A watched video is not a sold product, and a sold product is not necessarily a profitable sale. Say you sell a $79 serum: your Meta video pulls 42,000 views in a week, Shopify records 31 orders, and the report still cannot tell you whether the advertising earned its keep.
Those numbers describe different events. Before I hire anyone to improve ROAS, I want four questions kept separate: did people pay attention, did the platform attribute purchases, did advertising cause extra sales, and did those sales leave margin?
Free tool: the ROAS Calculator works out your return on ad spend, your break-even ROAS and the budget a revenue goal takes.
Meta’s latest attention research makes that distinction more important, not less. The company is testing Aggregated View Time, an early-alpha tool that adds up the time a person spends with a brand across a week, rather than grading each view on its own. Alongside it, Meta shared research claiming five seconds of attention accumulated across exposures drove 32% higher offline sales than one continuous five-second view.
That is a study result, not a forecast for the serum account. I’ll keep that $79 bottle on the counter throughout this piece, because the easiest way to get lost in measurement is to forget what the business actually sells.
Start with attention: a timer is not a receipt
A view records time spent with an ad. It does not record why someone stayed. The serum clip can collect 10,000 three-second views from people who find the before-and-after oddly satisfying and have no intention of buying skincare that week.
That is the first distinction: watchable creative and persuasive creative are not the same thing. Strong watch time tells me the ad held attention. It does not, by itself, tell me the ad did its sales job.
Accumulation is still worth understanding. Independent attention research found that cumulative attention explained eight times more brand-metric movement than frequency alone. In one comparison, two campaigns both at frequency 5 delivered 9 seconds versus 71 seconds of actual attention. Same frequency. Very different exposure.
The same work put median thresholds at around 21 seconds for awareness and 47 seconds for purchase intent. That does not give the serum a stopwatch setting for guaranteed demand. It shows why counting impressions or isolated views can miss how much exposure someone actually receives.
My takeaway: accumulated attention can make the exposure picture more useful. It does not turn that picture into a sales ledger.
What Meta’s 32% result does—and does not—give you
Aggregated View Time adds up the seconds one person gives a brand across several impressions in a week. The total becomes the unit of analysis, rather than a single three-second view.
Meta is testing it as an early-alpha measurement tool with no general-availability date. It is not a generally available Ads Manager setting or something you can simply choose to bid toward. Treat it as a research direction, not an optimization lever.
The 32% figure needs the same discipline. Meta reported higher offline sales from five seconds split across exposures than from one continuous five-second view in a specific study. The coverage also noted that those figures reflect specific studies, not guaranteed outcomes.
Other research makes the limit plain. A study with Kroger Precision Marketing found little to no correlation between attention metrics and sales lift. Even the often-quoted 8x attention finding explained only about 8% of survey-based brand lift, not sales.
For the serum, accumulated seconds might help explain recognition or intent. They still leave three questions unanswered: which purchases got credited, which purchases were additional, and what those additional purchases earned.
Four questions hiding inside “Did the ad work?”
I use this chain in order. Each step asks something the previous step cannot answer.
1. Attention: did someone spend time with the ad?
For the serum, this includes accumulated seconds, ThruPlays, and three-second views. These describe exposure, not checkout behavior.
The distinction shows up in broader account data. Across 88,329 sales-objective Meta video ads and $122 million in spend, hook rate rose from 24.42% to 25.44% while average ROAS fell from 2.41 to 2.17. Health and Beauty had the highest CTR and only 1.82 ROAS in that set.
Those figures do not explain why every account changed. They do show that stronger attention metrics and weaker sales returns can coexist. A better hook is not automatically a better business result.
2. Attribution: did the platform connect an exposure with a purchase?
Attribution is a counting rule. It decides which purchases qualify for credit after an interaction with an ad.
Meta’s reported default for website purchases is 7-day click, 1-day engage-through, and 1-day view-through. Its reporting rules have also changed: longer view-through windows stopped returning data in the Ads Insights API, and click-through was redefined around link clicks, with five-second video views moved into engage-through.
For the serum owner, the practical issue is comparability. A report can change because the counting rule changes, even when the underlying business does not. A buyer who watches, clicks, and purchases can receive different treatment depending on the window selected.
Before comparing ROAS, write down the attribution window. Otherwise, I am comparing labels rather than performance.
3. Incrementality: would the purchase have happened anyway?
Suppose Meta attributes 31 serum orders to advertising. Maybe 19 buyers already intended to purchase after a friend recommended it, then saw a retargeting clip before checking out.
The exposure can qualify for attribution without creating the sale. That is why attributed ROAS cannot answer the causal question.
A lift test uses a randomly assigned test group that can see the advertising and a holdout group that cannot, then compares purchases. The difference estimates the additional sales caused by advertising. Strong Ads Manager ROAS can coexist with low lift, especially when retargeting reaches buyers already likely to convert.
Attribution assigns credit; incrementality tests whether the advertising made a difference. I need both concepts, even when I cannot run a credible lift test immediately.
4. Economics: did the extra sales leave enough margin?
Now put the bottle back on the counter. Say the $79 selling price has these costs:
- $24 for product, pick, pack, and shipping.
- $11 for fees and expected returns.
- About $26 in advertising at a 3.0 ROAS.
That leaves about $18 before overhead. If ROAS slips to 2.2, advertising costs about $36 per bottle, leaving roughly $8.
At those margins, selling 40% more bottles still produces less total contribution. More orders, less money left. Not much of a victory lap.
Translate ROAS into dollars kept per order. Even genuinely incremental sales can disappoint if acquiring them consumes the margin.
Why optimizing views can disappoint the person paying for them
If I ask Meta to find more three-second views, I am asking it to find people likely to give three-second views. I have not asked the same question as “Who will buy the serum profitably?”
The satisfying texture video may attract skincare enthusiasts who watch every routine and purchase nothing. Hook rate climbs, cost per view falls, and the order rate goes nowhere. The optimization target matters because it defines what counts as success.
Repeat exposure creates a second complication. It may build recognition in a new buyer, or remind an existing customer to reorder. Both can be useful. They are not interchangeable when I am trying to acquire customers.
Retargeting past buyers and site visitors can produce attractive reported returns while claiming credit for purchases already likely to happen. View-through attribution on past visitors is particularly vulnerable to this inflation.
Then the reporting window adds another layer. The same week might show 3.1 ROAS on 7-day click plus 1-day view and 2.2 on click-only. Those are different counts, not competing verdicts on causality. Click-only is useful for comparison, but a click does not prove the sale needed the ad either.
I want the owner to know which number improved, why it improved, and whether that improvement survived the trip to margin.
What I would ask a Meta agency to put in writing
A company promising better ROAS should explain how it will work through the chain, not jump straight to a creative deck. I would start with four questions:
- Is purchase tracking reliable? Are the pixel and Conversions API sending purchase value and currency, with each purchase counted once?
- Are new and existing buyers separated? Prospecting should not hide behind reorders.
- Which attribution window judges performance? Put it in the proposal, not just the dashboard settings.
- What is the comparable baseline? Show the last 30 days on click-only as well as the agreed reporting window, without pretending either proves incrementality.
That is the instinct behind my pre-flight checklist before trusting ROAS: establish what the numbers mean before promising to improve them.
If those answers are missing, I cannot tell whether the ROAS promise describes a performance improvement or a reporting choice.
Next, I want creative decisions tied to purchase evidence. Keep the serum comparison simple: the same audience, budget, and week, with two hooks. Version A is the satisfying texture close-up. Version B shows a hand holding the bottle next to a 30-day calendar, with the price on screen.
For that comparison, I would judge cost per new-customer purchase on a consistent click-only window, not hook rate. Then I would translate acquisition cost into contribution per order. The prettier version might win both. It might not.
The rule is straightforward: cheaper views do not earn more budget unless the sales economics support it.
Then I want a credible incrementality comparison. A small holdout can help, but “turn something off somewhere and see what happens” is not automatically a lift test. The comparison needs a group eligible to see the advertising and a comparable group held out from it, with purchases measured across both.
I would not divide purchases by people reached: the holdout is deliberately not reached by the advertising being tested. The question is whether the test group produces additional purchases relative to the holdout, not whether exposed people look good in a platform report.
If Meta reports 4x ROAS while Shopify shows 1.8x, I would reconcile the purchase records and reporting rules before scaling. Shopify’s order count matters to the business, but neither dashboard alone tells me how many sales the advertising caused.
Finally, I ask who does the work, how quickly, and where I can see it. Who can pause the texture video on a Tuesday when new-customer cost per purchase rises 20%? Who investigates when Shopify and Meta disagree? Where is the record of what changed and why?
Periodic reviews and percentage-of-spend fees do not answer those questions. Spending more pays the vendor more, whether or not the owner keeps more margin. I want continuous execution, a plain-language action log, and one human who owns the outcome.
No service can guarantee Meta’s 32% study result or a fixed ROAS next month. A vendor does not control the serum’s price, repeat rate, or how many buyers were already coming. If the pitch skips those limits, I would leave.
Where groas fits—and how I would judge the next $5,000
The connection to groas is the operating principle, not a promise to run this Meta workflow: continuous execution, visible changes, and human accountability rather than a monthly explanation of the dashboard. Execution should serve the business outcome, not make the first metric look impressive.
That distinction also matters when an agency can raise ROAS while new-customer growth stalls. An average can improve because cheap reorders replace more expensive new buyers. The report looks better while the acquisition job goes unfinished.
Now give the serum owner $5,000 for next month and a requirement to keep at least $15 per bottle before overhead. I would reason through the same four steps:
- Attention: use accumulated seconds to diagnose whether the creative holds people, not to forecast orders.
- Attribution: compare new-customer acquisition costs under a consistent window, keeping reorders separate.
- Incrementality: use a credible holdout comparison to check whether the advertising creates additional purchases.
- Margin: keep only the spending that clears the owner’s contribution target.
The earlier economics make the last decision concrete. At roughly $26 in advertising per bottle, the serum keeps about $18 and clears that $15 requirement. At $36, it keeps about $8 and fails—even if accumulated attention improves.
That is where the model earns its keep. A week of watching may explain something useful about exposure. If it cannot clear attribution, incrementality, and margin, it stays a viewing stat.

