Comment tester efficacement sur Meta Ads (et payer moins cher)

Summary

J7 Media’s central argument is that continuous, structured creative testing is essential to keeping Meta Ads efficient. Meta is an auction-based platform: past performance does not guarantee future costs, and ads that lose relevance can attract fewer clicks, raise acquisition costs, and increase frequency.

The agency says it manages more than $50 million annually on Meta Ads. It also cites a Meta study attributing 57% of campaign effectiveness to factors outside advertisers’ control, with creative identified as the biggest success factor.

Account-Maturity Framework

J7 Media groups accounts by the stability and number of proven winners—not by budget alone:

• Young: KPIs and conversions are not yet stable; there is little historical data and no clear winning ads or offers. Often around €100/day or less.
• Intermediate: KPIs and conversions are stabilizing, with 1–3 winning elements and meaningful account history. Often around €500/day.
• Mature: KPIs are highly stable, conversion volume is increasing, the best products and offers are identified, and there are at least five winning elements. Often around €1,000/day.

An account spending less can still behave like a mature account if it has several proven winners.

The agency rejects the idea that one campaign should handle both testing and scaling. Its examples include an account spending $16,000/month across five active campaigns and another spending €86,000 in 30 days across 16 campaigns. In its approach, higher spend generally means more active, dedicated campaigns.

Testing Campaign Playbook

• Optimize for the deepest meaningful conversion in the business model. The speaker discourages optimizing for traffic, add-to-cart, or poorly qualified leads when better conversion signals are available.
• Use a campaign-level Advantage+ budget so Meta can allocate spend among ads and accumulate learning within a shared campaign, rather than fragmenting learning across many ad-set budgets.
• Set a budget that can generate several conversions per day. For a young account, the speaker suggests €20–€40/day if higher spend is unsuitable. Avoid spending €100–€200/day when the cost per conversion is much higher than €80–€100.
• For a young account, initially test just two audiences: broad and one relevant interest audience. Once the better audience is identified, use it for later tests instead of repeatedly testing audiences. More mature accounts should start with their best-known audience.
• Exclude existing customers for acquisition. Early-stage accounts may also include site visitors to establish sales, but should move toward cold audiences.
• For attribution, the speaker recommends 7-day click/1-day view for young accounts. As spend grows, click-only attribution may provide a closer view of performance.
• Start with 2–3 ads. At around €200/day, the agency may test 5–10 ads, depending on CPA.
• Give each test batch a dedicated campaign or ad set and budget. Don’t turn off a working ad just to make room for a new test; build a new campaign around it. If one ad takes most of a test campaign’s spend, move under-delivering ads into a new test campaign.
• The agency says audience overlap is not a major concern for its larger accounts because Meta can handle campaigns targeting the same audience.

Deciding Whether an Ad Is a Winner

Judge ads by spend, time in delivery, conversion volume, and CPA. A good early CPA alone is not enough: with too little delivery, one additional conversion can substantially change the average.

• For e-commerce, J7 Media generally wants at least three days of delivery, more than 10 purchases, and a CPA comfortably below the business’s maximum.
• For lead generation with leads costing less than about $10/€10, it looks for roughly 20–30 conversions at a stable CPL.

In an example e-commerce account with a maximum allowable CPA of $30:

• An image spent about $400 over seven days, generated 13 purchases, and averaged $30 CPA. It met the volume threshold but was already at the maximum CPA, leaving little room to scale.
• A video spent $220 over four days, generated nine purchases, and averaged $24 CPA. The recommendation was to wait for more than 10 purchases while maintaining a comfortable CPA.
• A carousel spent $70 over two days, generated three purchases, and had a low reported CPA. However, it was too early to judge because the sample was small.

Operating principle: Wait for stable, profitable performance before declaring a winner or scaling it.

What to Test, in Priority Order

1. Offer, product, service, or subcategory
2. Marketing angle
3. Who the ad speaks to—the intended customer persona, not simply platform targeting

An ad can be broken into components for testing: offer, product, angle, audience, concept, format, and—particularly for video—hook.

Examples

• E-commerce offers: One advertiser tested free delivery on its best product. It performed strongly at first, with an average CPA of about $52, but later cooled. A gift-with-purchase offer was also tested as an incentive that avoided cutting the product price.
• Consumables retailer: Each ad set tested one product through several ads, helping the team identify which product attributes merited further creative development.
• Car insurance lead generation: A review of five angles found “low price/best price” below 2 ROAS, while “save money” and “stop overpaying” exceeded 2 ROAS. The team inferred that the strongest audience was likely people switching because they felt their current insurance cost too much. It planned campaigns dedicated to that angle in 2026.
• Beauty e-commerce: Ads were built for distinct customer segments, such as younger women, skeptical buyers, women with Black or mixed-race skin, and people concerned about water retention. The creative was tailored to each segment—for example, by using models with relevant skin tones—to improve relevance and learn more about customer groups.

Creative Testing and Iteration Guidance

• Make test ads meaningfully different in format and concept. Repeating the same format when it is not working wastes time and budget.
• Avoid “phantom testing”—combining multiple creatives or copy variations inside a flexible ad and then treating the result as a controlled test. The speaker argues that this obscures which combination drove performance.
• Build ads that can engage cold audiences, not only people who already know the brand or understand the product. The speaker estimates that in 90% of cases, advertisers assume too much prior awareness. This can produce ads that work for the first 48 hours and then fade.
• Example: A cookware advertiser’s existing ads assumed viewers already understood the product. The recommendation was to explain the cooking problem immediately and give unfamiliar prospects a reason to enter the brand’s ecosystem.

Once an ad wins, extend its performance by isolating what worked—such as the hook, angle, concept, design, or format—and making variations. To reduce the risk that Meta treats a variation as too similar, the speaker recommends changing at least two major elements. Examples include changing the product image and layout, or changing a video’s hook, script, and shot order while retaining the core story.

Scaling Playbook

• Intermediate accounts should use proven elements to extend performance and grow spend gradually.
• Once an account has several stable winners and spend has reached roughly €700–€800/day, the agency recommends a more aggressive scaling step.
• Put stable winning ads into a new, high-budget campaign, starting at 2–3 times the usual minimum test-campaign budget. The rationale is that the ads have already been developed through successive tests of products, personas, sub-personas, and angles.

Presenters and Sources

• Presenter: The speaker is not named in the subtitles; they speak on behalf of J7 Media / Social Scaling.
• Sources and examples mentioned: A Meta study on campaign effectiveness; J7 Media’s account audits and client examples; e-commerce advertisers selling wedding rings, consumables, beauty products, and cookware; a car-insurance lead-generation advertiser; and Calm, used as an example of a UGC-style video ad.