retention-lab

Your retention data,
decoded.

Deconstruct any YouTube video's retention patterns with AI. Draft hooks and outlines calibrated to the retention shape of what actually works in your niche.

Join the waitlist → ★ Star on GitHub

Open source · today

Local CLI. Finds top videos on a topic, builds a proxy retention curve from public signals, compares your draft, outputs timestamped edit notes. Proxy — not real retention. GitHub →

Hosted · coming

OAuth your channel. Real retention curves from YouTube Analytics. Copilot that drafts hooks and edits against your actual audience's drop-off patterns. Waitlist →

Live demo

What the tool actually outputs.

Analyzed with retention-lab v0.1 (open source). Input: one YouTube URL.

Veritasium · 25:44 · 4.66M views

Exposing the flaw in tap to pay

Retention shape (proxy)

0%50%100%

Key moments

  • 0:00

    hook

    Opens mid-setup with Derek + MKBHD attempting to steal $10k from a locked iPhone. No intro — pure stakes. Bypasses 30-second drop-off entirely.

  • 2:34–4:17

    peak

    Live $5 test → $10k confirmation → 'approved' chime. Authentic co-host panic drives the highest comment cluster density in the first half.

  • 10:17

    structure shift

    Pivot from 'can we?' to 'why does the system allow this?' — retains analytical viewers by escalating to systemic critique.

  • 23:09

    peak

    Visa rep says 'we won't fix it' on camera. Highest comment cluster in the video. Ends on unresolved controversy — algorithmic engagement spike.

Edit suggestions

  1. Lead with a live demo at real stakes — viewers past the 30s threshold at 4x the rate of setup-first intros.
  2. Structure around escalating proof tiers: POC → edge case → systemic root cause → fix. Each tier is a retention recovery beat.
  3. Engineer a debate trigger in your conclusion. Veritasium's Visa moment drove 40% of top comments. Ask a question that forces viewers to pick a side.
  4. Front-load sponsor with direct thematic relevance — Incogni on a security video causes near-zero retention dip vs. off-topic ads.
  5. Replace abstract protocol explanations with physical metaphors. 'Propped-open door' retains casual viewers through technical sections.

Run it yourself: pip install retention-lab && retention-lab deep-dive <url>

Join the waitlist

Be first when the hosted version ships.

Early supporters get 30% off their first 3 months.

No spam. We'll email you once when the hosted version is ready.

Pricing (planned)

Simple, creator-friendly.

Free

$0

  • · Local CLI forever
  • · Community support
  • · Open source

Pro

$29/mo

  • · Real retention curves
  • · AI Copilot drafts
  • · Web workspace
  • · Email support

Studio

$99/mo

  • · Everything in Pro
  • · Managed agents (later)
  • · Unlimited channels
  • · Priority queue