Meta scales the best-watched ad. Money follows attention.

We keep making one claim: for video ads, watch rate is the metric worth managing. Some time ago, a performance marketer pushed back. His argument stayed with us because it was good — grounded in the job as it’s actually lived, and strong on reason. It deserved better than a dismissal. It deserved evidence.

Outcomes are the business, he said. CPA, ROAS — whatever the yardstick, the scoreboard that pays the bills is the only one that matters. And my process is built around that: I launch a few videos, Meta scales the one it favors, and I check that winner against my targets. If it holds, I put more behind it. If it doesn’t, I kill it and test fresh creative. More volume, more shots at a winner.

How Meta chooses between my videos is Meta’s business. Whatever number it runs on inside that machine, I can’t spend it — I can spend conversions. When you hand me a new metric to watch, all I hear is homework: more to track, nothing promised at the end. And I’ve no room for it — ads cost more every year¹, my targets don’t move. Tie your metric to the money — what I spend, what I earn — or it’s one more maybe I can’t afford. Nobody has shown me one.

Can you show me the link?

It’s a strong argument. Every step follows from the one before — and the whole of it is almost true. Almost is the interesting part.

We do one thing with beliefs like this, ours included: test them. His challenge couldn’t be answered from a dashboard. So we did the work: on a real account, we read every race Meta ran between competing videos — who got the budget, and what the winner had that the loser didn’t.

What came back isn’t a debating point. It’s a finding: Meta’s pick is readable from the outside, and the number that predicts it is one he already has.

The question everyone stopped asking

Notice what his argument rests on. Not on conversions being the best metric — but on not knowing what else to use. It’s a resignation: how Meta chooses is unknowable, so manage what you can spend. And the resignation isn’t his; it’s the industry’s. Ask why one video took all the money and you get folklore: that’s just the algorithm — it backs the ads it likes and pours in the budget. The fix on offer? More volume — test more, feed Meta more videos, let it choose. Somewhere along the way, “how does Meta decide which ad to scale?” became a question nobody expects an answer to.

It is not a small question to abandon. The pick decides where the budget goes; the budget decides which ad gets to produce results; the results are the business. Everything he banks flows downstream of a decision he’s agreed not to understand. His operation has no room for maybes — his words — yet it runs on the biggest maybe of all: which video Meta will back. That is the question we answered.

The answer

Stated plainly: Meta scales the best-watched ad. On one live account, across 33 head-to-head races, it gave most of the budget to the highest watch-rate video in 28 of them — 85% of the time.

Meta shows its hand every time it chooses between ads — you just have to be looking at the ads’ numbers when it does.

And the choosing happens in plain sight. Advertisers launch a handful of videos into one ad set, where they compete for a single budget. Meta tracks each one from the day it goes live, and when it finds one performing better, it scales it: most of the budget, most of the impressions. Every step — launch, tracking, verdict — leaves a record in the account. See the same pattern race after race, and you’re looking at a real signal — Meta’s machinery showing how it works.

We read those records on a real advertiser’s account — 33 races, inside ad sets that launched two, three, or four videos at once. Meta scaled the best-watched video in 85% of them — a pattern that consistent, across that many races, doesn’t happen by accident.

Every race met the same conditions: the videos started together; each drew at least a thousand impressions; one provably out-watched the rest; and decided meant one video holding seventy percent or more of the contest’s impressions — a bar the winners usually crossed in the first day or two. Re-scored on a stricter bar — that dominance held three days or more — the match with watch rate rose to 93%. Details in the last section.²

This isn’t a claim about Meta’s insides — nobody outside sees those. It’s the record of what Meta did: it concentrates budget on the videos that hold attention.

And we didn’t start with watch rate and work backwards. We put every metric an advertiser could reasonably watch through the same two questions: was it even legible when Meta made the pick — and if so, did it call the pick right?

Read the table left to right, the way his own argument runs. First: was the metric even legible when Meta made the pick? Watch rate always was — all 33 races. Then: when it was legible, did it call the pick? 85% of the time. Multiply the two, and you get the number that actually matters — across every race, watch rate was there and right 85% of the time. Nothing else is close. Watch depth, its own family, comes second at 76%. Conversions and clicks aren’t wrong — they call the pick 92% and 86% of the time — they just arrive later, legible in only about a third of races, which drags both down near 36%. The hook and the cheapest CPM fail the other way: wrong even when present — a coin flip, and a number that points backwards.

That’s the crack in his argument. At the moment that decided everything, his conversions had nothing to say — they hadn’t happened yet. Watch rate had. Whatever Meta was doing inside, watch rate was there on his own dashboard, already seeing the pick coming.

What the answer changes

Replay his argument with the answer in hand. He manages conversions because conversions are the business. But which video gets to produce them is Meta’s pick — and Meta’s pick tracks watch rate. So he’s better off reading watch rate than waiting on conversions. On three counts.

It reads first. Watch rate speaks from a video’s first days — every video, every race, in columns he already owns. Conversions, checked that early, spoke in barely one race in three. He still checks the winner against his CPA target; watch rate front-runs that check — it shows which way Meta is leaning days before conversions can, so he stops paying to run the dead videos while he waits.

It’s makeable. Watch rate is a property of the video — how it opens, how it holds, how it lands. A team can do something about it: brief for it, cut for it, test for it. “Make ads that convert” names a destination and offers no road. “Make videos people watch longer” is a brief a creative team can take.

And it pays. He asked for a link to what he spends and earns. Watch rate is it. The proof is next.

The link he asked for

He asked for a number tied to what he spends and what he earns. Watch rate is that number — and the link runs both ways.

Better-watched videos convert better. On the same calendar days, in the same placement, the better-watched video won 71% of nearly a thousand head-to-head matchups. On this account, doubling a video’s watch rate means install costs roughly a third lower in Instagram Reels, two-fifths in Facebook Reels, and half in Facebook Feed — more installs from the same impressions.⁴

And better-watched videos win their auctions with a lower bid. Meta’s auction doesn’t sell to the highest bidder — it favors the ad people are more likely to act on, and lets it win the slot for less.³ A better-watched video, converting better, is exactly the ad that rule favors.

So the link runs end to end. His budget follows watch rate in — Meta scales the best-watched ad, and wins its auctions with a lower bid. His money follows it out — the same impressions convert better, more installs once he has them.

What he’d do differently

Two changes.

Read watch rate from day one. Conversions take three or four days to stabilize; by then Meta has usually chosen. Watch rate shows the pick coming from the first day. So he reads it early and kills the videos headed nowhere, instead of paying to run them while conversions catch up.

Set his creative targets by watch rate. Watch rate is a brief his creative team can actually own. Conversions never were: they ride on the offer, the price, the audience, the landing page as much as on the video — most of what moves them was never the creative team’s to touch. Watch rate is different. It’s made of creative choices — how a video opens, holds, and lands — the part that’s genuinely theirs. Point the craft at the metric the craft controls, and the team finally has a real target to build toward.

He came with a fair challenge: tie the metric to the money, or don’t hand it to him. So here’s the link, end to end — one number, readable from the first day his ads go live, in a column he already owns. It moves his money three ways. It wins his auctions with a lower bid. It turns his impressions into more installs. And it tells him which videos to kill before he’s spent three days finding out. The lottery he’d made peace with was never a lottery. He just wasn’t reading the number that called the draw.

And it’s bigger than his desk: as far as anyone has published, no one had measured how Meta picks — until now.

For the statistically curious