Automatically cutting every ad below target ROAS looks disciplined. Often it protects budget. Sometimes it destroys valuable information—or creative capable of supporting growth.
An ad does not directly produce ROAS. It enters a chain: impression, attention, click, visit, intent, purchase. ROAS gives the final result without locating the broken link. Before cutting, ask: where does the creative lose momentum?
ROAS delivers a verdict, but does not explain the failure
ROAS (attributed purchase value / spend) shows the revenue Meta claims for each euro invested. It locates neither lost clicks nor abandoned carts.
It does not measure profit. Margin, VAT, returns, payments and logistics can turn flattering ROAS into a loss. In lead generation, diagnosis fails without qualified leads or sales fed back from the CRM.
The figure also depends on attribution windows. Meta allows comparing post-view and post-click results, including over 1 or 7 days. The same creative can therefore look different across time windows.
ROAS is a warning light. It does not identify which engine part is overheating.
The advertising chain reveals where performance breaks
Analysing each level in order avoids blaming creative for site failures or protecting a poor ad behind an imaginary problem.
| Link | Metrics to watch | Likely failure |
|---|---|---|
| Delivery | CPM, reach, frequency, placements | Auction, audience or delivery |
| Attention | 3-second views, hook rate, hold rate | Hook, pace or readability |
| Click | Link CTR, outbound CTR, link CPC | Insufficient promise or desire |
| Arrival | Landing page views / link clicks | Slow loading, redirects or tracking |
| Intent | ViewContent, add to cart, checkout, lead | Offer, proof, price or qualification |
| Conversion | Purchase rate, CPA, average order value, ROAS | Delivery, payments, trust or sales |
Hook rate and hold rate: agree on a definition
Meta defines three-second video views and ThruPlays, but no universal formula called ‘hook rate’ or ‘hold rate’.
A team may calculate:
- Hook rate = 3-second views ÷ impressions
- Hold rate = ThruPlays ÷ 3-second views
Other dashboards use two or fifteen seconds. A calculation changing definition between reports is worthless. Compare similar lengths, formats, placements and markets: a seven-second Story differs from a forty-second Feed demonstration.
Two creatives with identical ROAS may require opposite decisions
Imagine two ads with 1.2 ROAS—an illustration, not a benchmark.
The first holds little attention, has weak link CTR and generates expensive carts. The chain breaks at the top: cut or rebuild it.
The second stops scrolling, generates qualified clicks and fills carts competitively. Only purchase drops off. Deleting it would hide the cause.
E-commerce example in Israel
A Hebrew video advertises a product for ₪199, with free delivery in Israel. After clicking, the page opens in English, displays euros and adds ₪39 at checkout. Despite good CTR and low-cost carts, conversion collapses: the journey contradicts the promise.
Likewise, a Facebook ad localised for Israel may reveal a two-week delivery at checkout. Changing its first three seconds fixes nothing.
France–Israel lead generation example
A campaign targeting French-speaking olim feeds the CRM at acceptable CPL. But sales fall: late callbacks, Hebrew-speaking staff, no WhatsApp contact. Cutting creative before auditing sales follow-up targets the wrong link.
Discovery creative can help, but ‘TOFU’ is not immunity
In the carousel, a creative showed 1.5 ROAS against a target of 2. Hook, retention, CTR and cart cost stayed strong; only conversion fell. Kept running, it spent €150,000 while the account exceeded 2.2 ROAS. Its role was feeding discovery and retargeting.
The decision paid off. But coexistence with a profitable account does not prove causal contribution: other ads may compensate. Comments prove attention, not incremental sales.
TOFU creative deserves a measured role, not unlimited tolerance.
Give it a reprieve if it recruits new users, generates qualified visits and improves overall results. Without proof, ‘it drives discovery’ becomes an excuse.
How to prove creative contributes to later sales
First compare attribution windows. A gap between 1-day and 7-day click signals timing to investigate, not proven causality.
Then combine evidence:
- new and returning visitors;
- new-customer share and margin generated;
- growth in branded searches;
- assisted purchases and engaged-audience growth;
- pause-and-restart tests over comparable periods;
- geo-tests or control groups when volume permits.
Meta presents Conversion Lift as a way to measure incremental impact by comparing exposed and control groups. For a more accessible hypothesis, its A/B testing tool isolates one variable between two strategies.
Do not judge creative over seven days if buying takes thirty. Base the delay on the real cycle, not a desire to rescue a loser.
Without reliable tracking, the funnel is fiction
In Events Manager, check ViewContent, AddToCart, InitiateCheckout and Purchase, then value, currency and product identifiers. Meta recommends the Pixel and Conversions API to support measurement and optimisation; verify deduplication.
For e-commerce in Israel, audit:
- correct ILS currency reporting
- domain or language changes
- late-added delivery fees
- declined or abandoned payments
- consistency across Shopify, GA4 and Meta
For lead generation, send back valuable stages: qualified lead, appointment, sale, revenue. Low CPL with no sales is not a creative victory, but an incomplete signal.
Keep, revise or cut: the decision rule
Cut
When spend is sufficient relative to target CPA, but creative earns no attention, qualified clicks or defensible intent.
Revise
When the hook mainly attracts curiosity. Show the price, qualify the audience or realign ad and landing page.
Keep under observation
When failure comes after strong intent. Cap budget while auditing the site, offer or sales follow-up. A reprieve without a protocol becomes endless spending.
Scale
When creative demonstrably contributes, directly or indirectly, to CAC, margin or incremental growth. Intermediate metrics guide analysis; they do not replace economic validation.
Checklist before cutting a Meta ad
- Is spend sufficient relative to target CPA?
- Are hook and hold rate formulas stable?
- Are you analysing link CTR rather than all clicks?
- Do clicks become landing page views?
- Where do add-to-cart, checkout and purchase drop off?
- Are price, currency, language and delivery consistent after clicking?
- Are Pixel, CAPI, values and deduplication reliable?
- Is the discovery role proven or merely claimed?
- Which asset will receive the freed budget?
Poor ROAS starts analysis; it does not finish it
The basic rule is sound: an ad producing no attention, intent or sales should make way. Applying it to ROAS alone is like repairing a chain by throwing away the first visible link.
For a French brand entering Israel or an Israeli business targeting French speakers, diagnosis must include language, shekels, delivery, landing page and sales handling. These frictions can destroy performance correctly initiated by the ad.
As a France–Israel paid acquisition agency, Alyads connects creative analysis, tracking, CRO and sales reality. The right reflex is not ‘cut or keep’, but: prove, then decide.
FAQ — Meta Ads creative analysis
After how much spend should you cut a creative?
There is no universal amount. It depends on target CPA, conversion rate, event volume and variance. A few euros cannot judge an offer with high expected CPA.
What is a good hook rate?
No cross-category benchmark exists. Compare identical formulas across similar length, format, placement, objective and market. Do not mechanically transfer French history to Israel.
Why can good CTR produce poor ROAS?
The promise may attract unqualified curiosity; the page may contradict it, load slowly or reveal price, delivery or payment friction.
How do you evaluate top-of-funnel creative?
Combine attribution windows, new visitors, cohorts, branded searches, margin, pause tests and, with enough volume, lift measurement or control groups.
