How it works
One creative, all the way around the loop.
From a raw file before any spend, to live results, to the learnings that sharpen the next ad. Every output is the next step’s input.
- 01Uploaddrop the raw file
- 02Reportthe score & the fix
- 03Script boothfix it, side by side
- 04A/B briefpair the rewrite
- 05Connectreal outcomes flow in
- 06Blended viewbeside real ROAS
- 07In-Flightscale · fix · retire
- 08Agentdecide, or delegate
- 09Gradeforecast vs reality
- 10Calibratedtuned to your account
- 11Trends + DNAyour winning formula
- 12Next briefthe sharper creative
- Back to a sharper startthe loop closes
Read it as a snake: the top row runs left to right, the live row folds back right to left, then the dotted arc closes the loop, carrying every learning into a sharper next brief.
Read it top to bottom: each phase runs in order, and the last step closes the loop, carrying every learning into a sharper next brief.
Read the creative and fix it before a penny goes live. No ad account, no pixel.
Run it, pull the real results back in, and act on every verdict: scale, fix or retire.
Grade the forecast against reality, calibrate to your account, and brief the sharper next ad.
What the Report step hands back
Step 02 on the map, shown on a sample ad. Your uploads get the same output: a score, the sections costing the most, and the fix.
- Weakest sections
Offer scores 13: the reviews badge on screen is its only reason to act, with no discount, urgency or guarantee. Call to action scores 55: vocal energy drops 32% at the ask.
- Where
The ask lands flat at 0:24: “link’s in my bio if you want to check it out.”
- The fix
Re-record the CTA standing, with a direct line: “Tap the link and start tonight.” Then give a reason to act: a guarantee or a first-order offer. Both briefed before any spend.
An illustrative sample analysis, hand-curated to show the report format — not the stored output of a pipeline run. Your ad’s numbers are read from your ad; nothing here is assumed about yours.
Your ads teach it what works for you.
- SignalsWhat happens in the ad
Pitch, energy, pace and pauses second by second, plus the words and sampled frames.
- MetricsThe same read for every ad
The same scored sections for every ad, so any two compare like for like.
- ForecastFrozen before you spend
Stored at analysis and never refit to the results that follow.
- Your modelWeighted to your results
Weights shift from research norms towards what tracks your results.
When it starts learning from you
- From your first upload
- Research-based weights, shared by every account.
- Your industry
- Re-weighted from 50 outcomes in your industry, pooled from accounts that share pseudonymised metrics (on by default, off in Account). A calibration needs 5 or more accounts, so it is never one advertiser’s data in disguise.
- Your account, from 20 results
- Signal weights start moving towards yours, reaching full strength at 50, and only if they rank your own held-out results better than the weights below them, by a margin noise rarely reaches. Otherwise you stay on those.
- Your account, from 50 results
- Forecasts can be labelled calibrated to your account, and a model trained on your ads alone can take over the headline score, under the same out-of-sample test.
The proof is on your numbers: from 10 results, the forecast ledger scores each forecast out of sample against simply predicting your account’s average. If it doesn’t beat that, it says so.