An AI systems studio
We build the agents that watch your Shopify and Amazon orders, write your ads and reconcile your marketplace finances, then hand you the two decisions a day that actually need you.
Book a free ops teardown01
Seven jobs that repeat, and one person they all wait on.
One day6:00 AM
A typical day0decisions waiting on you
Support tickets
Forty tickets. Thirty five of them are the same four questions.
02
Eight finished frames from two brands, and the strategy underneath them.
Your designer can make twelve ads. Finding a winner takes more tries than that.
Creative is a queue in the operating layer like any other, and it backs up like any other. A batch is not a hundred versions of one ad: it is the same product argued to a different woman, on a different channel, for a different reason.
03
A live account, read every morning by something that is not you.
The campaign has been delivering to nobody since Tuesday. You find out on Friday.
A matcha brand on Amazon and Shopify. The agent reads the account every morning, decides what is a real problem and what is a reporting artifact, and speaks only when a human has to choose.
An integration that fails surfaces as an alert, not as silence.
Four notes from the live account, set as field reports.
Session data gap: all three SKUs showing n/a sessions. This clears as Amazon traffic reporting catches up, typically 24 to 48h post-listing activation. Not alarming yet; flag if still n/a tomorrow.
Whisk and ceramic holder velocity: both show n/a days of runway because units sold is zero. This is a data gap, not confirmation of healthy stock.
Delivery check today is the single most important operational task.
Approve, or reject.
Nothing publishes until someone answers. Raised three times, and it has been waiting fifteen days.
Amazon's ad reporting is not a source of truth. Spend and sales figures get revised after the fact, attribution is dated to the ad click rather than the purchase, and the console deletes granular daily data after 60 to 90 days depending on the report.
On one account we found three weeks where the numbers had quietly stopped updating. Nothing errored. Every report still returned a figure. It was just the same figure, and nobody noticed until we went looking. Three weeks of recommendations were made on data that had stopped moving.
The system now reads and stores the account every day, keeps its own copy, and says out loud when the numbers disagree with themselves.
Option B (benefit) balances brand voice with benefit-forward hooks. Default recommendation.
Option A (feature) leads with tangible product attributes. Strongest fit where features rank higher in conversion.
The agent does not just produce options. It says which one, and why, per channel.
04
Reconciliation. Entirely rules, which is why it goes first.
A month of settlement lines, and your accountant is waiting on you to classify them.
Marketplace tax reporting is a monthly reconciliation nobody wants and everybody has to do. It is also entirely rules, which is what makes it the first thing worth handing over.
40+ hours of manual work a month, off the founder's desk and off the bookkeeper's invoice.
05
Your hours, your rate, your arithmetic. No benchmarks.
Nothing ships without you, which means nothing ships while you sleep.
We are not going to forecast your return from another brand's numbers. Put your own hours in and do the division yourself.
These are outcomes from systems in production, not projections.
| Outcome | Before | After |
|---|---|---|
| Owner oversight | Constant | Exception-only |
| Creative production | One at a time | 134 variations per batch |
| Manual reconciliation | 40+ hours a month | Automated |
06
An agent with no stated limits is a liability. These are the limits.
Nothing ships without you. That stays true, on purpose.
Four rungs. Everything we build sits on one of them, and you decide which.
Reads the account and says nothing unless something is worth saying. The daily memo lives here.
Names the option it would take and why, per channel, and waits. It does not act on its own recommendation.
Builds the thing and holds it. Creative is rendered, copy is written, the queue fills up. Nothing is published.
Only after a person answers. Approve or reject, by name, on the specific item. There is no rung above this one.
You do. The workflows, the prompts and the account configuration are yours, and they leave with you.
On your infrastructure and your accounts wherever the platform allows it, so nothing depends on us staying in the room.
An integration that fails surfaces as an alert, not as silence. The system is built to stop and say so rather than continue on stale data.
Documented, and walked through with whoever will hold it. If your team can't run it without us, it isn't finished.
Good fit
Not a fit
07
Three ways in. Fixed price, scoped before anyone starts.
One workflow, scoped and shipped. The bottleneck you already know about.
Live in 1 to 2 weeks
From$1,500
A single agent in production, with approvals, error handling and monitoring.
2 to 4 weeks, depending on how many integrations it touches
From$5,000
The full system: orders, inventory, ad creative and the daily memo.
First monitored workflow live in week 2. Whole system by week 4 to 8.
The range is how many systems it touches, agreed before anyone starts.
Range$10,000to $18,000
Fixed price, scoped up front. No retainer to start.
If your bottleneck isn’t one of these, the teardown still applies. Most builds start as something not on this page.
08
A studio of one who ships, not a roster you are assigned from.
A note from Shah
Every system on this page started as something I was doing by hand for a client, badly, at the wrong hour. The agent exists because I wanted the evening back. It turned out the founder wanted theirs back more.
You are not buying a platform, and you are not being assigned an account manager. Whoever scopes your build is the person who builds it, and the person who answers when it escalates.
Book a free ops teardownShah DurranFounder
How I work
A client asked for a deliverable I took the money for and then could not build the way it needed to be built. I credited it back rather than shipping something that would have technically satisfied the brief.
Another wanted to buy a tool I would have been paid to integrate. It would not have earned back what it cost them at their volume, so I said so, and we did not buy it.
When a client's numbers and my own database disagreed, I went and checked against the live platform instead of defending my figures. Mine were the ones that were stale.
None of those were good for the invoice. All three are why the same people keep coming back.
nootri and Mouchiie, whose systems and creative appear above.
A free ops teardown is one call. We map what is worth automating first and roughly what it returns. If nothing is worth it, we say so.