Build a TikTok Creative-Fatigue Triage Board

01

1. Build one row for every ad under review

The board should preserve enough media evidence to support a decision and enough creative context to produce the next asset.

Create one row for each ad that needs review. Start with identifiers: review date and account time zone, campaign, ad group, ad name or ID, objective, audience, landing page, offer, creator, format, concept, angle, hook, and current lifecycle state. Lifecycle matters because a stable ad, a recently edited ad, and an ad still learning should not be interpreted in the same way.

Add two explicit comparison windows: a current window and a baseline window. Make them comparable in duration and commercial context. A promotion weekend, inventory shortage, site outage, or abrupt spending change should not quietly become the baseline for normal performance. The board should also show conversion volume; a percentage movement based on very few events may not support a confident creative decision.

TikTok defines reach as an estimate of the unique accounts exposed to an ad and frequency as the average number of exposures per unique account. TikTok also cautions that most reach figures are estimates, so use them directionally rather than treating them as a precise census (TikTok’s Reach metric guide).

Finish the row with operational fields: suspected cause, confounders checked, decision, decision statement, linked backlog item, owner, due date, next review condition, launch date, and result. This prevents a dashboard observation from dying in a meeting without becoming a controlled action.

1. Identify the campaign, ad group, ad, audience, objective, offer, and landing page.
2. Describe the creative as a concept, angle, hook, creator, format, scenes, caption treatment, proof mechanism, and call to action.
3. Record the current lifecycle: learning, stable, recently edited, scaling, constrained, or paused.
4. Choose current and baseline periods that are comparable.
5. Assign an owner and specify when or under what evidence condition the ad will be reviewed again.

02

2. Record evidence without inventing universal fatigue thresholds

Use your brand’s prior stable period and acceptable economics as the reference point. Do not copy a universal frequency, CTR, or CPA cutoff.

The evidence portion of the board should contain four groups. Delivery includes spend, impressions, delivery status, and budget utilization. Audience includes reach or daily new-user delivery and frequency. Attention includes CTR and the video-view or retention indicators your team routinely uses. Outcome includes conversion rate, CPA, ROAS where relevant, conversion volume, and revenue tracked in the platform.

Read these as a pattern. For example, rising frequency combined with weaker reach, CTR, and acquisition efficiency is more persuasive than a CPA increase by itself. TikTok’s web-auction guidance describes fatigue using several signals together, including declining CTR, rising frequency, declining reach or impressions, and worsening CPA or ROAS. The same guidance notes that creative is not always the cause; targeting, bidding, and budget can contribute (TikTok Web Auction Best Practices).

Avoid rules such as “refresh at frequency X” or “kill after Y days.” A defensible boundary depends on the objective, audience size, spend, conversion volume, attribution approach, promotion calendar, product economics, and the ad’s historical behavior. Instead, write a brand-specific evidence condition: “Review when the decline persists across two comparable reporting windows and at least one audience or attention indicator corroborates the outcome change.” That is a process rule, not a universal performance benchmark.

1. Has the decline persisted across comparable periods rather than one isolated day?
2. Is delivery falling, or is the campaign simply unable to spend?
3. Is new-user reach weakening while frequency increases?
4. Is CTR falling? If CPA rose, did CTR, conversion rate, or both contribute?
5. Is there enough conversion volume to interpret the movement responsibly?
6. Did TikTok surface a fatigue, delivery, approval, or Diagnosis alert?

03

3. Rule out non-creative causes before opening production work

A creative-fatigue workflow must contain a confounder check. Otherwise, every account problem becomes an expensive request for more videos.

Record recent changes to budget, bids, bidding strategy, targeting, attribution, schedule, creative mix, offer, price, inventory, landing page, checkout, and tracking. Also note account balance, approval, or delivery restrictions. The operator does not need to prove that every item caused the decline; the goal is to expose plausible alternative explanations before the team commits production capacity.

TikTok states that performance can fluctuate during the learning phase and that significant changes to budget, bids, bidding strategy, or targeting may retrigger learning (TikTok learning-phase guidance). A short-term decline following one of those changes is therefore weak evidence of fatigue on its own.

Use a simple confounder status for each row: ruled out, plausible, confirmed, or unknown. If a site problem is confirmed, repair the site and hold the creative decision. If targeting changed and the campaign is learning again, schedule another review. If the account is stable and several creative-related indicators continue to deteriorate, proceed to triage.

1. Check budget, bid, bidding strategy, audience, schedule, attribution, approval, and delivery changes.
2. Check whether the ad group is learning or relearning.
3. Check the offer, price, stock, landing page, checkout, and measurement setup.
4. Label each possible confounder as ruled out, plausible, confirmed, or unknown.
5. Do not create a creative task until the reviewer can explain why creative is a credible cause.

04

4. Choose Hold when the evidence is incomplete

Hold means making no immediate creative change while preserving the ad as a control and scheduling a deliberate recheck.

Choose hold when the decline is brief, isolated, low-volume, or occurring during learning. Hold is also appropriate when a recent budget, bid, targeting, schedule, approval, tracking, offer, or landing-page change offers a stronger explanation. If CPA moves but CTR, reach, frequency, and conversion rate do not corroborate creative decay, the board should not manufacture certainty.

Holding is an active decision, not neglect. Write down the suspected confounder, the evidence needed to resolve it, and the next review date or condition. Preserve the current ad as the control against which later variants can be compared.

Do not pull an item into production yet. Instead, mark promising hooks, captions, scenes, and concepts as ready if the decline persists. This keeps the team prepared without spending production time on an unconfirmed diagnosis.

1. Set the decision to Hold.
2. Document why the current evidence is insufficient.
3. Name the possible confounder and the next evidence required.
4. Preserve the ad as a control.
5. Set a review date or minimum evidence condition.

05

5. Choose Iterate when the concept is still viable

Iterate when you can name a specific weak component and preserve the underlying premise, audience, or offer.

Iteration fits cases where attention is weakening but downstream response remains comparatively healthy, or where frame-level review suggests that an opening, transition, explanation, proof moment, caption, or call to action is the likely weak point. The operative test is whether the team can state a component-level hypothesis.

Pull from one backlog lane. A hook iteration can change the first spoken line, opening visual, objection, question, claim framing, or pattern interrupt. A caption iteration can change context, benefit framing, proof, offer language, or call to action. A scene iteration can replace the opening shot, move the demonstration earlier, add a proof moment, or tighten the sequence. A recombination can retain the concept while changing one or two clearly identified elements.

TikTok’s web-auction guidance identifies new creative iterations and updated captions as possible responses to fatigue, while emphasizing elements such as the hook, proposition, overlays, transitions, graphics, and call to action (TikTok Web Auction Best Practices).

The brief must say what remains constant and what changes. “Make three fresh versions” is not a testable brief. “Keep the creator, audience, product demonstration, offer, and landing page constant; replace the opening problem statement with three objection-led hooks” is much more useful. After launch, classify the result against that hypothesis.

1. Name the component believed to be weakening.
2. Select one hook, caption, scene, or controlled recombination from the backlog.
3. List every element that must remain constant.
4. Write the expected effect, such as improved early attention without changing the offer.
5. Link the produced variants back to the parent creative and triage decision.

06

6. Choose Refresh when the premise needs replacement

Refresh means introducing a materially different full concept, not disguising the same ad with a new first line.

Choose refresh when delivery and new-user reach deteriorate as frequency rises, attention and downstream efficiency weaken together, several iterations of the same premise have declined, or the team cannot form another credible component-level hypothesis. A platform fatigue notice can support the decision, but it should be recorded alongside the observed trend rather than used as the entire rationale.

Pull from the full-concept backlog. Change the premise, narrative structure, visual environment, spokesperson, use case, proof mechanism, or core desire. The replacement should be different enough to explore a new creative route while still serving the intended audience and commercial objective.

TikTok recommends refreshing creative when delivery shows a consistent declining trend or daily new-user delivery is low. Its guidance also recommends adding new creative to the existing ad group rather than automatically creating another ad group (TikTok creative best practices). Apply that platform guidance in light of your own campaign structure and constraints.

Do not delete the intellectual history of the old ad. Preserve its winning elements, decline pattern, iterations attempted, and failure hypothesis. A fatigued creative can still contain reusable hooks, proof, scenes, or audience insights.

1. Set the decision to Refresh.
2. Write why another component-level iteration is not credible.
3. Select a materially different full concept.
4. Define the new premise, audience desire, proof mechanism, narrative, creator type, and offer treatment.
5. Archive the old creative’s context and reusable learnings.

07

7. Maintain four linked creative backlogs

Separate component ideas from complete concepts so the triage decision naturally determines the size of the production response.

Use four queues: hooks, captions, scenes, and full concepts. Hook records contain spoken openings, visual interrupts, objections, desires, questions, or proof-led entries. Caption records contain overlays, subtitles, context, benefit framing, proof, offer treatment, and calls to action. Scene records contain demonstrations, reactions, close-ups, comparisons, transitions, and proof moments. Full-concept records contain the complete premise, audience, angle, narrative, creator type, proof mechanism, and offer.

Every item should carry its parent creative family, intended audience and desire, variable being changed, supporting evidence or previous learning, estimated production effort, readiness, linked triage decision, and post-launch result. This structure allows a hold decision to reserve ideas, an iterate decision to pull a component, and a refresh decision to pull a full concept.

Use this decision statement in every approved row: “We chose [hold / iterate / refresh] because [observed pattern]. We ruled out [confounders]. The next creative action is [specific backlog item], intended to test [hypothesis]. We will reassess on [date or evidence condition].”

1. Create separate queues for hooks, captions, scenes, and full concepts.
2. Link every component to a parent concept or creative family.
3. Record the evidence or learning behind each idea.
4. Attach the selected item to the triage row and production brief.
5. Return the result to both the ad record and the backlog item.

08

8. Keep human review even when TikTok automates replacement

Automation can change delivery, but your team still needs a record of what was varied and what was learned.

TikTok describes Smart Creative as using fatigue detection and auto-refresh logic to pause videos showing fatigue signals and introduce waiting-list assets. TikTok also notes that some advertisers may see Automate Creative instead (TikTok Smart Creative).

That does not eliminate the need for a triage board. Record the original concept, which elements the platform varied or replaced, the platform warning or action, the team’s interpretation, and the learning that should influence the next brief. Otherwise, automation can increase asset turnover without improving your organization’s creative memory.

Your review cadence can be daily for urgent delivery issues and weekly for broader creative planning, but the board should distinguish observation from decision. A recurring review should end with a documented hold, iterate, or refresh status—not a vague request for “more creative.”

1. Record whether automated creative features are active.
2. Capture any platform fatigue or replacement action.
3. Identify the original and replacement creative families.
4. Document what changed and what remained constant.
5. Feed the outcome into the next manual brief.

09

9. Run the workflow in ATIYO without pretending it reads TikTok performance

Media performance remains in the ad platform, and ATIYO preserves creative context and learnings.

ATIYO does not connect to ad accounts, diagnose fatigue from live TikTok data, buy media, calculate ROAS, or know your performance unless someone records it. Operators should review TikTok Ads Manager, manually bring the relevant indicators into the triage record, and preserve the source period and review date.

ATIYO can then organize the creative side of the process: the ad’s concept and brand context, the hold/iterate/refresh decision, hypotheses, linked hooks and angles, production briefs, assets, iterations, owners, roadmap work, and reusable post-test learning. This keeps a media observation connected to the work it creates without implying automatic performance analysis.

A practical setup is to use a roadmap status for triage, connect the approved decision to a brief, attach the selected backlog item, route the work to an owner, and record the result after launch. The Brain is BYOK through OpenRouter, but any generated assistance should be grounded in the context your team supplies; it does not independently know TikTok performance.

If you do not use ATIYO, the same workflow can run in a spreadsheet, project-management tool, or database. The essential requirements are stable fields, explicit decision criteria, links between parent concepts and variants, named owners, and a closed loop from evidence to production to learning.

1. Review and export or transcribe the relevant indicators from TikTok.
2. Record the comparison windows and decision evidence manually.
3. Connect the decision to the parent creative family and selected backlog item.
4. Create a brief that specifies constants, variables, owner, and due date.
5. After launch, bring the result back into the creative record as a reusable learning.