ع
Start Topics Teams Reference What's new Saved
playbook

The monthly performance review loop

Turn a raw export of post or campaign metrics into a clear read on what's working, why, and what to do more of — a repeatable monthly ritual, not a one-off.

medium ~40 min
when to reach for this

Most teams have dashboards full of numbers and no time to find the story in them, so they keep doing what feels good instead of what works. This loop turns a metrics export into an honest monthly read: the real winners, the pattern behind them, and a short list of bets for next month. Run it the same way every month and the comparisons start compounding into actual learning.

gather this first
  • A metrics export as CSV — posts-last-90.csv with columns like date, channel, post, impressions, engagement, clicks, conversions.
  • Last month's review (if you've run this before) so Claude can compare and track whether bets paid off.
  • Context Claude can't see: any post that was boosted, went viral by luck, or rode a news moment — flag these so they don't skew the read.
the workflow

  1. Get the lay of the land

    Open the folder with your metrics export in Claude Desktop and ask in the chat — no terminal needed. Start with an honest summary before any analysis, so you can sanity-check that Claude read the file correctly and you spot any junk data early.

    you ask
    Read posts-last-90.csv. Give me a plain-English summary: how many posts, across which channels, and the median + top/bottom for each metric. Flag anything that looks like a data error or an outlier.

    what you get back A grounded overview with medians (not just averages, which outliers distort) and a flag on anything suspect — your data sanity check.

    Always anchor on the median. One viral post can drag the average up and make a bad month look great.

  2. Find the pattern, not just the top row

    The value isn't "this post won" — it's why. Ask Claude to find what the winners share, which is the thing you can actually repeat.

    you ask
    Identify the top 5 posts by engagement rate (not raw impressions). What do they have in common — format, topic, length, hook, time of day? Then do the same for the bottom 5. Give me the contrast as a short table.

    what you get back A winners-vs-losers contrast — e.g. "top posts: how-to, under 120 words, open with a number; bottom posts: announcements, long, open with 'We're excited'" — the repeatable signal.

  3. Turn the read into bets

    Convert the pattern into a few specific, testable actions for next month. Vague takeaways die in the doc; bets get tested.

    you ask
    Based on this, give me 3 specific things to do MORE of and 2 to stop, each phrased as a testable bet for next month with a metric to watch. Be concrete enough that I could tell in 30 days whether it worked.

    what you get back Five concrete bets, each tied to a metric — "ship 2 how-tos a week, target +20% engagement" — not platitudes.

  4. Close last month's loop

    If you ran this before, this is the step that makes it a system: did last month's bets pay off? That accountability is where the compounding happens.

    you ask
    Here's last month's review. Did the bets I made then actually move the numbers this month? Mark each as worked / didn't / unclear, and tell me what to keep testing.

    what you get back A scorecard on last month's bets so winners get doubled down and dead ends get dropped — learning, not just reporting.

make it your own
  • Channel deep-dive: narrow any step to one channel when a platform is behaving strangely and you need to know why.
  • Campaign view: swap posts for campaigns and add spend to get a rough cost-per-result read alongside engagement.
  • Automate the ritual (Power Track): save the prompt chain and run it on the 1st of each month; advanced teams can wire it to a scheduled agent (see the Capabilities tab).
watch out for
  • Correlation isn't cause. "Tuesday posts do best" might just be what you happened to post on Tuesdays. Treat findings as hypotheses to test, not laws.
  • Feed Claude context it can't see — a boosted post or a lucky viral hit will otherwise be read as a repeatable win.
  • Mind what's in the export. Reach and engagement are safe to analyze; a CSV that carries individual customer names, emails, or lead-level data is not. Strip it to aggregates first, or keep the analysis inside a workspace your company has approved for that data.
  • Don't let the metric become the goal. Optimizing engagement can quietly trade away the conversions that actually matter — keep an eye on the downstream number.

you'll end up with A repeatable monthly ritual that turns raw exports into an honest read on what's working and a short list of tested bets — so next month's content is informed by this month's, not by vibes.

Questions people ask

What format should my metrics be in?
A CSV export works best — something like `posts-last-90.csv` with columns for date, channel, post, impressions, engagement, clicks, and conversions. If you've run this before, bring last month's review too so Claude can score whether your bets paid off.
Why anchor on the median instead of the average?
Because one viral post can drag the average up and make a bad month look great. The median tells you what a typical post actually did, so you're optimizing for repeatable results rather than for a lucky outlier.
Should I trust a finding like \"Tuesday posts do best\"?
Treat it as a hypothesis, not a law. Correlation isn't cause — "Tuesday does best" might just be what you happened to post on Tuesdays. Phrase each finding as a testable bet and confirm it next month.
What context does Claude need that the export doesn't show?
Flag anything the numbers can't explain: a boosted post, a lucky viral hit, or a post that rode a news moment. Without that, Claude will read those spikes as repeatable wins and steer your bets wrong.