I built an AI Micro SaaS product in less than a month and grew it to a 1000$ a month in revenue. In this video, I will be showing you everything, how we did, how it got started, what was our tech stack, our real stripe numbers, server costs, development costs and more. Along the way I will also share with you the five key lessons about launching an AI SaaS that I personally had to learn the hard way.
This video is probably just going to be me doing a lot of screen shares, and showing you all of the secrets so you can really avoid potential pitfalls where so many products fail. So, let’s get started with the product demo, this is the current version of this product right now, and it’s called oai-widget, where OAI stands for you guessed it OpenAI. Essentially, this is just a custom AI website widget built on top of the latest openai assistants api, and it can answer your customers questions based on files that you upload.
So if I ask our homepage widget what makes this product different, as you can see it provides me as a potential customer with a pretty good response. But that's not it. You can actually connect these widgets to virtually any APIs.
So this widget is actually connected to our customer support API, which means that if I have a problem with the product, it can submit a customer support request for me. So let me now say that for example my widget is not openning on my website. It then asks me to provide a bit more details like error messages, my name and email.
Once I do that, it tells me that the customer support request has been submitted. So if I open now our customer support inbox, you can indeed see this new customer support ticket from myself. Now let’s click on a dashboard tab, this will take us to the main UI, where we have a few tabs on the left.
To create an assistant, we need to go to the assistants tab. Then click create new select your API key, and click create. Now you can customize them, just like on custom GPTs, you can enter name, description, instructions, and also upload your knowledge and add any APIs.
Then unlike with custom GPTs, you can actually take out this assistant and integrate it into other channels, like a widget. You can also customize it to match your website by changing colors, icons, start message, and so on. I thought this was a pretty cool and simple SaaS idea that we could build really fast and some people find it really useful, however, what we surprisingly figured out later on from customer support requests is that users were trying to use these widgets as standalone applications or as sections on their website.
Some users actually just embedded them on empty white pages in order to use them separately and share them with others. So this was the first most important lesson we learned the hard way: always listen to feedback. Feedback was hands down the most valuable thing from the entire launch, because it allowed us to really understand where we could take this product next.
I honestly never thought that it would become something more than just a widget initially, but because of customer feedback I realized that there is a lot more potential to this. So, this is why we decided to release this feature next called custom gpts. Essentially, it's standalone chat applications with more highly requested features like white labeling, managing user access, adding multiple assistants, accessing chat history, using voice, uploading files directly in the messages and more.
And you actually can see a noticeable bump on our Stripe dashboard after we released it, which I'm going to show you next. But before I do that, just so you know, this product now looks completely different. We've completely rebranded it, and now it runs on my open-source framework with a ton of new features like Slack integrations, which I am gonna talk about at the end.
Okay here is our stripe dashboard. As you can see we peaked at around 1000$ a month and then it slightly went down a bit. This is the bump I talked about earlier when we released custom gpts and sent out a small email campaign.
Literally, the whole marketing for this product was just me releasing one video about 8 months ago and nothing else. I just literally released one video and then I was also attaching a link to this product under some of my other videos, but I never promoted it again. At first, to be honest I was really disappointed by these numbers, because you know my B2B agency right now makes 50 to a 100 time more, but what I realized later on is that it’s actually not that bad.
Because if you look at our vercel, which is our front end hosting that we’ll go over later, as you can see we haven’t published anything for more 4 months, okay. That’s insane. Because imagine if we did this in our agency, right?
Imagine if we completely forgot about our service for more than 3 months. We would probably have had like a dozen lawsuits right now and it would immediately go down to 0, but this product just keeps working by itself with barely any involvement on our end, which is just insane. So, this is the second key lesson that I learned is that it takes much longer to grow a SaaS product then you think but it’s much easier to maintain it.
This is why SaaS products can be sold for such tremendous amounts of money. You don’t have to constantly hire and train new people so you can scale it as much as you want. Okay now let’s take a look at how we actually built this product and what was our tech stack.
So for the front end we are using Next JS, which is one of the most highly respected frameworks out there for building reactive web applications, and Material UI which is a component library. I definitely recommend you guys consider a component library if you're tight on the budget, because they allow you to build front ends much faster. Essentially, they provide you with a set of pre-made commonly used components, like buttons, different tabs, tables, and so on, so you don't have to create everything from scratch yourself.
Other options include bootstrap and tailwind. For our backend we’re using firebase which provides you everything need to develop scalable apps like this including authentication, database and backend API in a single SDK. We are running our backend primarily on firebase serverless functions, which in my opinion are perfect for AI apps because with AI you typically have some services that are used a lot more frequently than others.
Like for example, in our app we have a get completion function that runs your assistant and returns a new message which gets probably like 95% of all traffic. And with serverless each function can scale independently, so you don’t even have to think about any server management. The disadvantage of serverless is that most likely you're going to be locked in with your provider for life.
However, the real reason we're using this stack is because of the developer experience. This is the third key lesson that I learned is that the developer experience is key for success in software business. You see, developer experience is often overlooked, but it's actually your highest cost when you're building a product.
At least at the start, this is most likely reason of why your product can fail: you simply run out of money to pay for the developers before you find your product market fit and become profitable. We'll go into costs at the end, but I'll tell you right now that all other costs are absolutely negligible compared to the development costs when you are just starting out. So this is why I like firebase so much, it only takes you like 3 minutes and 5 lines of code to set up authentication.
You can access your database directly from front end, which saves a ton of time, and all you have to deploy your entire app is just run a single command. Then you never have to think about it again because firebase will scale and manage everything for you. Not many people know that Fireship's channel was actually named this way because he started from Firebase tutorials.
So if you want to avoid building a clunky three-page SaaS product for $200,000, I definitely recommend checking out some more modern tech stacks. If you prefer an open source alternative, there is also Supabase, and pocketbase. However, even with the most modern frameworks with all those SDKs, most SaaS products still fail, because there is one more component that all first time founders tend to forget.
And that is the people component. The fourth lesson that I learned the hard way is that every SaaS business is also a business. Sure, you can grow a product by yourself to a certain extent, but if you really want to keep expanding it over time, I guarantee that at some point you’ll get stuck.
Instead of working on long-term vision and strategy you'll be stuck in low-value tasks like customer support, finances, and bug fixes. To avoid this you need to put some systems in place. Personally, we use Notion for our management combined with a PARA system by Tiago Forte.
Essentially we organize our notes, tasks and files by projects, so whenever people work on this project, they simply open this project in Notion and instantly they have everything they need in one place. This helps them to actually focus on doing the work, rather than going around and distracting one another while looking for API keys or whatever else they might need. Additionally, we have extensive playbooks which are like training manuals for our employees that help them to onboard on projects and collaborate with others much faster.
We also have automated CI/CD workflows on github that automatically run all tests on every commit, so we can catch bugs as soon as they appear. A great practice that you can implement right now is test first. It has been shown in many studies that developers' performance increases significantly in any case when tests are written first.
So don't stress about this in the beginning. You can easily get started by yourself. And once you begin to grow, simply record how you do things and then hire someone else to do them for you.
That's the whole idea. All of these processes and playbooks are literally just hours of recordings of myself on how I used to do things when I was all by myself, just getting started. Now, let's take a look at the costs Okay, first here are the OpenAI costs.
On our platform, as you saw before, users actually provide their own API keys, which is why the only OpenAI costs we incur are for this homepage widget. Additionally we are running some tests on the backend that also use openai, but the total OpenAI costs are only at around 20$ dollars month. This is not a lot but on the first months after release in Deccember and January we actually got a much larger bills for this widget at 140 and 100$.
It was still using a previous turbo model however, so with new gpt-4o it should like 4 times cheaper. Now, let’s checkout the firebase costs. So as you see it’s around 90$ a month, however, most of these are actually non-Firebase services.
This is because for some critical cloud functions, like get completion, we are setting minInstances to 1 to reduce the latency. This is not necessary but can slightly improve user experience. So, this means, that if our application keeps growing, those $60 will stay the same and only the $30 on top will go up.
Vercel which is our front end hosting also costs us 20$ per developer per month, so this adds another 60$, however, you don't really have to add all your developers there. You can, for example, just add yourself or only your front end dev. So if we completely stopped working on this product right now It means that our profit would be at around 600$ a month.
However, as I said in the beginning, This is not the cost that you should be worried about. If you're just getting started and if you don't have funding development costs is where you need to really pay attention. So, for the development costs, we keep track of all our budgets in notion, so I can tell you exactly how much it took us to build this version.
The last release was on April 22, so as you can see it took us 8000$ to develop this product up to that point. And the net volume on stripe so far is 6000$, so with all the costs and everything combined we are now at a loss of around 3000$ at the moment. I am sure we could have taken this product much further if I just posted more videos, but the primary reason we are still so low on revenue is because 95% of our active users are using this product for free.
And I know there is a lot of hype in software right now about freemium models, however, if you don't have funding, if you are bootstrapping your SaaS yourself, you really can't afford to spend a year building a product without revenue. So, the final lesson I wish I learned earlier is that you should not consider a freemium model if you are bootstrapping a SaaS. This is what we are going to do in our next huge update, which should be already be out by the time you watch this video.
So the bonus tip I guess for you is that you need to have a very large vision when you are building a product. Otherwise, there is absolutely no reason to starting a SaaS. Don't be swayed by those microSaaS trends like I was.
They may be "micro," but they still require a big of effort. There are so many other business models that can make a lot more money with a lot less effort much faster. The only reason for building a SaaS is if you are planning to go big.
So your vision needs be aligned with that. This is what I am going to do now, no more micro saas, now this product is going to be called Agency AI, and it will be fully based on my open source agent framework to become the only platform for deploying reliable AI agents fast. It is now fully based on my open source framework, and it is designed for developers aiming to help businesses unlock a new level of AI automation.
We added a ton of new features and improvements like advanced tabs for widgets and custom GPTs, streaming support, advanced parameters for agents, more models, API integrations, Slack integrations and skool community. I'll probably be doing a whole separate video on that later, but the price for this product is now $79 per month, no freemium. We will now be doing releases weekly, so if you want to become an agent developer and you want to support us now, checkout the link below.
However, we are only opening 200 spots at the moment because we want to offer the best possible support. Once we open more spots, we’ll send you an email. Thank you for watching and don't forget to subscribe.
