all the standard stuff like user accounts, auth, etc.
• Open AI

Payment Gateway

• Lemon Squeeze

Privacy Policy & other legal docs

• Termly (https://termly.io)

THE LAUNCH

Based on my experience, I knew that delaying the launch or creating more features than were needed for the MVP made almost no sense. Also, as I have no audience and no plans to run PPC from the start, I will have plenty of time to fix and add more features after the launch if needed. Plus, understanding that the product is live and someone is using it adds additional pressure and keeps you focused on what needs to be worked on.

I had never tried React or NEXT JS before, but with the help of Chat GPT, I was able to deliver and really learn fast. Launching the first version of Grammar AI took me around 60 hours of work, and in two weeks, it was ready to be launched. At that point, 90% of the frontend code was copied and pasted from Chat GPT output, but it will change over the next few months while I am learning.

I had a lot of basic API things already done and ready on the backend side from older projects.

I was able to launch in mid-March. Here is what the test generator looks like:

OPEN AI PROBLEMS

As you already know, Chat GPT is good in some cases, and in others, it’s terrible.

The first problem I faced was that Open AI API could be slow in response generation or even dead. I didn’t want a user to have a bad experience when they needed to wait ~30 seconds for a test to be generated or not get the test because the Open AI API was dead. So, I decided to generate 10K test questions and store them in the DB.

The second and biggest problem was AI-generated wrong tests or answers in 22% of the cases. It was very frustrating as how can you learn and count on an educational app with the wrong tests and explanations, nonsense. No one needs an app that could have wrong tests cmon… I decided to write a few AI checkers and AI validators and was able to down the wrong tests generation % to 4% instead of 22%, a small win but still… In my view, it’s a fail and now all that I can do is add human validation to it to make it perfect.

All those test generations and validators skyrocketed my Open AI cost from $5-$10 to $134.02 in April. Also, it took additional time to fix and rebuild all of those.

ADDING FLASHCARDS FEATURE

After I fixed the test generator, I started to think about what else I could add useful for the users to bring more value to the app using AI and help people to learn English. And I decided to go with a flashcards generator where you can pick a predefined topic or write whatever you want to practice.

Polishing the whole app, rewriting the test generator, and creating the flashcard generator took all my free time in April.

APRIL RESULTS

During March, I submitted Grammar AI to all AI curated libraries that I could find and started tweeting about it. The app started getting non-relevant traffic because people were interested in checking another AI app and not improving their English. Still, it was enough to test the main user flows and fix some obvious bugs.

Here is the app analytic data for April:

And here is the financial success for April 🤑:

PRODUCT HUNT LAUNCH

Inspired by the zero dollars revenue month, I decided it was the perfect time for Product Hunt to enter the game 😎.

As I had no audience to support my PH launch, and I’m not a fan of reaching out to everyone on Twitter asking for upvotes, I had no illusions about the success of the launch.

My main goal was to get a backlink from the PH website for better ranking on Google. So, everything was ready to be published in 10 minutes or so, using ChatGPT and the standard screenshot tool. In the end, I even got six upvotes, and the backlink 😁

SEEMS LIKE MISTAKE

In May, I had time and data to check where I got most of the traffic for April.

I thought it would be nice to translate the site content into different languages to make it easier for users to interact with the app. I want to say that no one asked me about it 😂 And now, when I changing some text on the site or add a new one, I need to translate or update seven localizations files… It’s so much work that it was not worth it.

ADDING ONE MORE FEATURE

As a true Indie Hacker, Instead of talking to users to try to understand what they really need or doing marketing, I decided that I needed another feature, “Role-Play Scenarios,” where you can text-chat with AI on any topic. Sounds cool, right? I also think so 😂 So, I built it in a few weekends. In my defense, I can say that this feature became the most used one. Check the “Role-play Created” even chart below. But still, it brings zero paying customers.

BECAME MORE AGGRESSIVE

At the end of May I was a little tired of having no paid users, and I decided to be more pushy and try to convince users to register an account for using features. So, I created this blocking pop-up.

And in fact, I started to get more and more registrations.

MAY RESULTS

Based on the May numbers, it looked much better than in comparing with April (you can check all the numbers below). But I still had zero paid users 🥲.

• Visitors
  475 -> 1.4K
• Views
  2.5K -> 6.2K
• Avg. Time on site
  01:18 -> 01:44
• Bounce Rate
  27% -> 32%
• Google Web Master
  8 -> 97 Clicks
  41 -> 625 Impressions
• Registrations
  69 -> 263

During May, the app also had some SEO traffic, as you can see on the graph from the Google Web Tool.

NEW ROLE-PLAY FEATURE

I think it would be a great idea to rework the text-based role-play feature using voice-to-voice communication, just like in real life. You don't text with a doctor or flight attendant, you talk to them. So, another weekend, I built a fully new voice-to-voice role-play feature, and it's currently my favorite one! ☺️

You need to choose the language you want to practice in as well as a scenario.

PIVOT

For the last month, I have been developing another feature that should help people in IELTS preparation. I thought that it will be really hard to keep IELTS prep and other features in the same product.

I have decided to move the related IELTS feature to a separate domain, as it is very difficult to build an all-in-one solution. I believe it would be much easier to create content solely around IELTS and focus advertising efforts around it as well. Hence, I have created https://ieltschamp.com, which still has zero paid users 😆.

Additionally, in the next few weeks, I will be moving role-play to a separate domain as well.

Final Thoughts and Takeaways

I apologize if my story seems a bit inconsistent, but I hope you found it helpful nonetheless. As I reflect on my journey, I realize that there were several personal insights that I gained along the way that I would like to share with you.

As someone who has read a lot about building a successful product, I know that there are certain principles that are often emphasized, such as avoiding useless features, building something that people want, and starting small. However, I have also come to realize that there is only so much you can learn from simply reading or watching others play, whether it's playing the guitar or building a product.

From my experience, I have come up with a list of my top five takeaways that I believe could be helpful to other aspiring Indie Hackers:

1. Launching a product without an audience is incredibly challenging. It's important to have a solid plan for building and nurturing your audience before you launch your product. Or you need to know where to find it.
2. Managing a PPC budget as an Indie Hacker can be difficult and overwhelming. It's important to have a strategy in place to ensure that your budget is being used effectively. Or not to do PPC at all at the start. I thought that with PPC I would be able to get initial traffic and test and maybe even get some feedback, but still not happened.
3. Focusing on a specific niche or audience can make it much easier to build and promote a product. For example, creating a product that is specifically designed for IELTS prep is likely to be more successful than trying to create a product that covers all aspects of language learning at once.
4. Building a successful product takes time, especially in the beginning. It's important to be patient and persistent, and to be willing to learn from your mistakes.
5. Lastly, it's important to be open to feedback and to be willing to make changes to your product based on what your users or Indie Hackers community are telling you. This can be difficult at times, especially if you have a strong vision for your product, but it's essential if you want to build something that people truly love and find valuable.

If you have any questions or just want to say hi, please DM me at @m_0_r_g_a_n_