3 things I learnt from my failed ML developer tools startup
Updated: Aug 3, 2020
I have been a startup founder for the past seven years. As a founder and an early startup employee, I have seen my fair share of ups and downs, have made a whole bunch of mistakes, & learned a lot in the process. Over the past few years, I have noticed that most of the first time founders make the same mistakes. Hence, I will be sharing my learnings with everyone that might help founders avoid these mistakes and guide them in the right direction.
Ridingo, TMinus, ModelChimp & T.I.T.O are the 4 different starups that I have worked on in the past. In this blog I will be talking about ModelChimp.
Idea: ModelChimp is like GitHub but for Machine Learning/Deep Learning developers.
Problem: Building a Machine Learning / Deep Learning model is an iterative process. To build a production-ready model, an ML / DL engineer has to iterate through 100's of experiments. After interviewing 100's of ML/DL engineers, we realized that most of them use Adhoc tools such as log files, excel sheets, or build in house tools to keep track of their iterations that go into building the model. We at ModelChimp were aiming at dramatically streamlining the process.
How did we decide on the problem to solve?
Prior to working on ModelChimp, Me and Deepak were working on a Social Countdown App called TMinus. Since TMinus wasn't gaining traction as a Social App, we decided to shut it down and looking for problems to solve in 2019. We had a long list of problems to solve but here are the reasons why we picked ModelChimp to work on:
Personal Experience: Samir (my co-founder), being a Data Scientist, had experienced the problem first hand. While building Machine Learning models, he would use log files, excel sheets, or other ad-hoc tools to keep track of his experiments. His first-hand experience with the problem and also our interviews with other data scientists validated the fact that ModelChimp is a problem worth solving.
Funding Validation: Startups in USA & Isreal like Weights & Biases, Comet.ml, Missinglink.ai had raised tons of money and were solving the same problem. Also, TechCrunch press releases reinforced the idea that it is was a good problem to solve.
User Interviews: Prior to building the product, we put together mock screens and started interviewing developers. In these interviews, most of the data scientist we spoke to said yes that they do use ad-hoc tools to keep a track of their experiments. Hence, we thought our solution made sense to them.
Why did we shut down?
Personal Runway: After spending 24+ months with out a salary my bank accounts ran dry.
Paying Customers: After interviewing about 50+ customers and running a few POC's, none of our customers were willing to pay at the end of the POC.
Cash is king. The best way is to turn sustainable with the help of customer revenues. Raising money is hard.
Solve a local problem. We were building ModelChimp for developers based out of the USA, Canada, Germany, and countries where AI/ML startups are widely present. Even though we were a developer tool, the ModelChimp sales process was a typical enterprise sales process, where we often pitched to the CTO or the Head of ML. One of the advantages our competitors had was physical access to customers - which helped them gain trust. I think in enterprise sales - this is important. Thinking Global is good, but solving a local problem will help build that trust & get to market faster.
Don't blindly listen to feature requests. Your customer needs to use your product. While we were pitching to clients - they would always say that build us this "must-have-feature" and then we will use your product. That's a trap. Every single time we reached out to our potential client after building that "must-have-feature" they would stop responding to us. One way to mitigate this problem is getting customers to commit. If a customer is willing to pay for the "must-have-feature" then you know that you are solving a painful problem.
Survive. Survive. Survive. SaaS is a long shot - depending on the ticket size it will take you some time to get to Ramen Profitability. Unless you have raised money and have a substantial runway for a few years, it is important that you take up some consulting gigs, or build a services division on the side(but that shouldn't sidetrack you from your goal). You are typically looking at a 24-month runway to build a sustainable SaaS business.
What am I doing differently now?
Going after an neglected market / overlooked oppourtunity: Currently we are building a Shopify-like portal for a Marriage bureaus. There's no market leader and a standard product in the market that caters to these bureaus.
India First Product: For the past decade, Indian SaaS startups have mostly been building products in India and selling to clients outside India. But now there's a shift in the Indian SaaS ecosystem as well. Indian startups have started building for the local market. Some startups are InstaMojo, Limetray, GoFrugal, ClassPlus. One of the biggest advantages of building for India is access to customers, which makes it really easy to establish trust, sell and iterate over the product.
Get payment upfront to confirm the customer's commitment: Every time we now think of building a new product, our benchmark to write the first line of code is five paying customers. Before we wrote our first line of code for Jothishi Matrimony, we onboarded five paying customers. Having those paying customers gave us enough validation that this is a painful problem worth solving.
Not burning out: I start my day at 9am and end at 6pm. I'm not burning myself out, but preparing for a longer haul. It's like training for Marathon vs a 100 mts sprint. Now meditate twice a day, code, blog, cycle and also facilitate meditaion session post work from 7:00 pm to 7:40 pm everyday.
If you are thinking of starting up or have started up and would like to bounce off problems to solve, GoToMarket, get your pitch deck reviewed, B2B sales strategies feel free to reach out to me. I'm available on : https://wa.me/919986347648