Skip to content

Creative testing for DTC: a weekly system for Meta and TikTok video

Run creative testing as a one-week loop. Brief on Wednesday, produce through Monday, launch on Tuesday and read that batch the next Tuesday, after about seven days. Each ad tests one change to the angle, hook, format or offer, gets a budget set from your target cost per purchase, and lives or dies by kill rules written before launch.

Meta and TikTok both have built-in test setups, and both offer automation that can change the creative you’re testing.

In short
  • Run creative testing as a one-week loop: brief Wednesday, produce through Monday, launch Tuesday, read the batch the next Tuesday
  • Each ad tests one change to the angle, hook, format or offer, and lives or dies by kill rules written before launch
  • New ads per week = weekly spend × 20% ÷ (target cost per purchase × 3). Count ads, not ideas
  • Turn Advantage+ and Smart+ enhancements off for tests, or they change the thing you are testing
7 daysminimum Meta recommends for an A/B test. Shorter tests may be inconclusiveMeta
5%chance an on-target ad shows zero purchases after spending three times target. That is your kill linePoisson, e to the power of minus 3
0.54 to 2.09×the range a 10-purchase cost per purchase can really be. At 30 purchases it narrows to 0.70 to 1.48×95% Poisson intervals

The weekly cadence

Three batches run at once: one being read, one launching, one in production.

The one-week loop
Wed: briefiterations on winners plus new angles
 
Thu to Mon: producecaptions, disclosures, pages, naming
 
Tue: launchtest ads start on the ad set schedule
 
Next Tue: read and decidekill and scale rules, owner and date
Hands off in between apart from hard kills. Editing during the learning phase resets learning.
  1. Tuesday morning, read last Tuesday’s batch into the scorecard. Meta recommends A/B tests run at least 7 days, as shorter tests may produce inconclusive results.
  2. Tuesday, decide. Apply the kill and scale rules, with an owner and a date for each decision.
  3. Tuesday afternoon, launch this week’s batch. In a Meta creative test, test ads start on their ad set’s schedule once published, and the test schedule only sets the window Meta measures.
  4. Wednesday, brief next week’s ads: iterations on winners, plus a few new angles.
  5. Thursday to Monday, produce and check captions, disclosures, landing pages and naming. New hooks from existing footage fit this window, but brief creators two weeks ahead for fresh footage.
  6. In between, hands off apart from hard kills. Meta says editing during the learning phase resets learning.

What a testable hypothesis looks like

Change one variable and hold everything else. Meta says results are more conclusive when versions are identical except for the variable you’re testing. Choose from four:

  • Angle: the reason to buy, such as time saved or a better deal.
  • Hook: the first two or three seconds, meaning opening line, first frame and on-screen text.
  • Format: creator to camera, hands-only demo, comparison or founder story.
  • Offer: price, bundle, shipping threshold or guarantee.

Write each test as one sentence: if we change [variable] from [control] to [variant], [metric] will improve because [evidence from reviews, tickets or past tests]. An offer test changes margin too, so judge it on contribution after ad spend, not cost per purchase alone.

Automation can break the one-variable rule. Meta says some Advantage+ creative enhancements may be on by default, and TikTok says Smart+ Automatic Enhancements can refresh music or hooks. Turn enhancements off for tests, or match them across test ads. Our Meta Advantage+ post covers the campaign side.

How many new concepts should you ship each week?

New ads per week = weekly spend × test share ÷ (target cost per purchase × 3)

Meta’s creative test setup suggests no more than 20% of the existing budget, so use 20% as the test share. The 3 is probability. If an ad’s true cost per purchase is exactly on target, the chance of zero purchases after spending three times target is about 5% (e to the power of minus 3), against about 14% after spending only twice target. So zero purchases at three times target is strong evidence.

Three illustrations with made-up targets, rounded down:

Daily spend Target cost per purchase Weekly test budget at 20% New ads per week
$500 $40 $700 5
$1,500 $45 $2,100 15
$5,000 $60 $7,000 38

Count ads, not ideas: a concept with three hooks counts three times. Meta allows 2 to 7 test ads per creative test, so the middle row needs three tests a week. You set the share per test, so tests on one budget split the 20%, about 7% each for three. At the $5,000 row, production is the limit. At the $500 row, five ads is about two concepts. In every row an even split gives each ad about three times target in a week, so an on-target ad lands about 3 or 4 purchases, and a 10-purchase read takes about three weeks. Set the target from first-order contribution and payback, as our ROAS, MER and POAS post explains.

New ads a week at a 20% test share
 0 ads10 ads20 ads30 ads40 ads$500 a day, $40 target $500 a day, $40 target: 5 ads 5 ads$1,500 a day, $45 target $1,500 a day, $45 target: 15 ads 15 ads$5,000 a day, $60 target $5,000 a day, $60 target: 38 ads 38 ads
 0 ads10 ads20 ads30 ads40 ads$500 a day, $40 target $500 a day, $40 target: 5 ads 5 ads$1,500 a day, $45 target $1,500 a day, $45 target: 15 ads 15 ads$5,000 a day, $60 target $5,000 a day, $60 target: 38 ads 38 ads
Made-up targets, rounded down. Each ad gets about three times target in a week, so an on-target ad lands 3 or 4 purchases and a 10-purchase read takes about three weeks.

Test setups on Meta and TikTok

Meta: creative tests and A/B tests

Meta’s page on setting up a creative test says you can test new creative inside an existing campaign, so winners keep the delivery system’s learnings. At the ad level, open the Creative test section, click Set up test and make 2 to 7 copies of the ad for your new creative. The campaign must use the Highest volume bid strategy. Meta names no other, so if you bid for value, check that the Creative test section appears, and use an A/B test if it doesn’t.

You choose how much of the existing campaign or ad set budget to spend across the test ads. Meta suggests no more than 20% and aims to spend that amount per day. Test ads may get more if no other ads are active. Meta doesn’t promise an even split or separate audiences, so check spend per ad after day one.

At the end, Meta emails any high performers and shows results in Experiments and under the beaker icon in Ads Manager. Test ads keep running on the original budget until you turn them off, and the results don’t include a confidence level.

For questions bigger than one ad, such as creator-led against studio-shot, use an A/B test. It splits your audience so nobody sees both versions, Meta recommends equal budgets, and tests run 1 to 30 days.

TikTok: split testing

TikTok’s split testing overview says it splits your audience into two equal groups that each see only one ad group. Creative is one of the test variables, and so is Smart+. TikTok uses a 90% confidence rate to pick the winner, which you can keep running. The overview doesn’t say how budget is set or divided.

TikTok: Smart+ creative options

TikTok’s May 2026 Smart+ guide says a campaign can hold up to 30 ad groups, each with up to 30 asset groups of up to 50 creatives, and reports by creative, text and enhancement. Three features change your creative:

  • Automatic Enhancements improves quality, resizes for vertical, refreshes music or hooks, and translates and dubs into more than 50 languages.
  • Recommended Creatives generates videos from your landing page, virtual avatars included.
  • Auto-add refreshes creatives while the campaign runs.

TikTok describes Automatic Enhancements and Auto-add as toggles. Keep them off while you test.

Which metrics to read, and when

Metric How to calculate it Read it after
Hook rate Meta: 3-second video plays ÷ impressions. TikTok: 2-second video views ÷ impressions. About 5,000 impressions
Hold rate Meta: ThruPlays ÷ 3-second video plays. TikTok: 6-second ÷ 2-second video views. About 5,000 impressions
Link CTR Link clicks ÷ impressions 10,000 to 20,000 impressions
Cost per purchase Spend ÷ purchases 10 purchases for a rough read, 30 to decide

Meta counts a 3-second video play at three seconds, excluding replays, and a ThruPlay at 15 seconds or completion. TikTok lists no 3-second metric. Its 2-second and 6-second video views are the closest, and the 6-second count includes anyone who engaged in the first six seconds. So don’t compare hook rates across platforms, and since a ThruPlay is a completion for ads under 15 seconds, compare hold rates only between ads of similar length.

Meta’s link CTR is link clicks over impressions. On TikTok, divide clicks (destination) by impressions. Judge cost per purchase against target and against breakeven, the most a first order can cost before it loses money.

When a read is noise

By standard 95% Poisson intervals, 10 purchases put the true cost per purchase anywhere from 0.54 to 2.09 times what you see, so a $45 result might really be $24 or $94. At 30 purchases the range is 0.70 to 1.48 times.

How wrong a $45 cost per purchase can be
 $0$25$50$75$10010 purchases 10 purchases: $24 to $94 $24 to $9430 purchases 30 purchases: $32 to $67 $32 to $67
 $0$25$50$75$10010 purchases 10 purchases: $24 to $94 $24 to $9430 purchases 30 purchases: $32 to $67 $32 to $67
The dot is what you see. The bar is where the true number can sit at 95% confidence. Ten purchases is a rough read; thirty is a decision.

Attention metrics settle faster. At 5,000 impressions, a 30% hook rate has a 95% margin of about 1.3 points, so 30% against 25% is a real gap. A 1.2% CTR against 0.9% at 5,000 impressions each could be chance, and at 20,000 each it probably isn’t. Treat these as best cases, since one person can see an ad several times.

Kill and scale rules

Write these into the brief before launch.

  1. Kill anything disapproved, broken or pointing to the wrong page at once.
  2. Kill on attention. After about 5,000 impressions, cut an ad whose hook rate is clearly below your recent median, outside the margin above, unless its CTR beats your median.
  3. Kill on purchases: zero after spending three times target cost per purchase.
  4. Kill on cost: 10 or more purchases at over twice target. Even the optimistic 0.54 multiple leaves that above target.
  5. Keep everything else running until it reaches 30 purchases.
  6. Scale ads that hold at or under target for 30 purchases and a second week. Leave them in place, since Meta notes that merging test ads into another campaign resets learning, and raise budgets in steps, since frequent budget changes can restart learning.
  7. Don’t pause the old winner because a new ad beat it. Meta’s creative fatigue guidance says keeping the original active may maximize results.

Iterating on a winner

Make the winner the control and change one thing per round: new hooks on the same body, then a shorter cut or new format, then a new creator on the same script, then the offer. Meta flags Creative limited when cost per result is above your past ads but under double, and Creative fatigue at double or more, but only in ad sets with one creative. Otherwise, watch frequency and cost per purchase yourself. Meta’s first suggested fix is a materially different ad, not a tweak.

Creator content: rights and disclosure

Get the license in writing

  • Paid usage: platforms, countries and term.
  • Handle access: running ads from the creator’s account, and for how long.
  • Edit rights: raw footage, cutdowns, new hooks, subtitles and translations, including those platform AI makes.
  • Exclusivity: which competitors, and for how long.
  • Renewal price, so a winner isn’t repriced mid-scale.
  • Disclosure wording, and your review before posting.

This isn’t legal advice. Have a lawyer review the template once.

Meta partnership ads

Branded content ads are now called partnership ads. They show the creator’s account, yours or both in the ad header and use signals from both accounts. You need the creator’s permission, which Meta says they can revoke at any time. Content-level permission covers one post, through the Allow brand partner to boost toggle on a paid partnership post or a partnership ad code. Account-level permission lets you create ads from the creator’s handle without an existing post and include or exclude their custom audience.

TikTok Spark Ads

Spark Ads run organic posts as ads, and engagement from the promotion goes to the original post. The creator generates a code in the TikTok app (three dots, Ad settings, Generate), you add it in Ads Manager, and the authorization period shows under each post. A private video becomes public once it’s used in a campaign. TikTok’s branded content policy requires creators to turn on the commercial content disclosure toggle for branded posts.

FTC disclosure

In the US, the FTC’s Endorsement Guides, updated in 2023, say an unexpected connection between endorser and seller that could affect how much viewers trust the endorsement must be disclosed clearly and conspicuously. The FTC’s guide for influencers counts payment, free product, and employment, personal or family ties. For video ads:

  • Disclose in the video, not just the caption, and in both audio and on screen if the endorsement is spoken and shown. On social media, the Guides say it should be unavoidable.
  • Don’t rely on a platform’s built-in label. Asked if that’s enough, the FTC answers not necessarily.
  • If a testimonial-style ad uses actors, say they aren’t real customers.
  • Never present AI avatars, or people who didn’t use the product, as customers. The FTC’s rule on consumer reviews and testimonials bans creating such testimonials, and buying or running them when a business knew or should have known.
  • Brief and monitor creators. The Guides say an advertiser can be liable even when the endorser isn’t. Outside the US, check local rules too.

A creative brief template

  1. Test ID, named date_angle_hook_format_creator.
  2. Hypothesis, in the one-sentence form.
  3. Variable under test, and what stays fixed.
  4. Customer and problem, in their words from reviews or support tickets.
  5. Angle: the one reason to buy.
  6. Hooks: two or three, each with first line, first frame and on-screen text.
  7. Format and length, with text clear of the platform’s buttons.
  8. Offer and landing page, which must repeat the ad’s promise (see our landing page checklist).
  9. Proof and claims you can substantiate.
  10. Creator, usage rights, handle access and disclosure wording.
  11. Primary metric and kill line.
  12. Owner, due date and launch date.

A weekly scorecard template

One row per ad, filled in every Tuesday.

Column What goes in it
Test ID Matches the brief
Variable Angle, hook, format or offer
Spend and impressions This week and since launch
Hook rate, hold rate, CTR Beside your recent medians
Purchases and cost per purchase Since launch, beside target and breakeven
Read quality Noise, directional or decision
Decision Kill, keep, iterate or scale
Owner and date Who does what, by when
Next test The hypothesis it feeds

If you want help

Our growth audit ($1,000) reviews your ad accounts, tracking, creative and landing pages against margin and sets the breakeven and target numbers this system runs on. Growth Marketing runs it for you from $1,000 a month, and the Growth plan from $3,500 a month adds creative production. Book a 30-minute call to start.

Get the breakeven and target numbers this system runs on

The growth audit reviews your ad accounts, tracking, creative and landing pages against margin. $1,000, two to three weeks. Growth Marketing runs the loop for you from $1,000 a month, and the Growth plan from $3,500 a month adds creative production.

Sources

Facts checked on 5 October 2026.

Contact Us