Ryan Petersen explains how he'd build Flexport differently with today's AI tools. He'd invest earlier in accounting to understand unit economics, prioritize workflow automation, and use LLMs to improve margins by roughly 35% on the net revenue line.
Did you want to come on? So I raised your hand.
I wanted to dig into something that you said earlier on in the conversation. You were talking about trying to keep the team as small as possible for as long as possible. If you were to start Flexport again, given all the tools that are available now, these AI, like every day a new AI tool comes out. Do you think you'd Do you, would there be any specific team that you think you'd now be able to keep much smaller? And how do you think, you know, the tools that you have at your disposal right now would change how you'd scale Flexport?
Yeah, to be honest, a lot of it is probably investing more in the accounting function, like I mentioned earlier, so that you can really drill into the unit economics. I think the, and then it's more where do you point your engineering effort? towards automation of work. We really, we built this like crazy powerful workflow platform to manage the transactions and then automate them. But we only got really serious about that in 2019 or 2020 even. So your first customers came in 2014 and for the first 6 years we were just focused on the user experience and it turned out A lot of the problems in the user experience come from lack of standard workflow and structured work and data there. So it's more about like, where do you point your resources? I do think we'd have better unit economics now and less people if we'd started that. LLMs are incredible for this, by the way. We're finding, I think we're gonna be able to improve our margins by like 35% or something under the net revenue line. have 35% more of it turn into profit over the next 2 years just from being able to use AI to automate the workflow. But the early tests of this are really crazy. So yeah, I think you could build a better version of Flexport with AI. I mean, we're going to build a better version of it, but if we'd had that, if only we'd invented large language models and had the NVIDIA H100 10 years ago, we'd have been a great business.
