Case study · AI Agents & Automation
Creative decisions made on evidence, not whoever updated the account last
The brand's ad creative changed whenever someone got around to it, and the decision to keep or kill a variant came down to a feeling, not a comparison. This is a system that generates variants, rotates them against real performance, and surfaces the winner.
- Client
- a D2C retail brand
- Problem
- Creative decisions were made on a feeling, by whoever last had time to look at the account.
- Approach
- A system that generates ad variants, rotates them against performance, and surfaces the winner with evidence attached.
The problem
The brand ran Meta ads continuously, but the creative behind them barely rotated. A new variant went up when someone had an hour free, not when the data said the current one had run its course. Whether to keep an ad live or replace it was a judgment call made quickly, usually by whoever happened to be looking at the account that week.
Nobody disputed that testing more creative, faster, would help. What stopped it from happening was that testing took time nobody consistently had: writing variants, launching them, waiting for enough data, then reading the results against each other by hand. Each step was small. Doing all of them every week was not.
The reality behind the problem
The instinct was to hire more hands to write and launch creative. That adds output, not judgment. More variants launched on the same ad-hoc schedule just means more ads nobody is comparing properly, not better decisions about which one to trust with spend.
The real constraint was the loop from idea to evidence. A variant has to be written, launched, left long enough to mean something, read against its siblings, and either kept or killed, on a schedule, not when someone remembers to check. Skip any step and the brand is back to deciding on instinct, just with more ads to look at.
What was assumed
Testing more creative means writing more creative.
What Stallwart asked
What if the system ran the whole loop, idea to evidence, on its own?
The desired outcome
The target was not more ads. It was a loop that runs on its own, every week, without needing someone to remember it.
- Variants generated against the brand's existing angles, not a blank page.
- Each variant rotated into the account on a schedule, not an ad-hoc one.
- Performance read automatically once a variant has enough data to mean something.
- A clear keep-or-kill call, with the evidence behind it visible, not just the verdict.
- A person approving any variant before it goes live, and any decision to pull real spend.
The system we designed
The system drafts creative variants from the brand's existing angles and assets, launches them into the ad account on a fixed rotation, and tracks each one's performance against the others running at the same time.
Once a variant has enough spend behind it to read cleanly, the system compares it to its siblings and recommends a keep-or-kill call, with the performance evidence attached rather than a bare verdict. A person approves every new variant before it launches and every call to pull spend from a live ad.
How it works
- 01
Generate
Draft variants from the brand's existing creative angles and brand voice, not a generic template.
- 02
Rotate
Launch on a fixed schedule so testing happens every week, not when someone remembers.
- 03
Read
Compare performance once a variant has enough data to mean something, not on a fixed day.
- 04
Recommend
Surface a keep-or-kill call with the evidence attached, for a person to approve.
What changed
Creative testing now runs every week without anyone having to carve out the hours for it. The brand's team reviews a short list of evidence-backed recommendations instead of starting from a blank page or a gut call.
Decisions to kill an underperforming variant happen faster, because the comparison is already done by the time a person looks at it.
- From ad-hoc creative changes to a weekly testing cadence
- From a feeling to a comparison with evidence attached
- From one person's judgment to a documented, repeatable process
In practice
Marketing Lead, a D2C retail brand
The team described the shift as no longer having to defend a gut call about which ad to keep. The system's recommendation came with the comparison already done, so the conversation moved from opinion to a quick check of the evidence.
Ownership & handover
What you receive, and keep.
- The creative-generation and rotation system, yours to run independently.
- The performance-comparison logic, documented and handed over.
- Full account access stays with the brand throughout, not routed through an agency login.
What we learned
The bottleneck in ad creative is rarely ideas. It is the loop that turns an idea into evidence, and most teams do not have the hours to run that loop consistently by hand.
A system does not need to replace creative judgment to be worth building. It needs to make sure judgment is applied to a comparison that actually happened, on a schedule that actually holds.
This work connects to
Frequently asked
Does the system write creative from scratch, or from our existing ads?
From your existing angles, assets, and brand voice. It is scoped against what has already worked, not a blank page.
Who approves a new variant before it goes live?
A person on your team, every time. The system drafts and recommends; it does not launch spend on its own.
See a weak ad caught before it burns more spend
Tell us what your current testing process looks like, and we will tell you honestly whether a system changes the math.