User Experience as a Moat
I have a habit of creating a photo book every year for our family — and also an individual one for any larger family trips we take. We just returned from a large family trip — and I finished the photo book for this trip in about 2 days. Most people will consider that a bit crazy (and it probably is). But I think there is also a lesson about product design that can be learned from my experience — one that will be applicable to the coming onslaught of AI companies.
There is one simple key for how I get these books finished quickly and consistently — I never let the perfect be the enemy of the good. I could use more complex layouts, do more post-processing to make images look better, etc and improve the quality of the book. But I know the best feature of the book is that it’s finished. Good and done beats better and in process.
To build these photo books, I consistently use Google Photos. I know there are services with higher quality prints and better layout options. But the key for me is making it easy. With Google, I can make an album for my trip and with one click it will turn it into a book. I can mess a little bit with the layouts and add some captions. But mostly I leave it alone. Because my key objective is getting it done — not making it perfect.
This focus on making things easy for the user is an often under-appreciated super-power for software. Some consumer software companies have learned this lesson — though by no means all of them. And poor user experience is a cliché for enterprise software (though some newer B2B SaaS companies are doing a better job here). However, in general — most software is still built for people who already know how to use it.
This idea of the power of simple user experiences was driven home to me recently when I was talking to a friend about the moats available to AI companies. I think many people in the AI space are focused not on building a technical advantage or a data advantage. Surely some form of better training data and more sophisticated models will win the day. But I’m not so sure. Look at some of the most recent examples of technologies that captured the imagination of consumers.
First, we can look at ChatGPT. The underlying language model GTP-3 was nearly 2 years old when ChatGPT came out. And while some in the tech community were enamored with it, the challenge in leveraging the model in a consistent way meant that few outside the tech community saw the power of the model. However, when it was wrapped in a chat interface, it became an overnight phenomenon. Simplifying the user experience was key to getting people to really leverage it.
Another example would be GitHub’s Copilot. This product also leverages GPT-3, but in this case with an additional layer of training and the specific goal of helping developers write better code. Here, GitHub integrated the experience into the development environment itself as an auto-complete option for developers when there is a sufficiently high confidence guess of what comes next. I tried Copilot and found it amazing — particularly when I could just write a comment of what I wanted the code to do and Copilot would suggest whole functions to achieve my objective. It’s truly mind-blowing. While we can use this as an example of one effective way to leverage AI to increase the capabilities of human creatives (vs replacing them), my point today is simply that there was tremendous power (and competitive advantage) in developing the right user experience around the model.
My final example is the mobile app Lensa. Lensa took an open source image generation AI model called Stable Diffusion and used it to generate AI-drawn avatars for individuals. They even charged a few bucks for it. Of course, anyone could do this for free by just using Stable Diffusion directly. But speaking as someone who spent several hours trying to install Stable Diffusion (unsuccessfully — apparently it’s not easy to do an an Apple Silicon Mac…), I can attest to the power of making the model easily accessible to users.
While so much development is happening in state of the art model development, it’s easy to think that the companies that end up as big winners in this space will have the most sophisticated models or the biggest proprietary training data set — it’s important not to underestimate the advantage that accrues to those companies that build the right experiences on top of the models. It may not seem like nearly as defensible a moat, but given how few companies really excel at building user experiences and how different a skill set and mentality it takes than building AI models, I believe such differentiation can be just as valuable.
