A. OverviewB. Projects I Worked On1. Better voice analytics for call centers2. Automated bundling for websites3. Demographically-targeted out of home advertising4. AI-generated and curated content for trade mediaC. LearningsD. EF: What worked and what did not1. What was great2. What was suboptimalE. What's next for me?
A. Overview
I am an AI-developer who has been running a bootstrapped startup with around S$100k in annual revenue (Loki.ai) over the last 3 years. I currently have a small (S$4.5k/month) amount of recurring revenue coming from products that I developed at Loki.ai, but it is more of an umbrella company for housing my technology instead of a company with a focused vision or a path to generate double digit millions (or more) in annual revenue.
I had joined EF with the intention to start a new, ambitious company with someone who could help supercharge the commercialization of my technology. I worked with some incredibly smart and ambitious people during the program, but could not land on an idea that was big enough to excite me.
B. Projects I Worked On
1. Better voice analytics for call centers
The Idea
Voice APIs for call-centers in Asia that could understand customer tones, and use them to better inform call-centre decisions
The Team
I worked on this with Arjun Arora, who I have known since 2005 and has been a close friend throughout high school, college, and the years since. Arjun is an expert at call centers and had led digital transformation efforts for Singtel's call centers.
Why we didn't continue with this
We were addressing a shrinking industry instead of addressing something with real tailwinds (atleast in Asia). Enterprise spend on voice-based solutions was falling rapidly, as these were progressively replaced with chat-based solutions.
2. Automated bundling for websites
The Idea
AI-assisted bundling for eshops and ecommerce stores. We could make it easier for companies to upsell more products every time customers land on their website
The Team
I worked on this with my Arjun Arora, who I have known since 2005 and has been a close friend throughout high school, college, and the years since. Arjun is an expert at call centers and had led digital transformation efforts for Singtel's call centers
Why we didn't continue with this
This seemed like a great project for consulting with large companies, but would’ve been hard to scale as a product without deep integrations with individual companies. It definitely has potential, though. Will likely come back to this post-EF and try it out as a solo developer with other companies
3. Demographically-targeted out of home advertising
The Idea
AI-powered digital signage that automatically changes what it displays to customers based on how they look (demographics, dressing style, body type etc) – works for both walker-bys and in-store users
The Team
I worked on this with Jeremy Au, who had run a childcare company in the US for Nanny Sharing and was a former Bain consultant
Why we didn't continue with this
This didn’t seem like a very large business moving forward. Digital signage is a crowded space, and many people (including large local players in Singapore) were developing similar solutions
4. AI-generated and curated content for trade media
The Idea
AI that creates content for business executives, and delivers it to them in a bite-sized, TikTok-like format. To do this, it looks at data released from primary sources in real time, converts this unstructured data into data signals that an algorithm can understand, and then automatically creates beautiful data visualizations and motion graphics out of this data.
The Team
I worked on this with Reuben Noronha, who had previously led Financial Services and Monetization for Zilingo (between their Series C and D rounds), and was a Bain consultant before that. He had also co-founded a real-estate startup called Nivaasa
Why we didn't continue with this
I loved this idea, but this didn't seem to gel well with EF's approach or that of EF's partner investors (like SGInnovate) – it's dynamics were far more similar to that of a B2C company rather than a B2B company. Moreover, Reuben was great to work with, but I felt that he would not add enough value to this particular idea and was a little reticent to work in an arrangement where he would have half of the company's shares.
C. Learnings
Technical
- Edge-ML is going to become a really big trend. Devices like the Nvidia Jetson series bring GPUs to a RaspberryPi like interface, and are likely to rapidly grow the "Smart IoT" space in the coming months
- Transformers (the neural network architecture that powers GPT-3) are incredibly powerful beyond NLP tasks, and can add a ton of value in recommendation engines
- Voice AI is likely to be dominated by the giant Cloud providers in the near future. The amount of data needed to train voice models is non-trivial
Sales
- Products should be built based on System 2 thinking, but sales pitches should appeal to System 1 thinking
- Balancing between B2B Growth (open-source projects, freemium) and enterprise sales is a bit of a tightrope walk. Going for a sales-first approach can tie up company resources into few customers and can impede revenue growth But without "growth hacking"-like things (like a working product with some open-source support), enterprise sales becomes difficult as enterprises don't take you as seriously.
Personal
- Chase rational opportunities in fields that you are irrationally excited about. Don't just go for opportunities that sound great on paper but you don't have a personal interest in
- Do the math for evaluating opportunities early. If you can't ever see something hitting $100M in annual revenue, there's no point pursuing it as a venture-backed startup
D. EF: What worked and what did not
1. What was great
Big-picture: The EF team did a phenomenal job at communicating ideas in a virtual setting.
- The early modules around thinking through ideas, doing customer development team, and team-building were fantastic
- "Sharing-sessions" that featured former EF founders added a ton of value
- The EIRs (Didier and Shiu) were helpful, candid, and able to transfer their learnings from interacting with other teams over the years in a fairly seamless manner
2. What was suboptimal
- The camaraderie and serendipity that results from water-cooler talk and bumping into people in a physical setting was missing, though there's nothing that EF could have done about this
- A lot of EF's suggestions were optimized for companies that require a long turnaround time to build a product. The general approach (validate the problem, whether your solution can solve it, and how much enterprises would pay for it before starting to build a solution) was extremely well-suited to a company that would need many months to build their first product But it seemed to preclude "throw stuff at wall and see what sticks, then iterate quickly" kind of startups – these include developer tools companies, content-creator tools companies, and many more. This "one-sized fits most" approach introduced some unnecessary friction for teams that were working on ideas that had a relatively quick turnaround time for iteration
- Software and hardware startups attended the same sessions. It would've been more impactful to have different sessions for different kinds of startups
E. What's next for me?
I am going to continue running Loki.ai (which is still generating revenue and is on autopilot), and will also start two other companies:
- Fullstackdata.com: This will be a set of mostly open-source tools and frameworks for managing the data science pipeline (data strategy → data scraping → data warehousing → insights, predictions, and algorithms → delivery → data flywheels) I will monetize this in the long-run with managed tools, developer APIs, and consulting/training
- Vizer.ai: This will be an AI-generated and curated platform for business news (think TikTok for trade media). This will use some of the tools I develop for Full Stack Data (company listed above).