How to Run a 7-Day Creative Test Sprint End to End

Write one product, one offer, one landing page, three genuinely different hooks, and two avatars. That is the 6-creative matrix. Add a kill rule in the same note: after 48–72 hours, pause anything with hook rate under ~20% or spend past 1.5× target CPA and no conversions. If the brief is still a Slack thread, the sprint has not started — put it in a dated doc before anyone renders.

The three hooks must disagree: a question or complaint, a proof or number, and a pattern interrupt or callout. Two avatars should differ on age band, setting, or energy so a casting signal can show up. Hold the body script, captions, length, and destination identical. Changing the discount on variant five 'because it might help' contaminates the week. Put names on roles even if one person wears all of them: brief owner, producer, media buyer, readout lead. A sprint with no named readout time on the calendar is a batch that will be judged from a phone at midnight.

Budget the week backwards from signal, not from leftover media. You want roughly 1,000 impressions per variant, which for many accounts is a modest equal daily budget across six ads in one ad set. Write the daily number, the test campaign name, and the Friday readout slot into the brief before anyone renders a frame. If the budget cannot support six variants, cut to three hooks on one avatar rather than starving every cell of the grid. A thin six-cell test is how you get six noisy stories and no decision.

Render the full 3 × 2 grid before anyone opens Ads Manager. Do not launch a partial batch and 'add the rest tomorrow' — late ads steal learning from the ones that started on time. QC the first two seconds, claims, captions, and file names in the same sitting so Wednesday is launch, not another edit day that quietly eats Thursday's data. A five-cell launch is a broken matrix — wait or cut the grid on purpose.

Work the matrix as a table on screen: rows are hooks, columns are avatars. Generate or film down the columns so the body stays locked. Name files product_hook_avatar so the dashboard maps to the brief without a decoder. Watch each cut on a phone, muted, and confirm the opening line is readable as burned-in text. If a claim would fail policy, fix it now; a rejected ad on Thursday is a missing cell, not a small delay. Rejects are the most common way a seven-day sprint quietly becomes a twelve-day mess.

Klip Kanvas can produce the six variants from one product brief when you lock the body and swap only hook and avatar, which is the entire point of Day 2. If you are mixing AI UGC with real B-roll, cut the B-roll once and reuse it across the grid rather than inventing new footage per cell. Extra difference in the middle of the video is contamination. End the day with six approved files in a launch folder, not with 'almost done except the second avatar.' A five-cell launch is a broken matrix: wait, or cut to three hooks on one face on purpose.

Create a single test campaign, one ad set, broad targeting, conversion objective, equal creative slots for all six ads. Turn Advantage+ (or the TikTok equivalent) off for the test so the platform cannot hide a weak cell inside a black box. Start delivery at midnight in the account timezone so Day 4 is a full day of data, not a partial evening that looks like a winner.

Do not split the six ads across six ad sets. Learning wants pooled conversion events; fragmentation starves every cell. Use automatic placements except for a possible Audience Network exclusion if junk inventory has flattered you before. Same audience, same bid type, same destination. Schedule a calendar block that says do not edit this campaign until the 48-hour mark. Every pause, copy tweak, or 40% budget swing resets learning and burns the sprint's only scarce resource, which is uninterrupted time in market.

Send a launch note to whoever might 'just peek': campaign URL, the six names, the no-touch rule, and the time of the first allowed check. If a client or founder likes to log into Ads Manager, this note is the sprint. Without it, someone will kill variant three at dinner because CPA looked ugly on 200 impressions. Day 3 success is boring: ads delivering, spend roughly even, no one touching bids. Put the note in the same thread as the brief so the readout on Day 6 can quote the rule you actually launched with.

Look only at whether each variant is spending and serving impressions at a similar rate. Do not crown a hook. Do not pause a 'loser.' Early CPA on tiny numbers is noise, and calling a winner here is the most common way a sprint picks the wrong cell. The 24-hour mark is a delivery audit, not a performance meeting — write three lines and close the tab.

The 24-hour audit is technical. Is any ad in review or rejected? Is one avatar getting all the spend because of a setup mistake? Are placements wildly lopsided? Fix delivery bugs; do not 'optimise.' If spend is uneven because the platform already prefers a hook, note it and leave it — that preference is data, not a reason to start moving budgets. Write a three-line status: delivery OK, any rejects, any action taken. Then close the tab and spend the rest of Day 4 preparing the next brief's hook list so Friday is not a scramble.

If a stakeholder asks who is winning, send hook rate and impression counts with a sentence that 24 hours is not a read. Sharing a ROAS leaderboard on Day 4 trains the room to overreact. The sprint's discipline is social as much as statistical. Your job today is to protect the 48–72 hour window you wrote on Monday. If you need to look busy, spend the hour drafting next week's three challenger hooks from reviews, not moving budgets on this grid. Protect the window the way you would protect a live shoot: no visitors on set.

At 48 hours, rank the six ads on hook rate (3-second views ÷ impressions) and hold (15-second views ÷ 3-second views), then CTR. You are allowed to mark likely kills and likely iterates. You are not allowed to double budgets or rebuild the offer. If impressions are still well under ~1,000 on several cells, extend the no-scale rule rather than forcing a story.

Use the funnel order. Hook under ~20% means the first two seconds failed — that cell is a kill or a new open, not a new landing page. Hook 30%+ with hold stuck well under 12% means the open over-promised and the body lost them. Strong hook and hold with weak CTR is a close or offer problem, which you will iterate after the sprint, not during it. Grey-band hooks in the mid-20s stay in market until 72 hours unless spend has already blown the kill rule. Write the ranking in the brief doc so Friday's meeting does not start from screenshots.

Resist the urge to 'help' the trailing avatar by raising its budget. Unequal spend is how you lose the matrix. The only legitimate Day 5 action besides a reject fix is pausing a cell that has already crossed the written kill line with no conversions and a dead hook. Everything else waits for the readout. Use the afternoon to draft two replacement hooks for any cell that is clearly failing the thumbstop, so Day 7 production is ready. Grey-band hooks stay in market; do not negotiate them into a pause because someone liked a different face in preview.

Hold a 25-minute readout. Sort every cell into kill, iterate, or graduate. Kill: hook dead or spend past the rule with no conversions. Iterate: strong open, weak middle or close — keep the first two seconds, rewrite what follows. Graduate: multiple conversions at or under target CPA — duplicate into a scaling campaign, leave the test campaign as the lab.

Read the matrix, not six isolated ads. If one hook wins on both avatars, the message is the driver — next week's sprint adds hooks in that family and can keep the winning face. If one avatar wins across all three hooks, casting is the driver — recast the loser and keep the openings. If nothing is clean after 72 hours and spend was honest, you learned that these three angles are weak; do not 'give it another week' without changing the first two seconds. Ties get 24 more hours, not a gut pick in the meeting.

Write decisions in the brief: paused IDs, the graduate's new campaign, the iterate's new body assignment, and the three hooks for next Monday. Scaling, if you have a graduate, is a 20–30% budget step every 48 hours in the scaling campaign — not a double overnight inside the test. The test campaign stays open for the next grid. Ending the sprint by dismantling the lab is how week two starts from zero. If two cells tie, add 24 hours rather than a gut pick; ties usually break themselves with one more honest day of spend.

Produce or brief the next six cells before you celebrate the winner. Fatigue on cold traffic shows around frequency 2.5–3.5; a graduate without a replacement is a countdown. Day 7 is inventory. If you only scale, week two is a panic refresh instead of a sprint, and the calendar you wanted becomes a one-off project. Queue the next 3 × 2 before you raise a single budget step.

Build next week's trio from the readout, not from a new brainstorm. Winning hook family plus two challengers is the default. Keep the winning avatar if casting won; swap the face if you are testing whether the line generalises. Same offer unless the readout explicitly blamed the offer — and even then, change offer in a separate sprint so you do not mix variables. Put the new brief at the top of the same doc, dated, with the old matrix archived below it. History is how a sprint becomes a playbook instead of seven busy days.

If you have no graduate, still ship a new grid. A week that only produced kills is a successful test if you actually killed them and replaced the openings. The failure mode is running the same six files into week two because production slipped. Protect Day 2 of the next cycle the way you protected the no-touch window. The sprint is the loop, not a one-off project named in a kickoff deck. Book next Friday's readout before you leave, even if this week's table was ugly — especially if it was ugly. Empty labs are how frequency 2.5–3.5 arrives with nothing in render.