Happy Sunday and welcome to InsideAI weekend! I'm Rob May, Founder/CTO at Dianthus. I'm also an active angel investor in the AI space, so send me deals if you see good early stage opportunities. As I announced previously, after 6 years here I'm starting to wind down and write at "Investing in AI". You'll see bi-weekly posts from me here for the rest of the year, and then in 2023 the new newsletter will be my focus.
This week I want to talk about what I call the "Business Limen Problem" in AI. In science, a limen is a threshold at which a stimulus is perceived. Below that threshold it isn't noticed. Above that threshold it is. While we don't talk about limens in business, they are in a sense there for new technologies. A new technology that is below the threshold of perception is not very useful or appreciated, so, when deploying new tech we want to make sure we cross the limen.
My latest startup, Dianthus, was founded on the principle that in AI, limens are hard to breach in a short time frame. There are so many massive opportunities to apply AI in e-commerce, and we see 3x increases to EBITDA in as little as 18 months in the companies we acquire, but, for the first 6 months it's often slow because deploying AI models is hard, and it often takes them a while to become effective.
I've written before about the PAC Framework, how to apply AI in a business context, but today I want to add something to that framework - the concept of business limens. Where in the business can you apply AI so that the stimulus is perceived and people are excited about what AI can do?
You need 3 things:
- Data (we all know this one) - without enough data to train a model, it won't work well and won't have an impact
- Patience (this is the hardest one) - AI models learn through more and more data, interactions with the world, and corrections from humans. They rarely perform well out of the box and often need to be tweaked. I'd give an AI project at least a year before deciding if it's effective.
- Performance - The model has to have an impact. If you are trying to use AI to optimize your pricing, for example, you might have all the data and patience in the world but if it only moves the needle on EBITDA by 0.2%, it won't break the business limen. Sometimes data doesn't have the predictive value you think it does. And other times a process is already pretty optimized and AI can't add much.
In my view AI is still woefully under-deployed compared to what it is capable of, at most organizations. Breaking through the business limens and showing how this stimulus can prove big performance increases is important. I've seen it work in the companies we have acquired here at Dianthus, and in many of my AI investments. Think about the best places for impact and choose your projects wisely.
Thanks for reading.
@robmay
Comments
Post a Comment