Why data-driven creative only starts after launch

Setting a brand campaign and leaving it is right about the thing that matters most, continuity, and wrong about which layer of the campaign stays constant. Everything that happens before launch, the research, the pretest, the approvals, is prediction, and data-driven creative only starts when the first impression is served and the feed begins returning observed behaviour, per ad, per week. A marketing director at a large consumer brand said it in a first call, we have no tools to understand what is working from a creative perspective, and why. Her account was not short of data, what it lacked was anything connecting an individual ad, live in a feed, to a weekly decision about what to do with it.

Predicting and knowing

Pre-launch research predicts. Post-launch data knows. No panel simulates the feed, a respondent paid to watch is doing something categorically different from a commuter thumbing past between a meme and a message. Pretesting ranks concepts against each other in a room, the feed ranks them against everything, with money on the line, and the spread is wide. Meta's own test data has the winning creative in 235 brand lift tests at 13% lower cost per ad recall than the average of the others in the same test, and across 14,780 A/B tests at 21% lower cost per website conversion. Those gaps exist inside sets of ads that all passed whatever gate preceded them. So the honest status of every launch is hypothesis, and a campaign that runs all year has the most weeks of evidence to act on.

The weekly decision

The signals that arrive first live in the performance tier of How to read paid social results, stop rate, hold rate and watched rate per ad, per week, on one platform and one threshold. They are leading indicators, a watched rate is not brand lift and only the lift study says the brand moved, but they arrive in days rather than quarters. In one measured account each weekly cycle retired the weakest work and extended the strongest, and the watched rate climbed cycle on cycle on the same budget and media plan, and What the numbers show when creative is managed weekly carries the figures.

The weekly decision has three moves. Pause, for ads where attention has decayed or never arrived. Remix, for ads where the idea is alive and the execution is leaving value unclaimed, a strong concept with a slow opening, a winning moment in the wrong format, the cheapest move in the loop and the most undervalued. Replace, for the net-new executions the portfolio now needs, briefed from what the data showed. None of the three touches the strategy. The brand idea, the distinctive assets and the message are what continuity protects, and they stay constant all year, while the executions carrying them rotate, which is Jenni Romaniuk's position, consistent assets, fresh executions, applied at the speed the medium reports.

Fatigue is a diagnosis

Ask how refresh decisions get made and the answers fall into two camps, instinct, the team is bored of the work, and calendar. Neither consults the only party whose boredom matters, the audience. Fatigue is a per-ad condition with a visible signature, watched rate slides for the audience the ad has been reaching, cost per watched impression climbs as the auction reprices the weaker response, frequency creeps as the system struggles to find fresh responders, three readings moving together. TikTok's definition is the plain one, an audience that has seen the same ad too many times and become less likely to engage, and both platforms now ship detection for it. Audiences tire of the same fifteen seconds, not of the brand or the idea, so the first move is remix, not new strategy. The reverse reading matters as much. An ad holding its watched rate at stable cost is not fatigued whatever the calendar says, and retiring it is retiring an asset that is still working. An execution is retired because it was not watched, never because it is old.

Pretesting as sufficient

The counter-position is the process most large brands run, and it has real evidence behind it. System1's star rating, tested on 101 Facebook ads and 50 Pinterest ads against platform lift studies, found high-star ads had a materially higher success rate on brand awareness gain than low-star ads. A strong forced-exposure score does predict how an ad performs once it has attention, and every ad in a well-run system passes a floor before media spend attaches. The word doing the work is attention. Forced exposure means the respondent was made to watch, so the score says nothing about whether the ad earns attention in a feed, and What pretesting can tell you, and what only the feed can takes the method apart properly. Pretesting is the gate. The feed is the exam, marked weekly, per ad. Media creates the opportunity to see your ads. Creative determines whether they work. Only the post-launch data ever finds out which creative that is.

What to do with this

Keep the campaign running all year, and move the decision from the calendar to the data. Export the last quarter from Ads Manager, one platform, one threshold, and compute stop rate, hold rate and watched rate per ad per week. Mark the ads whose watched rate is sliding while cost per watched impression and frequency climb, the fatigue candidates, and the ones holding steady, which stay live whatever the plan said. Then put a thirty-minute weekly review in the diary with one owner and three columns, pause, remix, replace, and record each decision against the reason for it. How to read paid social results is the standard the review runs on, and it is the natural next read.

Questions

It is a brand campaign. Why would we change it every week?

You do not change the campaign, you change which executions carry it. Continuity is the right instinct for brand building, Byron Sharp's work on reach and continuity over bursts holds on social as it does anywhere, and the idea, the distinctive assets and the message stay constant all year. What the weekly review decides is which of the twenty or thirty executions carrying that idea are still being watched, which need a new opening or format, and which gaps in the portfolio need a new execution.

Left alone, the same fifteen seconds reaches the same people more often while fewer of them watch, and the media buys repeat skips of a message those people already have.

How do we know an ad is fatigued rather than having a bad week?

Three readings move together. Watched rate slides for the audience the ad has been reaching, cost per watched impression climbs as the auction reprices the weaker response, and frequency creeps as the system struggles to find fresh people. One reading on its own can be noise, a holiday week, a competitor's burst. All three for two consecutive weeks is the diagnosis, and the first response is a remix of the same footage with a new opening, not a new strategy. An ad whose watched rate is holding at stable cost is not fatigued, whatever the calendar says.

Is ninety days too early to judge a brand campaign?

Too early to judge effectiveness, which lands over quarters and is read in lift studies on their own clock, and the right window to judge the portfolio. Month one casts the net and the reading is coverage and signal quality. Month two starts the cull and is usually when stop and watched rates move visibly. Month three is when the surviving portfolio is concentrated in what earned attention and the next briefs are written from evidence. Decide before launch which metrics are decision-grade inside the quarter and which are out of bounds, so neither a mixed first month nor a flat lift read gets mistaken for a verdict.

Sources

• Meta, "Creative and Media Excellence", circa 2023. Winning creative against the average of other creatives in the same test, 235 brand lift tests (April 2021 to 2022) and 14,780 A/B tests (April 2021 to April 2022). Meta internal test data, Meta sells the media.
• TikTok for Business creative guidance and TikTok Ads Manager documentation. Definition of creative fatigue and the platform's fatigue detection and auto-refresh tooling. Platform documentation, TikTok sells the media.
• System1 test of 101 Facebook ads and 50 Pinterest ads against platform brand lift studies, as cited by Tom Roach, "The 3rd age of effectiveness", circa 2023. High-star ads and brand awareness success rate. System1 sells the test.
• System1, Kantar Link and Swayable methodology pages. Forced-exposure pretesting predicts response under attention, not attention capture in a feed. Each sells the test.
• Jenni Romaniuk, Building Distinctive Brand Assets, 2018, and Byron Sharp, How Brands Grow, 2010. Consistency of assets with variety of executions, continuity over bursts. Ehrenberg-Bass Institute, sponsored by large advertisers.
• One measured account, anonymised. Weekly cycles retiring the weakest work and extending the strongest on the same budget and media plan, from the account's own reporting. Figures carried in What the numbers show when creative is managed weekly. Our own account, interest declared.
• A marketing director at a large consumer brand, anonymised, first call, 2025.