LLMs and Self-Storage - similar businesses
Last week ChatGPT rolled out improved memory to many of its users* - with ChatGPT able to remember useful information across chats including your likes, dislikes, places you’ve been or the car you drive.
Yesterday Grok also released their memory capability to many users*.
*(I say many users because for those of us in the EEA, UK, Switzerland, Norway, Iceland, and Liechtenstein we are ‘protected’ from these technical advancements for the time being. But at least we have sticky bottle tops!)
The business of self storage
A few years back I was watching a documentary about a self storage business.
The owner was describing why he loved the business model - because the friction for a customer to cancel was so high.
“If you want to cancel you need to do something.
You need to figure out where else you are going to put all of this stuff.
And then you need to move it.
For most people the effort is just too much and they keep paying.”
I’ve always remembered that, as it was such a good physical embodiment of switching costs.
Memory becomes the moat
Until recently there was zero switching cost from one AI model to the next.
I can pop the same chat into multiple models at once and see which response is best.
In Cursor I can flip from one model to another if one starts lagging.
At a company level we could switch from one model to another to get better API rates.
“LLMs is a race to the bottom” people would say. “There is no moat".”
Memory starts to change that.
Compare two conversations:
You meet up with your good friend that you went to school and have stayed in touch with.
The conversation flows, you stick to topics you are both interested in, and reminisce about ‘that’ trip to Weston-super-Mare!
Later on you find yourself at a work conference sharing a lunch table with someone you’ve never met.
The conversation is stilted - you jump from jobs, to where you live, to family as you try to find some common ground. Soon it grinds to a holt as one of you picks up your phone.
Of course the first conversation is better, and more enjoyable for both parties.
Early feedback from ChatGPT users with memory enabled is that the conversations feel more human, they feel more like speaking to that good friend and less like the forced networking at the conference.
With the models themselves now indistinguishable from each other in terms of intelligence for most tasks, it is the ‘feeling’ that will be the differentiator for most users.
I like speaking to this model more than that model, in the same way we like speaking to this person more than that person.
Memory is your ‘stuff’ in storage
Tying this back to the self storage business.
Every time you speak with ChatGPT, or Grok, with their memory enabled, you are putting something in storage.
The tech you were researching
The calculation you were trying to solve
The employee you were hiring
For you to switch models means having to ‘move’ that memory into another provider.
Or just leave it there and return to the ‘forced networking’ conversation.
For many the friction will be too much, and it is just easier to keep using the model that holds their memories.
It will be interesting to see how this plays out - especially in the context of organisational accounts where the user’s memories are ‘owned’ by the company.
Get Started
Whenever you are ready there are three ways I can be helpful:
Model 101 Playlist: 20 minute guides to using ChatGPT, Claude, NotebookLM for work
AI Inspiration Briefing: Show your people the path forward in this 90 minute live session
Kowalah: Buying platform to help you pick the right AI tools for your business