The Shooting Star Startup
We built a speech-to-text summary startup that went viral, but our success didn't last for long.
Introduction
Shooting Star Startup: a small enterprise that experiences the peak of it's success upon launch, then begins to fade away.
This is the story of my own Shooting Star Startup, as told in 4 parts:
I. Feather's Origin
II. MVP to v1.0
III. Shutting Feather AI Down
IV. Why Shooting Stars Aren't All That Bad
My objective for this piece is two-fold:
1. Help folks learn from my experience.
2. Encourage folks to feel more comfortable with failure.
Bonus: Entertain you along the way with some details on just how whacky this journey was.
Hope you enjoy!
I. Feather's Origin
It started with Ben's Bites Hackathon. Ben Tossell is an entrepreneur and the creator of Ben's Bites, a daily newsletter on all things AI.
In December 2022, Ben announced that he would host a hackathon. Teams had 1 week to put an AI solution together. Judges would then pick the top 3 projects and hand out cash prizes to the teams.
I immediately reached out to my former colleague, Dave, to collaborate. How come?
Dave and I had built successful AI products in the past, we knew each other's skill sets, and we knew how to effectively work with one another asynchronously.
Dave did the engineering. I do the product/design. We're both creative in different, complimentary ways.
I went to Dave with an idea that solves a problem I myself have as a podcaster - writing the show notes.
Enter Feather AI - an audio-to-text summarizer.
The biggest pain when producing a podcast is trying to type up a summary of everything that happened in the episode.
I thought to myself, "What if you could summarize a conversation so that it could serve as the 1st draft of your show notes?"
That way, you could decrease production time and become a more prolific creator.
Suddenly, more use cases past podcasting emerged:
- What if you were a government consultant who wanted to summarize congressional hearings so that you could stay on top of legislative affairs?
- What if you were an investor who wanted to summarize earnings calls/fed meetings so that you could stay on top of earnings/interest rates?
- What if you missed a zoom call but wanted to read a quick summary of what was said so that you didn't have to rewatch the entire recording?
- What if you missed a webinar but you wanted to read a quick summary so that you could still extract value from the speaker's session?
Time to take Feather AI to Twitter.
We didn't place in the Hackathon, but we wanted to share Feather AI with more folks than just our family, friends, and fellow Hackathon members.
I thought back to the whacky thing I did this past November: I published a thread on twitter every single day.
This exercise taught me how to write engaging tweets that tell a good story, so I applied this skill to Feather AI and published a Twitter thread on our nifty little product.
The thread went viral. 600k+ saw it, 125k+ watched our demo video, 1k+ signed up for our beta.
The product kept breaking due to high usage. We kept fixing it. It was awesome.
First, the servers were overwhelmed (so we upgraded them.) Then, one of the components of the AI model gave out (so we rebooted it.) In comes the server again with another issue (so we performed another upgrade.) Then, gmail capped our account for sending too many emails (time to make a new email address.) Up. Down. Up. Down. Up. Down. Up. Down. Up.
At it's peak, we were processing 4 summaries per minute. It was so dang cool to see people across the world engaging with a product that was just an idea a week prior, and the feedback was incredibly positive too.
We even did our best to contact everyone who made a summary request when Feather AI was temporarily down so that we could: 1. Thank them for giving us a try. 2. Apologize for our lackluster performance and 3. Invite them to give us another shot.
Biggest takeaway from those 24 hours - ready-fire-aim.
Ready, aim, fire? No - "ready, fire, aim," according to Michael Masterson.
Masterson is a successful entrepreneur who argues that balancing calculated risks/quick decision making skills + careful planning/intentional strategy is the key to business success - not one or the other but both.
We had planned, strategized, and prototyped all week long, but this product wasn't going to improve itself from the sidelines.
Did we stress test our product? Yes. Did we think it'd get this much traffic? No. Is having "too much traffic" a great problem to have? Yes, and we met the moment.
David and I had to fire in order to uncover the server issues, API problem, and gmail cap so we could then aim in on them quickly. The result? We captured 1k+ beta users in our first week.
I didn't know where Feather would end up, but I knew that building, adapting, and surviving was fun and insane in all of the right ways.
We didn't know it at the time, but this was Feather's peak.
II. MVP to v1.0
Fast forward a month later into January 2023. It's time to turn the MVP into v1.0.
Now, I’ve gone from minimum viable product (MVP) to v1.0 a few times in a corporate role, but this is my first time doing it as a bootstrapped founder.
Here are the differences.
MVP to v1.0 in a corporate role:
- Pros: resources to lean on (financial, intellectual) means that neither cost nor lack of knowledge stand in your way
- Cons: access to resources makes you more comfortable and less creative in finding ways to make things work, decisions require layers of approval which means that you won’t be able to move fast
MVP to v1.0 in a bootstrapped startup:
- Pros: decisions that take weeks in a corporate role take just minutes in a startup, you can accelerate feedback loops much faster, and you can change direction much quicker. Plus, having a user find value in the product you willed into existence is 100x more fulfilling than pushing a product you don’t care about.
- Cons: resources are tight, so you’ll need to teach yourself more than just your domain. You have to become a generalist (legal, accounting, marketing, etc.) Things are also much more volatile. It’s a rollercoaster of emotions. One minute, you love what you created. The next, you think it’ll never work.
- Yeah, but those things above aren’t actually cons: I’m here to learn about entrepreneurship. In fact, documenting how to build cool things was the impetus of my podcast and this newsletter. So, shoot, I’m not complaining that I have to teach myself new things on the fly. It’s fun (this includes learning how to manage the rollercoaster of emotions - lol).
*Just right* doesn't exist.
Whereas a corporate entity would want to fine tune things until they’re *just right* for launch, a startup puts stuff out there as soon as possible (before it’s fully ready tbh).
There are no rules for a startup (other than build a product that users love). The only way to do that is iterate.
Take the newsletter I sent on 1/26/23 titled “We Need Your Help.” Technically, this was a soft launch. We needed family and friends to test what we had.
We had centered the entire Feather experience around a Chrome Extension - great for avid Chrome users, not so great for folks who aren’t.
Our family/friends launch taught us that we needed to take a step back and bring the "Paste a YouTube Link” functionality from our MVP back into our v1.0.
Basically, relying solely on the Chrome Extension prevented users from getting to an “Aha” moment quickly (when folks recognize value in the product).
Solution: we built My Feather - the one stop shop to generate summaries. When you registered, you received a personal page to generate summaries by pasting a YouTube URL, uploading an mp3, or downloading the Chrome Extension.
We launched My Feather on Tuesday, 2/21/23 to our Beta email list (see “Feather AI Is Here”). I wouldn’t call it a hard launch as folks sorta/kinda new about us - more of a warm launch, if you will.
The point? Launch. Learn more about what the user needs. Build what’s needed. Get ready for the next launch. Lather. Rinse. Repeat.
Yet, as we uncovered user needs, we discovered a hard truth - Feather wouldn't be needed for much longer.
III. Shutting Feather AI Down
We proved that podcast summarization was a service that the market wanted, but we quickly realized that we wouldn't be the ones to deliver this service.
The message we sent to our users breaks down why we closed our doors...
"Here’s the bottom line up front:
AI products moved so quickly that Feather AI was no longer needed.
More on all of this below, but before all of that we want to say this - thank you.
We deeply appreciate your willingness to give us a shot, and we’re going to steer you towards the proper tools below to ensure that your summarization needs are met.
What happens next.
To our monthly paid users: your last month of billing will be June 2023.
To our annual paid users: we’re sending you a prorated refund for the remainder of your 12 month plan that won’t be active.
To folks still in need of audio/video summarization services:
- Sign-up for Google Bard (it’s free)
- Prompt it to summarize any YouTube URL
- Ask Bard follow-up questions on your summary in real-time
More info. on why we’re shutting down.
AI is getting commoditized, and we can’t keep up.
Here was our process at Feather:
- Create a transcript of your audio/video file
- Summarize the transcript with a Feather-specific template
- Send you this summary via an email service provider or via Notion. The automated process took 2-10min (depending on our system’s usage/the size of the audio file).
We uncovered that users primarily used Feather to summarize YouTube videos. So, when LLMs like Google Bard got access to the open internet, our main use case had been solved for.
Now, you can use an LLM to access YouTube, pull transcripts, and generate summaries of videos in real-time.
Are the summaries that LLMs generate perfect? No, but here’s why summarizing audio/video with Bard is better than Feather:
- Faster processing time
- You can ask the chatbot follow-up questions about the summary
- And the most important point - it’s currently free (and will likely be low cost if they decide to charge in the future)
The commoditization of AI is a good thing. It means that the capabilities at your fingertips are going to continue to increase, and the costs are going to get driven down.
This also means that Feather is no longer needed ¯\_(ツ)_/¯.
Our goal was to build a product that users loved, and we achieved that.
150k+ watched our demo video. 2.5k+ of you became Feather users. We proved that the world needs audio/video summarization. Plus, we learned a whole bunch too.
So, thanks for giving us a shot. We had a lot of fun, and we’re grateful to have had the opportunity to serve you.
🪶 Wishing you the best in your summarization journey,
IV. Why Shooting Stars Aren't All That Bad
When we first set out, my co-founder and I thought this was a 12-18 month window of opportunity. We knew that additional many more major players - dare I say "Big AI" - were coming.
We just thought it would be a little while until they arrived.
We were wrong.
LLMs are moving faster and faster. They don't solve perfectly for our use case. Bard still has some hallucinations when you provide it with a YouTube URL to summarize.
Regardless, I trust that Google's team of talented engineers paired with their full/complete access to the YouTube libraries will make for a strong YouTube summarizer.
In fact, I would expect that YouTube will soon have Bard capabilities that are native to the platform. Features like "Auto-generate a description for my video" for creators and "Auto-generate a guide to the video" for consumers could be commonplace.
Yet, for a moment in time, we had the momentum and the confidence that this could become a sustainable business. Sure, it was fleeting. Even still, that shooting star moment in and of itself is an incredible experience.
Our shooting star taught me a few things about building something from scratch:
- You have to be consistently vigilant and resilient to remain in the game.
- #1 is exhausting, so folks who aren't committed get washed out.
- The market is impartial. If just wants what it needs as conveniently and affordably as possible.
- Your first hypothesis is rarely (if ever) right. The same is likely true for your next 50.
In all, our shooting star lasted from December 2022-July 2023 (with it's peak in December).
Wins:
- 2500+ beta users
- Turned MVP into a *barely* profitable product
- New B2C marketing experience for digital products
- Gained the support of strangers from across the internet
Losses:
- Didn't turn it into a sustainable business
When folks would ask, "How's the startup going?", I'd say the following with a smile: "This turned out to be a 4-6 month opportunity, and we're in month 7."
My words didn't quite matter when I informed folks it wasn't going to work out. My attitude did.
If it doesn't work, it doesn't work. ¯\_(ツ)_/¯
"Sometimes you win. Sometimes you lose. Sometimes it rains." - Bull Durham
"Sometimes a shooting star rips across the sky and flares out fantastically despite your best efforts to keep it alive." - Me
Win. Lose. Rain. Shooting star.
You'll experience each of these at some point if you're living a life well lived, and the losses will teach you more about your character than the wins. That alone makes it well worth it.