The leading RSI algo for content creation
Scan every second of your video, up to 3 minutes, and see exactly where viewers drop off. Fix it before you publish, so every post and every ad dollar keeps people watching.
We watch. You fix. You publish.
Shorts, Reels, TikTok and video ads, up to 180 seconds.
Real output from the internal demo. The same scan runs across the whole video, up to 180 s.
01_/8_
The videos that win keep people watching to the end. We help you find what does that for your brand, and make every post and every paid campaign work harder.
Every video you scan shows what holds your audience and what loses them: openings, pacing, pauses, endings. Over time that becomes a playbook you can repeat, built from your own content.
A creative that loses viewers early wastes budget. Fix the drop-off points before launch, and spend behind the versions that hold attention to the end.
02_/8_
Viewers can leave at any moment, not only in the hook. We scan the whole video, up to 3 minutes, and give the opening extra attention.
A pause isn’t always bad. Some build anticipation; others just stall. We tell them apart.
Fix the weak moments while you can still edit, not after the video has already lost its audience.
03_/8_
Drop in the video or ad creative before it goes live. Up to 180 seconds, mp4, mov, webm or mkv.
From the first frame to the last, moment by moment, through our three-level approach.
Hook score, drop-off curve, flagged moments with a reason, and the top three fixes. About half a minute in the demo.
Once your video has had time to gather views, bring back its real retention. Your next prediction gets sharper.
04_/8_
No single method is right on day one and still right at scale. So we combined three. Each level reads the same video at a different depth, and together they get sharper every time you publish.
Reads your video the way a seasoned editor would.
Predicts how an audience will watch, before anyone has.
Our own model, built from real audience behaviour.
You get a clear read from the very first video, before there is any history to learn from.
Every published video makes the next prediction more accurate, starting with your own channel.
All three levels work from the same understanding of your video, so their findings line up.
A deeper level runs quietly alongside and has to prove itself before it takes over.
Proprietary approach. Methods and models are confidential.
05_/8_ · RSI
RSI = recursive self-improvement. Every video you make teaches the system, and the system makes your next video better. Results flow back into every level, so each turn of the loop starts smarter than the last. It compounds.
You make a video or an ad creative.
Every second is read and a prediction is recorded before it goes live.
It goes out. About two days later, the real audience result comes back.
The result feeds the levels. The next prediction, and the next video, start smarter.
Sets the starting line. A strong read from the very first upload.
Sharpens with every result. Each published video tunes the next prediction.
Grows from what the loop collects. More results, more precision.
Your viral formula isn’t written once. It’s discovered, tested and refined, loop after loop.
Proprietary approach. Methods and models are confidential.
06_/8_
How likely people are to stay past the first 3 seconds, with what weakens it.
Risk or predicted retention for every moment of the video, up to 180 seconds, start to finish.
Each with a severity, watch risk likely swipe, and a reason you can act on.
Plain-language “why” and “fix” for the moments that cost the most viewers, wherever they are in the video.
Two videos with the same 5-second pause. Most tools flag both. We only flag the one that loses viewers.
risk Flagged: nothing gives the viewer a reason to wait.
07_/8_
In the app, Settings → Prediction engine has one switch per level. Only one is on at a time. Try it below.
Level 1, the expert read. Risk curve, hook score, flagged moments and fixes.
Level 2, audience simulation. A predicted retention curve with a range, sharper after you publish.
Level 3, learned retention. In development; shown greyed out in the app for now.
On: Rules + AI judge
08_/8_
The app runs as a local demo for now. This page explains it; testing happens in the demo.
Ask the team for a session on the demo machine, or run it yourself: make demo, then open localhost:3000.
Enter your email and click the link (in the demo, the team reads it from the server log). No passwords.
weak_hook → low hook score, early “likely swipe”.strong_hook → strong hook, few or no moments.dead_hold → the still shot is flagged.suspense_hold → the same still is not flagged.Anything you’d post: up to 3 minutes. Compare the flags with your own instinct.
Settings → Prediction engine. Turn on Simulation (Level 2) and analyse the same clip again.
FAQ
Videos up to 180 seconds: YouTube Shorts, Reels, TikTok and short video ads. You upload the file; it doesn’t post anything for you.
No. It scans every second of the video, up to 3 minutes, and flags drop-off points wherever they happen. The opening gets extra attention because that’s where many viewers decide.
Paid views that drop off early don’t deliver your message. Checking creatives before launch lets you fix those moments and put budget behind the versions that hold attention. We don’t promise a number; we help you spend on stronger creatives.
It predicts where viewers are likely to leave, not how many views you’ll get. Level 1 shows risk. Level 2 shows a predicted retention curve with a range, which gets sharper as real results come back.
About half a minute per clip on the demo laptop. Longer videos take longer.
In the internal demo it stays on the demo machine, except when it’s sent to an AI service for analysis.
Level 3, connecting a YouTube channel, and a public sign-up. This is an internal preview.