---
name: "social_analytics"
description: "Read social media analytics and run a weekly review loop: which metrics actually matter, how to compute engagement rate correctly, and how to turn the numbers into make-more-of-this content decisions."
---

# Social Analytics

## Purpose
Turn a week's worth of native platform analytics into concrete content decisions: which formats and topics to make more of, which to fix, which to kill.

## Workflow

1. **Set the goal first.** The goal picks the outcome metric:
   - Growth → reach per post, follower growth rate, non-follower reach share
   - Community → saves, shares, comments per reach
   - Traffic → link clicks per reach
   - Video → impressions click-through rate + average view duration / retention
2. **Pull the numbers.** Trailing 7 days vs the prior 7, per platform, from native analytics. Keep organic and boosted separate — boosting distorts every rate. Definitions and formulas live in `references/metric-definitions.md`.
3. **Compute per-post rates, never raw counts.** Engagement rate = (likes + comments + shares + saves + clicks) / reach × 100. Also save rate, share rate, and link-click rate — all per reach. Use `bin/score_posts.py` to rank posts and aggregate by format.
4. **Compare like with like.** Reels against Reels, carousels against carousels; organic against organic. Judge a format only with a real sample (~4–6 posts minimum); two posts is an anecdote, not data.
5. **Find the signal.** Top 3 / bottom 3 posts for the week, annotated with what made them different (format, topic, hook, posting slot). Flag any post more than ~2 std dev below its format's median for review — that is an anomaly, not a verdict.
6. **Issue one verdict per format/topic pillar:** MAKE MORE / KEEP / TEST TWEAK / KILL. Each verdict names the evidence: the rate, the sample size, the comparator.
7. **Feed the content plan.** The verdicts go directly into next week's calendar as assignments ("3 Reels in the X format, 0 carousels"). A brief nobody acts on is theater. Full walkthrough with a worked example: `references/weekly-review-loop.md`.

## Output Contract
A weekly analytics brief following `assets/weekly-brief-template.md`: week-over-week numbers, per-format verdicts with sample sizes and rates, top/bottom 3 posts, anomalies, and the concrete next-week assignments. No verdict without a cited rate and a sample size.

## Operating Rules
1. **Always state the denominator.** "Engagement rate 4.2%" means nothing — write "4.2% per reach" or "per impressions". The two formulas give different numbers and cannot be compared.
2. **Raw follower count is a vanity metric.** Report follower *growth rate* and reach *per follower* instead.
3. **Impressions > reach is normal.** Impressions/reach is frequency (repeat views), not a bonus score.
4. **Likes are the weakest engagement.** Weight saves, shares, and comments above likes when ranking what resonated.
5. **No invented benchmarks.** Never quote a "good" engagement rate from memory. If a benchmark is wanted, look up a current attributed industry report for that platform and cite it.
6. **Decisions need 4-week rolling evidence.** One good week is a lead; four weeks of the same signal is a verdict.