Facebook and Instagram Video Ad Metrics: Reading the Quartiles

"The video underperformed" is not a diagnosis. Video completion data turns it into one, and it is one of the least-used parts of Meta reporting.

What Meta reports

For every video ad: plays, thru-plays, and the number of people who reached 25%, 50%, 75%, 95% and 100% of the run time, plus average watch time.

Plotted, those five points form a drop-off curve. Where the curve falls tells you which part of the creative failed.

The quartiles are proportional, not absolute. The 25% mark on a 12-second ad is three seconds in; on a 90-second ad it is 22 seconds in. That single fact decides how you read everything below, and it is the reason two creatives of different lengths cannot be compared on quartile percentages alone.

What video ad completion metrics actually count

A quartile number is a count of people, not a rate. Meta hands you the number who reached the halfway point. To get the percentage everyone quotes, something has to divide it, and the denominator is your choice.

Three denominators are in common use, and they measure different things:

• Divided by impressions. Includes everyone the ad was served to, whether the video ever started or not. This is the strictest reading and the one closest to "of the people who saw this ad in their feed, how many got to the halfway point".
• Divided by video plays. Excludes people who scrolled past before playback began. This flatters the creative, because the hardest part of the funnel has already been removed.
• Divided by the 25% figure. Gives conditional retention, which answers "of the people still here at 25%, how many stayed to 50%". Useful for isolating the middle of a video, misleading if you quote it as a completion rate.

None of them is wrong. Mixing them is. If completion looks like it halved between two months because someone switched denominator, the creative did not change and the report is worthless. Pick one, write it in the column header, and never change it mid-campaign.

Autoplay is why this matters. Playback can begin without any deliberate act from the viewer, so the play count contains people who never chose to watch anything, and dividing by plays treats them as an engaged starting audience.

The quartiles explained, and what a drop at each one indicates

Read the curve as five checkpoints on the same journey. The useful number at each point is not the level, it is the step since the previous point, because that step is the only part of the creative that sits between the two checkpoints.

The 25% mark

Everything from the first frame to a quarter of the way in. On a short ad this is the hook and nothing else. On a long ad, the 25% mark is already well past the hook, so a healthy 25% figure on a 90-second video does not mean the opening worked, it means the opening plus the next 20 seconds survived.

A steep drop before 25% means the ad was rejected on sight, and the rest of the creative is therefore untested. You have no information about your offer, your proof or your call to action, because almost nobody reached them. Fix the opening before judging anything else.

Because the 25% checkpoint is too far in to see the hook on longer videos, add a sixth point at the front of the curve from the shorter play metrics: 2-second continuous plays and 3-second plays, divided by impressions. That reads at a fixed number of seconds rather than a fixed fraction, which is what a hook diagnosis needs.

The 50% mark

The step from 25% to 50% covers the body of the argument: the explanation, the demonstration, the reason to keep watching. People at 25% have accepted the premise. People who leave before 50% accepted the premise and then stopped believing it was going anywhere.

A drop concentrated here points at the section rather than the audience, because the audience already self-selected by staying to 25%. Two mechanisms produce it. Either the pace slows below the point where the viewer is getting new information per second, or the opening set an expectation the body does not match, so the viewer waits a few seconds for the promised thing, does not get it, and leaves.

The 75% mark

The step from 50% to 75% is the one that separates interest from intent. Viewers here have absorbed the main content and are deciding whether the rest is worth their time. A drop at this point points at content that continues after the case has already been made: a second example when one was enough, a recap, a stretch of footage without a new idea in it. The viewer has what they came for and the video has not finished.

This is also where a video that is simply too long for its content reveals itself. There is no penalty for a short ad, and the 50-to-75 step is where the padding shows up as a number.

The 95% mark

The 95% checkpoint exists because the last moments of a video are noisy. Playback can be interrupted, a viewer can swipe a fraction of a second early, and the player itself may not fire the final event. It gives you a reading just before the end that is less sensitive to all of that.

Treat 95% as "reached the ending" and 100% as "sat through the ending".

The 100% mark

Full completion. On its own it says less about the creative than it appears to, because it is dominated by duration: short ads complete, long ads do not, and that tells you about the edit length rather than the quality.

Where it earns its place is in the gap between 95% and 100%, covered below.

What the gap between 95% and 100% says about your end card

If 95% and 100% are close together, the ending is doing its job. People who reached the last stretch stayed for it.

A wide gap between them is a narrow, specific signal: viewers who watched almost the entire ad still abandoned in the final seconds. The mechanism is that the content ended before the video did. A logo sting, a legal card, a held final frame, a slow fade or a silent outro all continue running after the viewer has understood that nothing more is coming, so they leave. Often that gap is measuring dead time at the end rather than disinterest in your message. The data tells you when viewers left, not why, so treat this as the first hypothesis to test rather than the conclusion.

That is worth knowing, because it is a cheap edit: trimming an outro requires no new footage.

It also affects where your call to action belongs. If the ask lives in the final seconds and a wide 95-to-100 gap exists, some of your most engaged viewers never received it. Make the ask earlier and repeat it at the end, rather than saving it for the end.

Thru-plays versus completions

A thru-play is 15 seconds, or the full video if shorter. It is a billing and optimisation unit, not a measure of whether anyone watched to the end.

For a 12-second ad, thru-play equals completion. For a 90-second ad, a thru-play means someone watched a sixth of it. Reporting thru-plays on long creative overstates performance considerably, and it is an easy mistake to make because the number is prominent.

How to read thruplay against the quartiles

Thruplay is measured in seconds. The quartiles are measured in fractions. To compare them you have to convert one into the other, and the conversion is a single division: 15 divided by the video's duration in seconds gives you the fraction of the video a thruplay represents.

Read one way, thruplay is a useful shorthand: on a 30-second ad, thruplays and the 50% count should land in the same neighbourhood, and if they do not, one of your numbers is on a different denominator.

Read the other way, the problem appears. On a 60-second ad, thruplay and the 25% quartile measure the same moment. Quoting thruplays as evidence of engagement there is quoting the 25% figure with a friendlier name attached.

Thruplay as an optimisation goal, not just a metric

Thruplay is not only a reported number, it is a delivery objective, and delivery optimises for exactly what you name.

Optimise a 60-second video for thruplays and you are instructing the system to find people likely to watch 15 seconds. It will do that well. Those people are not necessarily the people who watch to the end, and not necessarily the people who buy. You can see it in the data: thruplay volume rises, cost per thruplay falls, and the 75% and 100% counts sit still. Nothing malfunctioned. The system optimised for the checkpoint you gave it.

So if the objective is thruplay, the honest way to report the campaign is on the quartile nearest 15 seconds, plus a separate note on what happened deeper in the video.

Three cross-checks worth running

1. Thruplays against the nearest quartile. On the same denominator they should be close. A thruplay count sitting below the quartile it ought to match means the two numbers were normalised differently.
2. Thruplays against 100%. For a sub-15-second ad the ratio should be roughly one. The further below one it sits, the more of your video lives past the point anyone is counted for.
3. Cost per thruplay against cost per outcome. If cost per thruplay is improving while cost per lead or purchase is flat or worse, the optimisation is buying attention at the 15-second mark that does not convert further down.

Sound-off playback and placement distort the completion curve

Quartile figures in a default report are aggregated across every placement the ad ran in, and placements do not behave alike. Feed, Reels, Stories and in-stream differ in whether playback loops, whether the next item auto-advances, and whether sound starts on. Each of those mechanics changes the count of people who reach a checkpoint without the creative changing at all.

Looping is the clearest example. When a short video restarts automatically, a viewer can pass the same checkpoints more than once, and the counts pick that up. A short ad in a looping placement can show retention a longer ad in the same campaign cannot, for reasons that have nothing to do with which is better.

So break the quartiles down by placement and by platform before diagnosing the creative. If one placement is carrying the drop, the fix is a placement exclusion or a placement-specific cut, not a rewrite.

Sound is the other split worth making. A video whose argument is carried by a voiceover will show a drop early in a sound-off environment, because the viewer is looking at footage with no apparent point. That looks identical in the quartile data to a weak hook, and the remedy is completely different: burn in captions, or carry the argument visually, rather than reshooting the opening.

Average watch time is the sanity check

Two videos can show the same 50% completion rate with very different average watch times. A low average with a decent completion rate can mean a bimodal audience - a small group watched it all, most left almost immediately. That points at the audience rather than the creative, and the quartile curve alone will not reveal it.

There is a second use for it. Average watch time is in seconds, so it survives the comparison that quartiles cannot make: two ads of different lengths. A 20-second ad with 11 seconds of average watch time held attention longer in absolute terms than a 60-second ad with 9 seconds, even though the second one might show a more flattering shape once you normalise by duration.

How to build the drop-off curve for a video ad

The whole diagnosis fits in a spreadsheet with seven columns.

1. Pull, per ad per day: impressions, 2-second continuous plays, 3-second plays, the five quartile counts, thruplays, average watch time and video duration.
2. Divide every count by impressions. One denominator, written in the header.
3. Add a column for the step between consecutive points: impressions to 3-second, 3-second to 25%, 25% to 50%, and so on. The steps are the diagnosis. The levels are just the running total of the steps before them.
4. Add a column converting each checkpoint to seconds, by multiplying the fraction by duration. This is what lets you put two ads of different lengths on one chart.
5. Mark the largest step. That interval of the video is the thing to edit.
6. Repeat weekly and keep the rows. How the curve changes over time is a separate signal from its shape today.

Two rules keep the sheet honest. Do not compare quartile percentages across ads of different durations without the seconds column, and do not read a curve built on a handful of impressions, where a single-digit count at the 100% mark swings the percentage on its own.

What a drop in video completion metrics does not tell you

The curve locates a failure in time. It does not name the cause, and several things that are not the creative can produce the same shape.

• Audience. Serve strong creative to people with no interest in the category and it drops early. That is indistinguishable, in the quartile data, from a weak hook. The tell is comparing the same creative across ad sets: if every audience produces the same early drop, it is the creative; if one audience does, it is the targeting.
• Frequency. People who have already seen the ad recognise it and leave sooner. A curve that degrades while frequency climbs is a saturation reading, not a quality reading.
• Placement mix. Covered above. A shifting placement mix moves the aggregate curve while every individual placement's curve stays put.
• Duration. Longer videos have a lower 100% figure by construction. Comparing them without normalising is comparing edit lengths.
• Conversion. Completion is not conversion. An ad can be watched to the end by people who had no intention of clicking, and an ad with a poor curve can produce cheap leads because the few who stayed were exactly right.

So do not pause a creative on its completion curve alone. Completion metrics diagnose the creative; cost per outcome decides the budget. Use the curve to explain a result you already measured somewhere else.

Turning video ad completion metrics into a creative test

A diagnosis is only useful if it makes the next edit obvious.

• Drop before the first checkpoints. Change the opening and nothing else. New first frame, new first line, or start the existing video a few seconds later so it opens on the point rather than the setup.
• Drop across the middle. Cut, do not rewrite. Remove the interval where the largest step sits and rejoin the two ends. If the argument survives the cut, the interval was the problem.
• Drop before the end. Shorten. Cutting the tail changes duration, so compare the new version against the old on average watch time in seconds and on cost per outcome, not on completion percentage, which will rise mechanically because the video is shorter.
• Wide 95% to 100% gap. Trim the outro and move the call to action earlier.

Change one thing per variant. If you change the hook and the length at once, the curve will move and you will not know which edit moved it. Hold duration constant where you can, and where you cannot, say so in the comparison so nobody reads the mechanical gain as a creative one.

Why this needs history

Creative fatigue shows up in the quartile curve before it shows up in cost per result.

A creative that held 40% to the halfway mark in April and holds 22% in August is dying, and the CPL may not have moved yet because the algorithm is compensating elsewhere. If your reporting only retains a rolling window, you cannot see that trend and you find out when the cost moves, which is later and more expensive.

Inflowave stores all five completion points plus average watch time per ad per day, retained, so the curve is comparable across months. The full metric set is covered in the 58-metric guide.

What to report to a client

Not five percentages. One sentence: "the hook is working, the middle isn't - we're recutting the section after the first five seconds," backed by the curve. The numbers are the evidence, not the report.

Related

• Meta ads reporting for agencies: the 58 metrics
• Meta's three diagnostic rankings
• Why your creative library should not live in Meta

---

Explore More

• Feature: Facebook & Instagram Ads - Every client's ad account in one view, with leads attributed back to the ad that produced them
• Feature: Analytics - Pipeline, conversion and revenue next to ad spend, per client
• Meta Ads Reporting for Agencies - Building the report clients actually read
• Pricing - Plans built for agencies running multiple ad accounts

\

More on running ads for clients\

Start here: Meta ads reporting for agencies - 58 metrics explained, and which ones a client actually cares about.
\

• Ad tracking and attribution software - knowing which ad produced the revenue.\
• Meta Business Portfolio explained - how agency access actually works.\
• Facebook ad account agency access - setup and permissions without the guesswork.\
• Google Ads manager account (MCC) - the same job on the other platform.\
• Automated ad rules - pausing a losing campaign without watching it.\
• Meta ad diagnostic rankings - quality, engagement and conversion, decoded.\
• Link clicks vs outbound clicks - the metric mix-up that flatters a report.\
• Video ad quartiles - where your creative actually loses people.\
• Cost per outcome on Meta - the number to report on.\
• Ad creative library for agencies - why it should not live inside Meta.\
• Meta partnership ads - setup, permissions and pitfalls.\
• What agencies charge for Facebook ads - pricing benchmarks.\
• Moving off Meta Business Suite - and what you gain by doing it.\