Closed-Loop Analytics Upgrade

What it does

A workflow is not a closed loop until it checks whether the change worked and updates the playbook from that evidence. This skill builds exactly that loop. The principle is blunt: human opinion is useful, but platform truth wins — you pull analytics after the change window and let numbers, not "I liked it," decide.

The core pattern has five steps: (1) Input — usage logs, the shipped change, platform analytics, owner feedback, cost/runtime; (2) AI action — compare baseline vs candidate, find repeatable signals; (3) Output — a candidate patch to a prompt, skill, brief template or scoring rubric; (4) Judgment — success rate, speed, cost, quality, human-correction rate and actual performance delta; (5) Self-improvement — promote only if it beats baseline, otherwise keep testing / rollback / unproven.

It knows the right metric set per surface: X (impressions, engagement rate, replies, bookmarks, profile clicks, follower delta), YouTube (CTR, average view duration, retention curve, watch time, subs), SEO/AEO/GEO (GSC clicks/impressions/CTR/position, GA4 sessions/conversions, Ahrefs rankings/backlinks, AI-search visibility), Revenue/Outbound (HubSpot pipeline, Gong call language, Instantly/Smartlead positive replies and booked meetings).

When to use it

• You changed a title/hook/thumbnail/CTA/sequence and need to know if it actually worked.
• You need a decision after a content refresh, schema/internal-linking change, or outbound sequence patch.
• You want evidence before promoting a playbook rule — "the author liked it" is not enough.
• You want to install a periodic readback rhythm with a next-readback date on every change.

Method / frameworks

Closed-loop measurement (baseline-vs-candidate readback), incrementality / iROAS logic (subtract organic demand to isolate the true delta), confounder auditing (seasonality, campaign overlap, dirty attribution, low volume — never conflate correlation with causation), and strict promotion rules: promote if candidate beats baseline on the primary metric OR exposes a repeatable audience signal AND downside metrics are not meaningfully worse. Do not promote on low volume, dirty attribution, seasonality-explained lifts, or author-only enthusiasm. Every promotion writes the mandatory readback fields (change, owner, windows, sources, primary/secondary, winner, caveats, decision, next patch, next readback). Safety boundary: read-only analytics pulls are fine; external writes (posting, CMS publish, ad budget/bid/targeting, CRM/outbound mutation, email/DM) require approval.