---
name: improve-hook-rate-and-hold
description: >-
  Defines and improves the two numbers that decide whether a Meta video ad gets watched: hook rate (3-second
  video views / impressions, the thumbstop rate) and hold rate (ThruPlay or 15-second views / impressions).
  Pulls both via mads_run_insights's video-view metrics, diagnoses whether the break is in the first 1-3
  seconds (weak hook) or in pacing/length after that (weak hold), and matches the fix to the stage -
  front-loading the payoff, captions, pattern interrupts, native feel, or trimming - not a generic re-edit.
  Every benchmark it cites (e.g. hook rate ~30%+) is DIRECTIONAL and time-sensitive, not a pass/fail bar.
  Reach for it when a video's cost per result lags and you need to know which stage is broken before
  touching footage, or to confirm a fix moved the retention curve. It ends at an improved retention curve on
  a shipped variant - it does not produce new creative concepts (produce-ugc-creative-brief) or diagnose
  decay on a creative that worked (diagnose-creative-fatigue).
---
# Improve Hook Rate and Hold Rate

## Purpose
On Meta's video placements, the first few seconds decide whether anyone sees the rest of the ad.
Hook rate and hold rate split that decision into two separately-fixable stages: did they stop
scrolling at all, and did they keep watching once stopped. Treating both as one vague "engagement"
problem produces re-edits that fix the wrong stage. This skill measures each stage precisely and
prescribes the fix that actually matches the break.

## When to run
- A video creative's cost per result is underperforming and the stage of the problem is unknown.
- Establishing a retention baseline for a new video concept before scaling it.
- After a hook or pacing change, to confirm the retention curve actually moved.

## When NOT to run
- The creative hasn't cleared learning or accumulated enough impressions yet, wait; see the
  judging floor in `run-creative-testing-framework`.
- Performance is declining on a creative that WAS working, that's `diagnose-creative-fatigue`, a
  different diagnosis from a creative that never hooked well.
- No video creative exists yet, `produce-ugc-creative-brief` first.

## Prerequisites
- A video ad with enough impressions to be non-noisy, a few thousand impressions minimum before
  reading hook/hold ratios as signal rather than noise.
- Access to video-view action types in Insights: 3-second views, ThruPlay (or 15-second views), and
  impressions.
- Cost per result already reviewed, hook/hold is a diagnostic layer underneath it, not a
  replacement for it.

## Procedure
1. **Pull video metrics.** `mads_run_insights` (poll with `mads_get_insights`, or use
   `mads_get_insights` for a single ad's fast pull) at the ad level for impressions, 3-second
   video views, ThruPlay/15-second views, spend, and results.
2. **Calculate hook rate** = 3-second video views ÷ impressions (the thumbstop rate).
3. **Calculate hold rate** ≈ ThruPlay (or 15-second) views ÷ impressions.
4. **Compare against directional benchmarks only.** A rough, time-sensitive goal is hook rate
   ~30%+; hold rate is judged against video length and the account's own trend, not a fixed number.
   Never present either as a guarantee.
5. **Diagnose the stage.** IF hook rate is low, the first 1-3 seconds are the problem regardless of
   what follows. IF hook rate is healthy but hold rate is low, people stopped and then left, pacing, length, or payoff placement is the problem, not the opening frame.
6. **Match the fix to the diagnosis**, see Decision rules. Never apply a hold-rate fix to a hook
   problem or vice versa.
7. **Ship the fix as a new variant, not a live edit.** `mads_create_creative` referencing the
   re-cut or re-captioned video, keep the losing variant paused (`mads_update_ad`) rather than
   deleted so the comparison survives, `mads_preview_creative` before submission, plain-English
   summary, explicit approval, then commit.
8. **Re-measure once the new variant clears the same impressions bar** as step 1, and compare the
   retention curve itself, not just cost per result, to confirm the targeted stage improved.
9. **Escalate concept-level fixes.** If the fix needs a genuinely new angle rather than a trim or
   caption pass, hand off to `produce-ugc-creative-brief` instead of iterating blindly on the same
   footage.

## Decision rules
- **Weak hook →** front-load the payoff or most visual moment into the first 1-3 seconds, open on
  motion or a pattern interrupt, cut any logo/branding lead-in that delays the hook, make it read as
  native to the placement rather than a polished ad.
- **Weak hold (hook fine, hold low) →** re-pace the middle (cut dead air), move the offer/CTA
  earlier, trim overall length, add captions/subtitles (most viewers watch muted), cut a slow
  explanation in favor of showing the outcome directly.
- **Both weak →** treat it as a concept problem, not an edit problem; route to
  `produce-ugc-creative-brief` for a new hook/angle rather than re-trimming the same footage.
- **Benchmarks decay**, platform and format norms shift; trust the account's own trend line over a
  fixed external number.
- **Insufficient impressions →** do not diagnose yet; the ratio is noise below a few thousand
  impressions.
- **Done means:** hook rate and hold rate are both calculated from a clean pull, the weak stage is
  identified, a matched fix has shipped as a new variant, and the retention curve is confirmed
  improved (or the attempt is logged and a new concept requested).

## Common failure modes
- **Diagnosing off too few impressions** and chasing noise instead of signal.
- **Applying a hold-rate fix to a hook problem**, re-pacing the middle when nobody got past second
  one is a wasted edit.
- **Treating a directional benchmark as a pass/fail bar** instead of a rough, decaying goal.
- **Comparing hook/hold rate across very different video lengths** without adjusting expectations.
- **Editing the live ad's creative in place** instead of shipping a new variant, losing the
  before/after comparison entirely.

## Related skills
- Before: `run-creative-testing-framework` (establishes which creative is worth this diagnostic
  depth).
- After: `produce-ugc-creative-brief` (when the fix needs a new concept, not just an edit).
- Related: `diagnose-creative-fatigue` (declining retention on a previously healthy creative is
  fatigue, not a hook/hold defect).