Quick Thoughts on ChatGPT and Why It Matters for SaaS Startups
Yesterday I published a brief LinkedIn post with a few thoughts on ChatGPT and the implications for SaaS companies. The posting prompted (no pun intended) many interesting comments, so I wanted to re-post a (slightly polished and extended) version here.
As we enter month 3 CE (ChatGPT Era), I’d like to share a few observations and questions on the ChatGPT phenomenon and what it might mean for SaaS startups.
What happened?
The idea of an artificial neural network is 80 years old, so if ChatGPT feels like an overnight success, it’s one that’s been decades in the making.
Speaking about Large Language Models (LLMs) more specifically, GPT-3 has been around for 2.5 years, and while it attracted lots of developers and enabled the spectacular success of companies like Jasper.ai, its speed of adoption was nothing compared to the never-seen-something-like-this-before growth of ChatGPT (5 days to 1 million users, I’m sure you’ve seen the chart).
There were several AI winters and AI resurgences before, so you could argue that this is just another “up” in a long series of ups and downs. I’m aware that this time is different are considered the four most expensive words in investing, but I wager to say that this is the iPhone moment of AI.
A friend of mine wrote me, and I fully agree:
I feel the same excitement as in the mid-90s when the web came, and behind all the problems of low-res graphics, timeouts, and expensive connections, you could see a lot of opportunity coming. It feels like this will be touching the lives of so many knowledge workers behind computer screens these days.
ChatGPT’s main innovation is the user interface. That might seem strange since (a) there’s nothing innovative about an instant messaging UI per se, and (b) chatbots don’t have a particularly good rep. But as we now know, the combination of a user interface that everyone is familiar with and the power of GPT-3/3.5 enabled ChatGPT’s meteoric rise.
Great design and UX have been a huge part of the SaaS investment thesis I started building in 2008. It led to my investments in Zendesk in 2008, Clio in 2009, and later Typeform, Loom, Factorial, and many others. I’m glad to see that user experience still matters. :)
What’s next?
1) There are at least 30–40 writing assistants similar to Jasper.ai. My guess is that many of them are small money-printing machines right now, but in the vast majority of cases, I’m struggling to see how they’re different from all the other ones. When the market cools off, and pricing comes down, I think only a few will survive and thrive (driven by critical mass and economies of scale).
2) There are exciting opportunities for writing assistants in highly specialized areas like bookkeeping, legal, and health. It’s unclear to me (and way beyond my technical understanding to predict) if specialization in a certain domain will give startups a sustainable advantage based on better model performance or if future versions of GPT and its competitors will be so good that they’ll excel across all domains.
However, returning to the point above, domain expertise and the ability to deliver a fantastic user experience are crucial. Vertical SaaS companies (like Petsapp, Graneet, NexHealth, Amenitiz, Jobber and others from the P9 Fam) win by
- having a deeper understanding of the customers in their industry
- solving their specific problems in the absolute best way
- marketing/selling exclusively to a narrowly defined segment of companies
- adding more layers of functionality over time, increasing ACV and stickiness
As a lawyer, you could run your practice using generic project management and CRM software. That doesn’t mean you should, and it didn’t prevent Clio from getting to $100M+ ARR.
Similarly, SaaS startups will be able to win in (large) segments against the big foundational model providers if they leverage unique insights, data, workflow integrations, and design talent to solve the largest customer problems and deliver the best UX.
3) Every SaaS company right now is thinking about how to add a conversational interface (if not, I think they probably should). Chat won’t be the best UI for all use cases or all users. Depending on what the user wants to accomplish, different types of UI will be the best fit. And different users have different preferences. The point is, text-based input, intelligently processed by the software, will be an excellent mode for some use cases and some users.
For example:
- Instead of going to Admin, choosing Settings, clicking on People, choosing Alex, going to the Roles section, selecting Owner from a dropdown, and clicking Save, the user will simply enter “make Alex an owner” and confirm her choice. This is a very simple example, and you could achieve the same with a Cmd+K functionality and a decent auto-suggest feature. But using specialized LLMs, people will come up with much more exciting ways to let users control applications using natural language.
- Generating reports and insights. This is super relevant for all enterprise SaaS companies, and GPT3 can be a real game changer here. As some people rightly pointed out on LinkedIn, the challenge with GPT3 is that it frequently hallucinates. As an enterprise customer, you definitely want your (artificial) data analyst to be sober, so this problem needs to be solved before. As far as I know, most AI experts are confident that it won’t take too long until this is solved.
There will be a build-or-buy question for many companies: Shall we build our own conversational UI/AI/co-pilot, using OpenAI and open source components, or shall we buy a solution from a SaaS vendor? I’m curious if there’s an Algolia-like opportunity for a conversational-UI/AI-aaS. If you’re building it, let me know. :)
I know I’m only scratching the surface here. Incredibly exciting things are happening in everything from molecular design to creative software, and I’m sure we’ll get back to some of these areas in future posts.
