Metlo YCombinator Application

[Didn't have one at the time, sent an update later]

Both of us do all the work. No one else has worked on Supernote.

We've known each other for almost 4 yrs. We met in college where we both worked in the AI and Vision lab and quickly became best friends. We've also been roommates in SF for the last couple years.

We made a consumer subscription app called [REDACTED] in December 2020 because we thought that the space was super interesting. We learned iOS development, designed and shipped v1 in 1 week and then launched a TikTok Ad campaign. The app reached $8K MRR in mid-Feb but we got bored and decided we wanted to do something much bigger so we stopped working on it as much. It's currently running on autopilot, we haven't done any work on it since mid-Feb.

We came up with the idea ~2 weeks ago, have done 15 user interviews since then and should have our MVP up and running by Saturday (3/26) To validate the idea we also asked some of our users to pre-pay $50 (USD) to get lifetime access for Supernote and 3 of them agreed so we have $150 in revenue (bookings? Not sure how to account it, but money in the bank)

We had the idea ~3 weeks ago. Since then, we've done 15 user interviews and are on track to ship our MVP by 03/26 to our first few users. Akshay has been working on this full-time for 2 weeks. Shri has been working nights and weekends for the last 2 weeks and full-time for 2 days (all the PTO he could take right now).

03/26/2021

Feb 2021: $150 Jan 2021: $0 Dec 2020: $0 Nov 2020: $0 Oct 2020: $0 Sep 2020: $0 Aug 2020: $0

All our revenue is from 3 pre-sales for lifetime use of our product. It's obviously heavily discounted but we couldn't think of a better way to validate our idea and asking for more seemed like a sure-fire way to get a "No." Additionally, 1 out of the customers who bought it in the pre-sale is a close friend of Shri's.

We did not apply for the last batch.

We haven't participated in another "accelerator" or "pre-accelerator"

How we came up with this idea: Our goal for YC build sprint was to figure out what we want to make, launch a product and get 1 paying customer. We didn’t quite accomplish that but we spent a lot of time talking to our friends who work at smaller companies (Series A/B) and non-tech F500s (Boeing, Medtronic, UnitedHealth Group) about their workflow and problems. Among all our conversations (27 convos), 4 people who were Data Scientists at startups brought up that they “could be managing our notebooks much better”. Shri works in python notebooks a lot at Facebook so he knew that they had a pretty good solution to the problems our interviewees brought up and Akshay made a lot of the data infra for Uber’s marketplace experimentation team so he knew how to make the back-end. We went back to the 4 DSs and pitched them Supernote -- they liked it so we did 11 more interviews with people we knew. We asked them to pay us $50 for life-time access to the product. 3 out of the 15 people paid us (1 of them is a close friend though) so we decided it was a good idea and we should work on it. Domain expertise: Shri knows exactly what product we need to build because he’s a power user of Bento notebooks and ANP (internal tools) at Facebook which is what we used to validate our idea with customers. Akshay did a *ton* of data work at Uber and knows exactly how to make the back-end -- otherwise we couldn’t get our MVP up in 2.5-3 weeks. How we know people need what we’re making: As described above we validated it with a bunch of people and got some pre-sales. Additionally, Facebook and Uber both have fairly large teams that work on these powerful notebooks as internal tools so we’re confident that there’s a real need here.

What’s new: A lot of our features exist in disparate open source software but there’s no one tool that a Data Scientist or ML Engineer can use to manage their whole workflow. These features individually solve annoying problems but not big enough that customers would want to pay for them. However, we think that a bundle of all these tools (Supernote) is much more valuable and crosses the threshold to something people will pay for. Substitutes: The most common DS workflow today is training a model on local jupyter notebooks or google colab notebook (for more savvy people who work at companies that allow it), testing it out with an example set of arguments and then putting it into a python script that they then upload to GCP cloud functions/AWS lambda to deploy it. There’s no good collaboration solution so people often just share screenshots in slack. The DS’ who are technically savvy and understand git sometimes use git for versioning, most just don’t save versions over time.

We don’t have any real competitors right now except Deepnote (YC S19) and Naas (https://naas.ai). We don’t really fear them. In the future we think Databricks, Domino Data Lab, AWS, Azure and GCP could all become competitors. We do fear Databricks changing their notebook product to be exactly the same as ours -- if they do this, we’re not sure what we’d do because they own so much of the data infra in a company’s stack that it would be very difficult to compete with them.

We think most people in the space don’t understand how overlooked DS tooling is and that if you capture the DS notebook (home for Data Scientist/ML Engineer), you can expand easily into other areas of the data and ML pipelines. There’s a ton of focus on DevTools for devs, not much on tools for pseudo-devs (DSs) -- we think this is a pretty big opportunity that we can go after.

We’ll make money with a SaaS subscription by charging $20-50/user/month and then charging for notebooks deployed as APIs on a usage-basis. We think we could be every DS’ daily driver tool so we could make a lot of money. For rough market sizing, we’ll assume that we can charge $25/user/month. According to LinkedIn there are 2.6M Data Engineers, 350K ML Engineers and 400K Data Scientists in the US alone, all of whom use Jupyter Notebooks everyday. For our notebooks product alone, we could make 25*12*3.35M = $1.05B/Yr. But more importantly, after we are successful in becoming the daily driver for DS, DE and MLEs, we can sell an E2E Data platform to our customers later which is a *much* larger market.

Most of our users initially will come from sales and marketing but in the medium-longer term, a free tier for students and individuals will help us get much more organic and word-of-mouth growth similar to Qualtrics and Benchling’s early GTM.

No

1. Feature Store for ML 2. Consumer subscription app studio (many are very successful in Belarus, Israel, China, Florida, and Bordeaux) 3. Sourcing tool for recruiters based on data merged from GitHub and LinkedIn

1/ The vast majority of subscription revenue comes from people who forgot they were paying for a service and consumers’ willingness to pay for stuff is way higher than people give them credit for. 2/ Paid marketing is a *much* better channel than people give it credit for. It often costs a heck of a lot more to get word of mouth growth and “organic” growth than it does to just use FB or Google Ads for customer acquisition. Few companies like Netflix, Wish and Spotify understand this deeply; most others just complain about CAC and give up quickly.