Reports & strategy · Runs on a schedule
Find which models and image styles perform best
Every Monday, Tolstoy compares your models and image styles on plays, completion and attributed orders, flags fatigue by fixed rules and suggests three tests.
What you need
- Images and videos in your Tolstoy library, with product tags where possible
- Tolstoy widgets on your store, so each asset collects exposure and plays
What you get
- A table of winning signals by model or treatment, with sample size and confidence
- A fatigue and underperformance list with the measured signal for each item
- Three controlled next tests, plus notes on missing data and coverage
How it runs
Tolstoy runs this playbook in 5 steps.
-
Set the window
The routine compares the last 30 complete days with the 30 days before, and records the exact dates and metric definitions.
-
Pull asset performance
It reads exposure, plays, completion, interactions and attributed order credits for each asset, then matches each asset to its library record and product tags.
-
Group the creative
It groups assets only by recorded or visible attributes: a named model, a treatment such as on-model or lifestyle, and the format and aspect ratio.
-
Compare fairly
It ranks a group only with at least 3 assets and 100 exposures or plays, and keeps smaller groups as directional evidence.
-
Flag fatigue and plan tests
It marks an asset fatiguing only when its metric falls across two comparable weeks at meaningful exposure, then proposes three tests that each change one variable.
Check the result
- Every row in the winning signals table shows its asset count, exposure or plays, and a confidence level.
- Each item on the fatigue list shows the measured signal and its sample, not only a low number.
- Each of the three tests changes one thing, such as the model or the background, and keeps the rest the same.
Know the limits
- It reads and reports only. It does not publish, replace or remove content, and you choose which tests to run.
- It measures assets that shoppers see through Tolstoy widgets on your store. Results from your Meta ads are not in this report.
- Comparisons are descriptive, not causal. A winning model is a signal to test, not proof that it caused more sales.
When to use this playbook
Use it when you produce many AI Studio images and videos and want evidence on which models and treatments to make more of, instead of choosing by taste. It needs content that shoppers see through Tolstoy widgets, because each asset needs storefront exposure and plays. For why tired creative costs more in paid social, read why Meta ads get expensive and how AI fixes creative.
Get better results
- Name your models in Tolstoy and tag assets to products in your library before the first run.
- Run the suggested tests as written. A test that changes the model and the background together cannot tell you which one mattered.
Schedule
Every Monday at 11:00 in your time zone; you can change the time. “Set up this routine” opens the routine in Tolstoy, where you check its connections and time. Nothing runs until you save it.
FAQ
What is the difference between fatiguing and underperforming?
Fatiguing means the metric fell across at least two comparable weeks while exposure stayed meaningful. Low performance with no downward trend is marked underperforming instead.
How does it know which model is in an image?
From the model name recorded in Tolstoy. It never guesses who a person is from how they look, so assets without a named model are grouped as "unknown model".
What do I do with an asset marked fatiguing?
Run one of the suggested tests on it. Each test changes one variable, such as the model or the background, so the next report shows which change moved the metric.
Run it on your own catalog
Start free, connect your store, and Tolstoy opens this playbook for you. You review every result before anything goes live.