The Video Ad Metrics That Actually Predict Winners

Picture a skincare brand running its first round of paid social. The media buyer, Dana, ships five video ads against the same audience at modest daily spend. By Thursday morning, none has enough purchases to call a winner with any confidence: CPA is empty, ROAS is a rounding error, and the dashboard shows impressions and view counts that all look roughly the same. The temptation is to wait another week. The better move is to read the early video ad metrics that have already declared themselves and reallocate budget by Friday.

That is the core problem with measuring creative early. The KPIs that feel reassuring tend to mean the least, while the ones that actually predict a winner sit a layer down. This piece sorts those two groups apart: which video ad metrics forecast performance inside the first 48 hours, and which vanity metrics are wasting your attention. The benchmarks are drawn from published 2025 to 2026 platform and agency data, cited inline so you can check them.

Why some metrics declare themselves early and others do not

A video ad has to clear three gates, in order: someone has to stop scrolling, then keep watching long enough for the message to land, then take an action that points toward revenue. Each gate has a metric that becomes statistically stable at a different point in the test, and the ordering is mechanical. Impression-based ratios stabilize fastest because impressions accumulate quickly, while a conversion rate needs conversions that trickle in. The metrics worth reading on day two are the ones whose denominators are already large by day two, which is why ROAS belongs at the end of a test rather than the start.

Gate one: hook rate, the scroll-stop

Hook rate is the share of impressions that turn into a counted early view. On Meta it is calculated as 3-second video plays divided by impressions, times 100. It is the same arithmetic as the thumb-stop ratio, just a different label, and it is the first thing the auction learns about your creative. Roughly 90% of viewers leave if the opening does not catch them in the first few seconds, which means a weak hook rate puts a hard ceiling on everything that follows.

One detail trips people up: the counting threshold differs by platform. TikTok registers a hook at the 2-second mark and Meta at 3 seconds, so a TikTok hook rate and a Meta hook rate are not comparable numbers. When the opening is the weak link, that is where to work first, and a tighter set of scroll-stopping openers shifts this metric more than any other single change.

Gate two: hold rate, the retention check

Hold rate is retention measured against attention you already earned: ThruPlays divided by 3-second video plays, times 100. Where hook rate asks whether the scroll stopped, hold rate asks whether the body of the ad keeps the people the hook caught. When the hook is strong but the hold collapses, the opening over-promised or the pacing sagged, which tells you the narrative is the problem, not the hook.

This is the gate where the script does its real work, and it is where most ads quietly die. The first three seconds get engineered carefully and the next ten arrive as an afterthought. Fixing it is a structural script problem rather than a thumbnail swap, which is why hold rate is one of the more diagnostic video ad metrics you can track.

Gate three: cost per outbound click, the first revenue signal

The earliest conversion-side metric that gets reliable at low spend is cost per outbound click: what you pay for a click that genuinely leaves the platform for your site, as distinct from one that merely expands the post. It becomes statistically meaningful far sooner than CPA or ROAS, which both need conversion volume you will not have on day two. It answers a useful question before the purchase data exists: is the attention you bought the kind that moves toward a checkout, or the kind that goes nowhere.

The 2026 benchmarks: what good looks like, with worked numbers

Benchmarks are reference points rather than targets, and they are platform-specific. Below are the published ranges, current as of 2026, by platform and placement.

Meta hook rate and thumb-stop

Hook rate. A common target band sits at 20 to 25%, with top ads above 30%. Motion frames strong performance as a 30 to 40% baseline, with anything under 25% signaling a creative problem, and Billo's tiers put best-in-class at 30%+ and elite at 35 to 45%.
Thumb-stop by placement. Because it is the same formula, the placement-level data is where it gets useful. Facebook Feed cold prospecting runs 18 to 28% with a median near 23%, Reels 24 to 36% with a median near 30%, Stories 22 to 32%, and warm retargeting 30 to 45%. Reels demand 5 to 10 points more than Feed because the swipe is faster.
The danger line is around 20 to 25%. AdSights notes that a thumb-stop ratio below 25% means most of your budget funds unseen impressions, because the auction will not grant real reach to a creative nobody stops for.

Vertical shifts the whole range. Gaming creative can reach 38 to 52% on Meta Reels, while SaaS and B2B carry a structural ceiling closer to 20 to 26%. Holding a B2B demo to a gaming benchmark will make a perfectly healthy ad look broken.

Meta hold rate and ThruPlay

Hold rate. The average lands at 40 to 50%, with top performers above 50%. Motion calls above 60% strong and under 30% in need of work, and Billo notes that an ad clearing 35% hook plus 30% hold is demonstrably performing.
ThruPlay is length-dependent, so compare like with like. A ThruPlay counts as a complete play for videos under 15 seconds, or reaching the 15-second mark on longer ones. That is why sub-15s ads benchmark at 70 to 95% with a median near 85%, 15 to 30s DTC prospecting at 15 to 30%, and 30 to 60s at 8 to 18%. Comparing a 12-second ad's ThruPlay against a 40-second ad's measures the threshold, not the creative.

TikTok: the 2-second hook and the 6-second view rate

TikTok's algorithm reads two thresholds. A 3-second view rate above 40% tells the system the hook landed, and it widens distribution; below 40% it restricts delivery. Then the 6-second VTR splits at 45% for sustained engagement, 25 to 35% as the industry average, and below 25% gets deprioritized. TikTok's own documentation defines a 6-second focused view as a play of at least 6 seconds, a full play if shorter, or an engagement within the first 6 seconds.

Here is the counterintuitive part for anyone over-indexed on clicks. A campaign with 0.7% CTR and 45%+ 6-second VTR outperforms one with 1.1% CTR and 22% VTR in algorithmic efficiency. TikTok's average CTR is only 0.84%, so chasing clicks while ignoring retention optimizes the wrong gate. For the format mechanics behind these numbers, the TikTok specs that actually convert are a more productive starting point than CTR alone.

YouTube: view rate, not raw views

YouTube's analog is view rate: views divided by impressions, where a view counts at 30 seconds, at the end of the video, or on a click. The overall 2026 average is 31.9%, with desktop at 35.4% and mobile at 33.2%, while skippable in-stream specifically baselines at 15 to 25%, with strong campaigns at 30 to 40%. Completion falls with length: 15-second ads finish at 70 to 85%, 30-second at 40 to 60%, and 60+ at 20 to 35%. Connected TV behaves differently again, with 90 to 97% completion and non-skippable pre-rolls at 98.6%, so a skippable in-feed ad should never be judged against CTV completion. The Shorts playbook follows its own set of rules.

The video ad metrics to ignore, and why they lie

Every metric in this section can look healthy on an ad that is failing. That is precisely what makes them dangerous, because they offer comfort while absorbing the attention the three gates deserve. These are the vanity metrics.

Raw impressions. Impressions measure how much budget you spent, not whether anyone engaged. A high impression count paired with a sub-20% thumb-stop means the platform served your ad and viewers scrolled past it. You paid for invisible reach.
Total video views without retention context. "10,000 views" tells you nothing on its own, because it hides where people dropped. As Motion puts it, impressions, likes, and views alone are misleading, and the real signal combines attention, engagement, and conversion. A view that ended at 3.1 seconds says nothing about whether your message landed.
CTR in isolation. The TikTok comparison above is the proof: a higher CTR can mask worse algorithmic efficiency. CTR earns its place as part of a pattern, read alongside view rate and cost per outbound click, never as a standalone scoreboard. For reference, traffic-campaign CTR on Meta averages 1.71%, but a high CTR with a poor hold rate is just an ad that wins the click and loses the visitor on the page.
Likes, comments, and shares. Treat engagement as a tie-breaker between two ads that already convert. As a primary signal it is close to useless.

One rule sorts the whole list. Any metric the algorithm does not use to decide distribution, and that does not sit on the path to revenue, is a tie-breaker at best. Hook rate, hold rate, view rate, and cost per outbound click each gate distribution or forecast revenue. Likes and raw impressions do neither.

The 2x2: a hook-by-hold matrix you can read in 48 hours

The most useful tool here is a single matrix that plots hook rate against hold rate. Pull both for each ad at the 48-hour mark, drop it into one of four quadrants, and the diagnosis is unambiguous. To make the quadrants concrete, here is a worked example for each, using a Meta Feed placement where the thumb-stop floor is about 23% and a healthy hold sits near 45%.

High hook, high hold (example: 34% hook, 52% hold). The creative works on both counts. Move straight to cost per outbound click. If that is also strong, say a click cost roughly 30% under your account average, this is your scale candidate. If the click cost is high despite that great retention, the leak is the offer, the landing page, or the audience, not the video.
High hook, low hold (example: 36% hook, 24% hold). The opening stops the scroll, then the body loses them. Keep the first three seconds and rewrite the middle. Vaizle calls this the most common fixable quadrant, because the hook is a proven asset and the script is the liability.
Low hook, high hold (example: 17% hook, 58% hold). Almost nobody starts, but the few who do stay to the end. The body is good and the opening is wrong. Swap the hook and you may find a winner that was hiding behind a bad first frame.
Low hook, low hold (example: 15% hook, 22% hold). Stop spending on it and do not iterate. Vaizle's guidance is explicit here: fix hook rate before hold rate, because a low hook prevents viewers from ever reaching the message you would be trying to improve.

The sequencing is the point of the whole matrix. Polishing the script on a low-hook ad is repainting a room nobody walks into. This is also why disciplined creative testing changes one variable at a time. If you swap the hook and rewrite the script in the same iteration, the matrix can no longer tell you which change moved the number, and your creative testing metrics lose their meaning.

Reading the order: what to check on which day

The practical version of all this is a fixed reading order for any fresh batch. The order matters more than any single threshold, because it stops you from rationalizing a loser or killing a slow starter too soon.

Days one and two, check spend before performance. Confirm each ad has cleared enough impressions for hook rate to stabilize. Below that point the numbers are noise, and judging them is guessing.
Gate one, hook rate or thumb-stop. Sort the batch by it. Anything under the placement floor, around 20% for Feed and 25% for Reels, goes on the kill list unless its hold rate is exceptional.
Gate two, hold rate within the same length bracket. Rank the gate-one survivors on retention, and remember to compare ThruPlay only within the same length bracket so the threshold mechanic does not mislead you.
Gate three, cost per outbound click. Of the ads that stop and hold, find the ones buying the cheapest qualified clicks. That is your conversion-side forecast before CPA is trustworthy.
Decide: scale, fix, or kill. Scale the ads with a strong hook, strong hold, and cheap clicks. Fix the ones with a clean single-gate failure. Kill the rest and recover the budget.
Only now look at CPA and ROAS, as confirmation. Meta ROAS commonly lands between 2.5 and 4.0 depending on customer value. Read it as the autopsy that confirms what the leading metrics already told you, not as the decision itself.

The catch nobody budgets for: throughput on a 48-hour clock

There is a structural problem under everything above, and it has nothing to do with reading skill. Suppose the matrix tells Dana, on Thursday, that her best ad has a strong hook and a dead middle, and a second ad needs an entirely new opening. Both diagnoses are correct and both are actionable. The catch is that acting means producing two new variants, and the test window that surfaced the signal is still ticking.

Made the traditional way, a new hook and a re-cut middle is a brief, a turnaround of a day or two, and a revision pass. By the time those variants are live, the original test has aged out and the data she is comparing against is stale. This is why many small teams stop iterating after the first round: they produce what they can afford, let the numbers pick the least-bad option, and treat that as the strategy.

Leading video ad metrics only create an advantage when production keeps pace with the signal. A 48-hour read is worthless if your fastest fix takes 48 hours to ship, because you are always acting on data that has already expired. The leverage is not cheaper creative; it is closing the loop between diagnosis and the next variant fast enough that what you learn on Thursday changes what is running on Friday. That is the argument for treating iteration speed as a competitive moat, and it is what lets a solo founder sustain a testing cadence that used to require a team. For an agency managing many accounts, the same throughput is the difference between feeding every client and overloading one editor.

Common questions about reading early

How long before I can trust video ad metrics?

Hook rate and thumb-stop stabilize first because their denominator is impressions, so they become usable within a day or two once an ad has cleared enough impressions. Hold rate follows. Cost per outbound click is the earliest reliable conversion-side signal. CPA and ROAS lag, because they need conversion volume you rarely have early, so use them to confirm a decision rather than to make it.

What is the difference between hook rate and hold rate?

Hook rate measures the scroll-stop, 3-second plays divided by impressions. Hold rate measures retention among the people who already stopped, ThruPlays divided by 3-second plays. The hook tells you whether the opening works; the hold tells you whether the body keeps them. A strong hook with a weak hold points to a good opening and a flat script, so you fix the middle and keep the first three seconds.

Why does my ad have a high CTR but bad sales?

A high CTR with a weak hold rate usually means the ad pulls clicks it cannot support: the hook over-promises and the body or the landing page does not deliver. Read hold rate and cost per outbound click together. If hold is low, the creative loses people before the message lands. If hold is fine but sales are not, the leak is downstream on the page, where an ad-to-landing-page mismatch is the usual culprit.

Sources

If the production side is your bottleneck, that is the specific gap Aitachyon is built to close: when the 48-hour read says "great hook, dead middle," you can generate the corrected variant and have it ready the same morning instead of waiting on an edit queue. Start a test batch and let the leading metrics actually change what runs tomorrow.

The Video Ad Metrics That Actually Predict Winners