Creative Space
Turn Ad Performance into the Next Creative Brief

Contenu

Provide: Authorized read access to ad accounts or backend-exported reports, plus the campaign goal, such as testing hooks or driving conversions.

Receive: A review grouped by creative approach: which angles/hooks/formats work, which may be fatigued, and evidence for every finding, separating facts and hypotheses. Include replacement candidates and the next production brief. Optionally hand it to a production inspiration or schedule a weekly review.

TikTok Ad Creative Studio begins with production and ends with feedback. This workflow begins with data and suits already-running accounts with mixed creative sources, including several platforms.

Tools and responsibilities

  • Ad-account connection: Read creatives and reports using the route below.
  • Host data analysis: Read/join/deduplicate reports, compare statistics, chart, and validate as available. No fixed analysis-skill name is required.
  • creative review: Creative-review business knowledge, asset ledger, relationships between performance and creative choices, and next-round briefs. This inspiration handles existing-account intake, review organization, and production routing.
  • Teardown — retrieve and read the full “Break Down a Reference Video into an Interactive Report” inspiration through Flova MCP by name: Inspect videos/copy to tag previously untracked creatives.
  • Production inspiration, optional: Make replacement creatives from the approved brief.
  • Review records: Local files or the user's Notion, Feishu Base, or Airtable.

Connect ad accounts

Check routes in this order; use the first suitable route rather than skipping ahead:

  1. Official platform MCP/plugin. Verify actual tools, permissions, report granularity, and returned metrics individually. A feature in the web dashboard does not prove MCP or the host exposes that data. If connection fails, continue using exported files to finish the review.
  2. Ad-data aggregation service, such as Supermetrics or Windsor.ai. These third parties hold authorization and charge their own fees. Explain that before connecting.
  3. Dashboard export. Request ad-level daily reports with date, account/ad-group ID, ad name/ID, creative preview/ID, spend, impressions, clicks, and conversions for the chosen goal. ROAS requires conversion value. Add frequency, reach, early video views/completion where supported. Missing fields narrow the questions answerable; do not require identical metrics across platforms.

Operational considerations:

  • Choose read-only permissions where available. Let the user reauthorize expired access; do not authorize on their behalf or store tokens.
  • Ask which accounts to inspect and review accounts separately.
  • Record each account's time zone, currency, and attribution window; do not directly sum incompatible account/platform results.
  • Delivery systems allocate budget to creatives predicted to perform best overall. Low-spend creatives may show lower average costs because of allocation (Meta's breakdown effect: https://www.facebook.com/business/help/770303663944673). This alone does not justify recommending that expensive creatives be stopped.

Boundaries

  • Read only: Do not change budgets/bids, start/stop ads, modify audiences, delete creatives, or create tests; do not call such connector write tools. Requested replacements are produced in the corresponding inspiration and launched by the user.
  • Creative review only: Performance can also change because of budget, bids, audience, seasonality, or landing pages. If signals suggest these causes—for example, every creative declines together, or conversions collapse while clicks remain healthy—state that the cause may not be creative and needs user verification. Do not attribute it automatically to creative.
  • No performance guarantee: Separate facts, interpretations, and evidence. Report association, not promised improvement from replacement.

Workflow

Confirm accounts/goals → add creative tags → review → fatigue/replacement candidates → next brief → optional production handoff → scheduling. Account/goals, first-time tags, and next brief are confirmation points.

1. Confirm accounts and goals

Ask platform/accounts, objective determining primary metrics (see creative review), product line/campaign, and dates. Default to completed dates over the past two weeks; treat recent conversions separately according to reporting maturity. Write config.md and read it on subsequent runs. Check actual host analysis/data-access capabilities first, then reuse the creative-review method's business method and Teardown analysis. A missing fixed Skill name does not prove the underlying capability is absent. Explain the concrete impact of any gap.

2. Add tags — first-time confirmation

Most existing assets lack a ledger. Add creative review ledger tags individually and mark them as post-launch annotations:

  • Use Teardown to inspect the opening seconds, overall format/presentation, and ad copy. Tag content before looking at performance, following creative review. If video is unavailable, retain readable names/copy only and mark visual tags “content not verified.”
  • With many creatives, start with the highest-spend batch, e.g. 30, and list the rest as untagged.
  • Agree the vocabulary (angles, hook types, formats) and show vocabulary/ledger for confirmation. Name future assets consistently to reduce retrospective tagging.

3. Review

Use creative review tags/business questions with host analysis tools for processing, comparisons, and validation. Check one creative mapped to several ads, dynamic combinations, attribution delay, and insufficient samples. For several platforms, write separate findings before summarizing similar approaches, retaining each comparison scope. Identify concrete next hypotheses from actual creatives; do not declare a universal cross-platform winner.

4. Fatigue and replacement candidates

Following creative review, compare trends across equal-length windows and consider alternative explanations. For each suspected fatigue/inefficiency candidate, record asset, evidence, ongoing spend, reason for candidacy, and proposed next test. Without trends or key definitions, label it pending verification instead of concluding.

5. Next brief — confirmation

Use the creative-review method's keep/replace/new-test format and ask the user to confirm. New tests change one variable at a time.

6. Production handoff — optional

After confirmation, route by creative type, attaching brief, ledger, and reference assets:

  • Presenter/spoken or real-life-style creative → UGC Ads.
  • Product showcase/scene advertisement → Product Video Studio.
  • Batch TikTok angle × hook tests → TikTok Ad Creative Studio.
  • Still-image creative → E-commerce Image Studio.

Create a new ledger batch, use naming standards, and leave launching to the user.

7. Schedule

  • Complete one manual review first. After approval of its format, default to weekly.
  • Scheduled runs perform steps 3–5 only: read configuration/ledger, review, draft candidate list/brief, notify the user. They neither generate creatives nor change ad delivery.
  • List newly launched, untagged creatives as pending tags and ask the user to confirm additions.
  • Use actual host scheduling capabilities. With exported reports, scheduling can remind the user to export; it cannot fetch data automatically.

Project directory

ad-review-<brand>/
├── config.md       Accounts, objectives, primary metrics, vocabulary, schedule
├── docs/           creative-tracker.csv, debrief-<date>.md, results-<date>.csv, briefs/
└── source/         Exported reports and creative-preview frames

Creative tracker and review

The host creates or reuses the tracker; a separate tracker-specification file is not required. Use UTF-8 docs/creative-tracker.csv or the user’s existing table. Record creative ID, batch, actual file, production sources, parent/version, previous review, product/topic, angle, hook type and first line/frame, format/presenter, duration/ratio, hypothesis/control, platform/placement, real post/ad/material IDs and links, publication time/state, and latest conclusion. Fill labels during production and real IDs after upload/publication. Preserve multi-platform/ad mappings and create new version rows. Agree batch vocabulary; when useful, align file/material/suggested ad names as <product>_<angle>_<hook>_<format>_<ratio>_<batch> (for example serum_dry-skin_pain_ugc_916_R2), retaining an established valid convention.

Use an authorized connection or a supplied/chosen report and join results to the tracker by actual IDs. Tag historical content before looking at performance; do not invent unseen visual labels. The host performs calculations/plots within stated objective, complete dates, audience/placement, attribution window/maturity, currency/time zone, spend and sample limits. Preserve many-ad mappings and dynamic combinations. Low exposure is not a proven loss; suspected fatigue needs a trend and checks for delivery-condition changes. Separate observations, explanatory hypotheses and evidence; insufficient data yields a next test, not a fabricated winner. Hand production a brief specifying what to keep, the exact part to replace, the next controlled change/metric and destination workflow, with actual material, parent and batch links. Analysis alone does not trigger generation, publication or campaign changes.

Processus créatif

  1. TikTok

    TikTok

    Read ad performance data

    Retrieve advertising reports and the corresponding video assets.

  2. Notion

    Notion

    Save the review

    Record creative performance and recommendations for the next iteration.