What I learned about genAI from failing to scale my AI startup

In November 2022, the launch of OpenAI’s ChatGPT captivated the tech world. It was one of those rare moments in which everyone in tech viscerally knew that the course of our industry had changed irreversibly.

I’ve only experienced this feeling on two other occasions: when the first iPhone launched, making touch screens usable, and before that, when I downloaded a movie over a broadband connection in an unimaginably short time compared to what my 56-kbps modem could manage.

Image credit: Timmy Loen

When a shift of this magnitude occurs, people of all stripes converge on the opportunity like moths to flame. I jumped onto the bandwagon with a side project, Scribbler, a podcast summarizer I spent the better part of 2023 working on with a friend.

The project is now on autopilot and earns about US$500 a month. That may not be a lot, but the lessons I learned are worth far more and instructive on the future of the generative AI industry.

SaaS redux

The most important question any entrepreneur has to answer is, “How do we make money from our product?”

Since we wanted to explore the possibility of turning Scribbler into a full-time gig – preferably without venture capital – we decided to assess the viability of profiting from our product, so we charged a fee for it from day one.

Our first instinct was to fall back on the time-tested subscription model. We wrote the code and started to charge our users US$10 a month for unlimited use, a pricing model consumers typically expect from the average software-as-a-service product.

This was a mistake. We severely underestimated both usage patterns and the eye-watering cost of computing needed to power today’s generative AI products. To make matters worse, these costs increased the more the product was used.

To illustrate the point, let’s use Scribbler and some napkin math.

Variable costs:

• Transcript: ~US$0.25/summary
• Summary: ~US$0.05/summary

Fixed costs:

• Hosting (UI): US$20/seat/month x 2 = US$40
• Hosting (back end): ~US$50/month
• Database: US$20/month

AI is a commodity

LLMs are building blocks