Analyzing Drop-Off Curves: How to Diagnose Exactly Where a Video Ad Loses the Viewer
· Paid Social

We split the 30 highest-spending video ads in a fragrance and body care account down the middle by hook rate this week. The half that stopped the most people paid more for every purchase.
The top half had a median hook rate of 48% and a median CTR of 3.4%, against 32% and 2.5% for the bottom half. On paper the first group wins on attention, yet its blended CPA came in at $31.03 against $28.75 for the group that hooked fewer people. That gap is nearly 8% across more than $630k in spend.
Hook rate tells you how many people you caught in the first 3 seconds, with nothing on who you caught or where you lost them afterward. That second part is where most of the money in a video ad gets made or wasted, which is why we read the drop-off curve before we read the hook.
Hook rate measures 1 moment in a 60-second video
Most of the ads in that account ran from 30 seconds to well over a minute, with the longest high-spend ad past 2 minutes. Hook rate, 3-second video plays divided by impressions, grades the first 5% of a 60-second ad and ignores the other 95%.
That would be fine if the people who stay for 3 seconds behaved like the people who stay for 30, which they rarely do. A hook built on shock or curiosity can pull in a large number of scrollers who were never going to buy, inflating hook rate and CTR while the purchase rate falls behind. Our read is that some of the highest-hooking ads in this account were doing exactly that. The drop-off curve is how you check the read instead of guessing at it.
The curve maps attention across the whole ad, so its shape tells you which part of the creative to fix, even when 2 ads share an identical hook rate.
You can build the curve from 7 columns already in Ads Manager
Meta gives you enough to sketch the curve at the ad level without extra tooling. This is the setup we use when auditing video creative.
Add impressions, 3-second video plays, video plays at 25%, 50%, 75% and 95%, and ThruPlays to a custom column set. Save it so the whole team reads the same view.
Divide every count by 3-second video plays instead of impressions. Each point then reads as the share of hooked viewers still watching, which keeps the hook from distorting the rest of the curve.
Convert each quartile into seconds using the ad's length. The 25% mark lands at second 7 on a 30-second ad and at second 30 on a 2-minute ad, a gap that percentages alone hide.
Plot your top ads by spend against the account median for ads of a similar length. Your own account is the benchmark to use, since another brand's numbers reflect a different audience and offer.
Filter to ads with meaningful spend before reading anything into the shape. A curve built on a few thousand impressions moves around too much to diagnose.
5 points make a rough curve. Meta's quartile figures also include plays that skipped ahead to that point, so treat the curve as a map of where the big drops happen rather than a second-by-second trace. That is enough to tell you which part of the ad to open up in the edit.
4 curve shapes cover almost every video ad we audit
The cliff at the hook. Hook rate sits below the account median, but the people who do stop keep watching at a healthy rate through the 50% mark. The first frame is the weak point in an otherwise working ad, so test new opening shots and opening lines on the same body before you touch anything else. This is one of the lowest-lift fixes in paid social, because 1 good edit can be cut into 5 or 6 hook variants.
The second drop.Hook rate looks strong, then a large share of hooked viewers leaves before the 25% mark. The hook made a promise the next beat didn't keep, usually because the setup is slow or the opening was bait unrelated to the product. We'd expect most of the high hook rate, high CPA ads in the account above to land here until the curves prove otherwise. The fix is tying the hook to the problem the product solves, so the right people stop.
The steady slope. No single cliff shows up anywhere, only a gradual leak from start to finish. The pacing is the problem, so cut dead air, tighten the cuts, and test a shorter edit before you write a new script. If a 60-second ad loses viewers evenly all the way down, a 30-second version can hold the same buyers for less.
The late drop.The curve holds well into the middle and then falls away at the product reveal or the offer. People were happy watching the story and left when it turned into an ad. Look at how the product enters, whether it arrives as the natural answer to the problem or as a hard pivot. Check whether clicks happen before the drop too, because a late exit after a click comes from a viewer who already did what you wanted.
Yapper ads, where 1 person talks straight to camera about a problem the viewer already feels, are built to avoid the second drop and the late drop. The problem carries attention all the way to the product, so the product arrives as the answer instead of an interruption.
The fix lives at the exact second the curve breaks
Once you know the shape, line the drop up against the script. Pull the transcript, mark the timestamp for each quartile, and read the line on screen where the biggest share of viewers leaves. In most accounts the problem becomes obvious once you look at the right 5 seconds, whether that is a slow product demo, a jump cut to a logo, or a creator restating the hook when the story should move on.
Do this across your top ads at once rather than one at a time. When 3 ads built on the same angle all lose people the moment the price appears, the offer framing is the problem, which changes what goes in the next creative brief.
The curve also changes how you brief iterations. An ad with a cliff at the hook needs new openings on the same body, while an ad with a late drop needs a new middle and ending on the same opening. A steady slope calls for a shorter cut of everything. Writing the brief from the curve stops editors guessing at what to change, so every iteration tests 1 thing.
Read the curve next to CPMr before you blame the creative
A winning ad that starts losing efficiency has 2 possible stories. The drop-off curve read alongside CPMr tells you which one you are in.
CPMr, or cost per mille reached, is a framework popularized by Phil Kiel, our Director of Paid Social. It multiplies CPM by frequency to give the cost of reaching 1,000 unique people. When it climbs week over week, your budget is cycling through the same audience instead of finding new people.
If the curve on a proven ad holds its shape week over week while CPMr climbs, the audience is running out on creative that still works. The levers here are exclusion hygiene, consolidation, and new creative built for a different territory, since new hooks on the same ad won't reach anyone new. Fresh territory also gives Andromeda, Meta's ad retrieval system, a wider pool to find.
If CPMr is flat and the curve itself starts sagging, with more people leaving at the second drop or the slope steepening, the people still being reached are tiring of the creative. That is the moment to cut new hooks and new edits from the body that already works.
Both problems show up as the same symptom, rising CPA on an ad that used to win, yet they need opposite fixes. Checking both numbers together takes 5 minutes and saves a week of iterating on the wrong thing.
https://x.com/PhilKiel/status/2029701498268176617
Start with your 5 biggest video ads this week. Pull the quartile columns for each one, convert them into seconds, and sort each ad into a shape. Whichever shape shows up most is the brief your editors should work from next. The high hook rate ad you were about to scale might be the first one you rewrite.
Read more on CPMr and how to calculate ithere.
Opening data comes from a single fragrance and body care DTC account, covering the 30 highest-spending video creatives over the 30 days to October 6, 2026, split at the median hook rate. CPA is blended spend divided by purchases for each group. All data anonymized.