# Creative Fatigue Detector > A Claude skill from hayesti54-eng/ai-media-buying-skills. - **Type:** Skill - **Install:** `agentstack add skill-hayesti54-eng-ai-media-buying-skills-creative-fatigue-detector` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [hayesti54-eng](https://agentstack.voostack.com/s/hayesti54-eng) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [hayesti54-eng](https://github.com/hayesti54-eng) - **Source:** https://github.com/hayesti54-eng/ai-media-buying-skills/tree/main/tiktok-ads/diagnostics/creative-fatigue-detector ## Install ```sh agentstack add skill-hayesti54-eng-ai-media-buying-skills-creative-fatigue-detector ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # TikTok Creative Fatigue Detector **Platform:** TikTok Ads **Category:** diagnostics **Tier:** pro ## Purpose Identify which TikTok ad creatives are entering, mid-stage, or fully in creative fatigue — defined as the degradation of performance metrics over time due to audience overexposure — before they become net-negative budget drains. This skill uses trend analysis across hook rate, CTR, CPM, frequency, and CPA to generate an early-warning fatigue signal and a replacement readiness score for each creative. ## When to Use - A creative that was performing well has shown declining metrics over 5–10 consecutive days - CPMs are rising on a specific creative or ad set without budget increases or audience changes - CTR is declining while impressions remain flat or grow — a textbook fatigue signature on TikTok - You are managing a high-volume creative pipeline and need to systematically forecast replacement timing - Post-launch (day 7–14) review of all new creatives to identify which need iteration before full fatigue sets in ## Inputs Required - 21–30 days of TikTok Ads Manager data broken down by day, at the creative level: Impressions, Reach, Frequency (or calculated from Impressions/Reach), 2-Second Video Views, CTR, CPM, CPC, Spend, Conversions, CPA - Creative launch date for each ad - Ad set frequency cap settings (if any) - Audience size estimate per ad set (from TikTok audience estimates) - Creative format labels: UGC, Spark, non-Spark, branded, influencer-style, product demo ## What This Skill Does This skill performs time-series trend analysis on each creative's core metrics, calculates week-over-week and 3-day rolling average changes, detects fatigue signatures (rising CPM + falling CTR + rising CPA on the same creative), and assigns a Fatigue Score (0–100) and a Replacement Urgency classification (Green / Yellow / Red) to each creative. It also projects a "kill date" based on current degradation trajectory and provides a replacement timing recommendation. ## Analysis Workflow 1. Ingest 21–30 days of daily creative-level data. Organize each creative into a time-series table. Calculate days since launch for each creative. 2. Calculate 7-day rolling averages for: CTR, CPM, Hook Rate (2s view rate), CPA. Compute week-over-week delta for each metric across all 3 rolling windows. 3. Flag fatigue signatures using a 3-signal detection model: - Signal 1: CTR decline of 15%+ week-over-week - Signal 2: CPM increase of 20%+ week-over-week (algorithm is paying more to find receptive viewers) - Signal 3: CPA degradation of 25%+ week-over-week A creative triggering 1 signal = Yellow. 2 signals = Orange. 3 signals = Red (active fatigue, immediate action required). 4. Calculate estimated audience frequency. If frequency reaches 2.5+ within the target audience size, flag regardless of metric status — overexposure is imminent. 5. Cross-reference fatigue signal with creative age. A 30-day-old creative showing Yellow signals is in normal decay. A 7-day-old creative showing Yellow signals indicates the audience was already saturated or the creative was never differentiated. 6. For each Red-signal creative, calculate the spend waste rate: daily spend × number of days CPA has been above target. This is the cost of delayed replacement. 7. Assign Fatigue Score (0–100): weight CTR decline (35%), CPM increase (30%), CPA degradation (25%), frequency level (10%). 8. Classify: Score 0–30 = Green (healthy), 31–55 = Yellow (monitor closely), 56–75 = Orange (begin replacement prep), 76–100 = Red (pause or replace within 48 hours). 9. Generate replacement timeline recommendation: when to launch replacement creative and what variant types to test based on the fatiguing creative's best-performing elements. ## Output Requirements - Creative-level fatigue scorecard: Fatigue Score, Urgency Classification, Days Live, estimated Frequency, week-over-week CTR/CPM/CPA delta - Red and Orange creative list with estimated daily spend waste and recommended kill/replace date - Replacement readiness brief per Red creative: preserve these elements (hook mechanic, format, offer), change these elements (visual style, hook opening, pacing) - Projected creative runway for Green/Yellow creatives based on current trajectory - Pipeline gap analysis: if all Red and Orange creatives are killed, how many replacement creatives are needed to maintain current spend levels? ## Platform-Specific Best Practices - TikTok audiences fatigue faster than Meta due to the For You Page's high content velocity. A strong creative may peak in days 3–7 and enter decay by day 14 on a narrowly targeted ad set. - Spark Ads tend to fatigue more slowly than non-Spark because organic engagement activity (likes, shares, comments) continues to accumulate, creating social proof that sustains CTR longer. - Creative velocity — the rate at which new creatives are introduced to the account — is the primary defense against fatigue-driven performance collapse. High-performing TikTok accounts maintain a 3–5 new creative introductions per week cadence. - Rotating creative angles within the same campaign can extend the effective life of a concept. Fresh hooks on proven offer structures reduce full replacement costs. - CPM spikes are TikTok's most reliable early fatigue indicator. The algorithm must bid higher to find users who haven't been overexposed — this is the platform's signal that the creative is losing relevance. - Broad audiences (Broad Targeting in TikTok) experience slower frequency build-up and therefore extend creative runway significantly compared to narrow interest stacks. ## Guardrails - Do not pause a high-CPA creative solely on frequency grounds if it is still converting below CPA target — frequency tolerance varies by audience size and campaign type. - Do not interpret a single bad day as fatigue onset. Require at least 3 consecutive days of multi-signal decline before escalating to Orange or Red. - Do not apply the same fatigue benchmarks to a $500/day account and a $50/day account — frequency builds at dramatically different rates. - Never replace a fatiguing creative without preserving and testing its highest-performing structural elements in the iteration. ## Example Requests 1. "Here's 30 days of daily creative-level data across 8 ads. Tell me which ones are fatigued, which ones are about to fatigue, and what I need to do about each one right now." 2. "My best performing creative is 22 days old and CPA has climbed 40% over the last week. Is this fatigue or something else? Walk me through a full fatigue diagnosis." 3. "I have a $1,500/day TikTok account with 5 active creatives. Build me a fatigue risk report and tell me how many new creatives I need in the pipeline this week." ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [hayesti54-eng](https://github.com/hayesti54-eng) - **Source:** [hayesti54-eng/ai-media-buying-skills](https://github.com/hayesti54-eng/ai-media-buying-skills) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-hayesti54-eng-ai-media-buying-skills-creative-fatigue-detector - Seller: https://agentstack.voostack.com/s/hayesti54-eng - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.