Case study · Ecommerce · Name withheld under NDA
0 to AI-native in 6 months
When we started, they were a top 5 ecommerce brand in their category by units sold, with ChatGPT open in a tab, a stock sheet that was always out, and a couple of hours every morning spent copying numbers between platforms. Within three months the whole team was up and running on Tobin OS, our AI operating system. Then we built autonomous agents to run the work daily, weekly and fortnightly.
How the work moved
Their tools
- Online store
- Marketplaces
- Creator data
- Product files
- Project board
- Team chat
- Call notes
Tobin OS
On Claude. Knows the products, the creators and the house rules.
The whole team works in it.
Runs on a schedule
- Creator reportEvery weekday
- Marketplace briefEvery weekday
- End-of-day listEvery weekday
- Stock forecastEvery week
- Ad reviewEvery week
- Winning ads checkEvery week
Where the hours went
Morning data
About 10 hours a week
A couple of hours every morning copying numbers from each platform into a spreadsheet.
None
A verdict on every creator is ready first thing, every weekday.
Marketplace work
Most of a day
One job could take up most of a working day.
About an hour
The same job takes about an hour in one chat.
Creators
The biggest few
Only the top creators got looked at, and a breakout could go unseen for days.
All of them
Every creator gets a verdict the next morning, small ones included.
Stock
A guess
A spreadsheet that was always out.
A forecast
A weekly forecast across every channel.
The tools
A tab
ChatGPT in a tab.
A system
Tobin OS, connected to the tools they already use and owned by the client.
How we did it
We mapped the work first
The goal from day one was a system the whole team could run the business from. So we started by watching. We spent a day in their office, then broke every workflow down: where the hours went, where work got stuck and where AI would pay back most.
We installed Tobin OS
Tobin OS is our AI operating system, and it runs on Claude. We set it up for the whole team. It knows the products, the creators and the house rules, and it connects to the tools they already use. Within three weeks the whole team had moved over, and they cancelled ChatGPT.
Checked by hand first, then autonomous
Each job went to the person who owns it, who checked the output by hand until they trusted it. Only then did it run on its own. In our experience, that order is what makes automation stick.
What the agents do
Creator report
Every weekday
Gives a verdict and a next action on every creator, small ones included, before the team starts.
Marketplace brief
Every weekday
Turns the daily marketplace report into a short brief on what moved.
End-of-day list
Every weekday
Gives the founder a list at the end of every day.
Stock forecast
Every week
Looks at what sold on every channel, the lead time and the stock on hand, and forecasts what’s needed.
Ad review
Every week
Reviews the ad account and proposes changes. A person signs them off one by one.
Winning ads check
Every week
Finds the ad creatives that are winning and flags each one once.
In their words
“I think what you’re doing is brilliant. Going into businesses, seeing their entire model and then being able to help them come up with ways to make it better.”
“It looks at how much we’re selling in a week on [every channel], then the lead time and how much stock we’ve got. It’s actually fully automated, which is class.”
“The workload I’ve taken on is more, yet somehow it’s less, because of everything that’s in the background.”
Where it landed
By the end of the build, six agents ran on a schedule and nobody had to start them. The client owns the whole system.
If your team has AI on every desk and nothing that runs on its own, let’s talk.
Our business is growing your business.
A family business, run by the people you’ll work with.
Book a call