mergisi Weekly MRR Digest
A weekly metrics report for solo founders and small SaaS teams who don't have a data analyst yet. Pulls Stripe, Mixpanel (or Amplitude / PostHog), and GA4, computes the core SaaS numbers, runs anomaly detection on each one, and sends a plain-English digest to Slack every Monday at 9am.
What this skill does
Every Monday at 9am (or whenever you schedule it):
- Pull from data sources — Stripe (subscriptions, invoices, refunds), Mixpanel/Amplitude/PostHog (events, cohorts), GA4 (traffic, attribution).
- Compute the core SaaS metrics:
- MRR (current week, +/- vs last week, MoM)
- Net new MRR (new subs minus churned)
- Churn rate (last 30d, last 7d)
- ARPU
- Signup → paid conversion (last 14d, last 90d trailing)
- Activation rate (you define what activation means in the config)
- Top traffic sources by signups, by paid conversion
- Run anomaly detection — z-score against trailing 30-day baseline. Flag any metric 2σ+ off in either direction.
- Add narrative context — "MRR dipped because 3 annual subs churned this week. Two were on the deprecated $19 tier. The third left a feedback note (link)."
- Surface 1-3 action items — based on what's anomalous, suggest the most leveraged thing to investigate this week.
- Send to Slack — to the channel you specify, formatted as Slack mrkdwn with metric blocks and a TL;DR at the top.
What this skill does NOT do
- It does NOT replace a real data analyst — for ad-hoc questions ("why did churn spike in Brazil?"), you still need someone who can write SQL.
- It does NOT build dashboards. For real-time visibility, use Mixpanel/Amplitude's native dashboards or pair with a separate dashboard skill.
- It does NOT predict future MRR. Forecasting is a separate problem; this skill is descriptive analytics, not predictive.
- It does NOT replace Stripe Sigma for deep financial analysis.
How to invoke
Scheduled (recommended): use Hermes' built-in cron.
/cron weekly "every Monday at 9am, run /weekly-mrr-digest and post to #standups"
Manual run:
/weekly-mrr-digest
Or with custom date range:
/weekly-mrr-digest --since 2026-04-01 --until 2026-04-30
Configuration
data_sources:
stripe:
api_key_env: STRIPE_SECRET_KEY
include_test_mode: false
mixpanel:
project_id: "12345"
api_secret_env: MIXPANEL_API_SECRET
ga4:
property_id: "987654321"
service_account_json_env: GA4_SA_JSON
slack:
channel: "#metrics"
bot_token_env: SLACK_BOT_TOKEN
metrics:
activation_event: "first_query_run" # what counts as "activated"
trial_period_days: 7
baseline_window_days: 30
anomaly_z_score: 2.0
schedule:
cron: "0 9 * * 1" # Monday 9am
timezone: "America/New_York"
Models
This is mostly structured queries with narrative summarization on top:
- Recommended: Claude Sonnet 4.6 for the narrative + anomaly explanations. The numbers come from APIs; the model writes the digest text.
- For raw query work: Haiku 4.5 is fine and ~10x cheaper.
- Total weekly cost: ~$0.40-1.20 in API spend for a typical solo SaaS at $1K-50K MRR (mostly the narrative pass).
Sample digest output
📊 Weekly MRR Digest — Week of 2026-04-29
TL;DR: MRR up 1.8% WoW. Anomaly: signup spike Tuesday (+47% vs baseline)
likely from a HN front-page mention. Investigate: activation dropped 14
percentage points after the Apr 24 deploy.
MRR: $3,847 (+1.8% WoW, +12% MoM)
• Net new: +$46 (3 new $216, 2 churned $170)
• Active subs: 247
Conversion (signup → paid, trailing 14d): 4.2%
• Industry SaaS median at your ARR: 2-5%, you're median+
Churn (last 30d monthly): 5.4%
• Industry SaaS 60-75th percentile, top quartile <4%
🚨 Anomaly: Activation rate dropped from 38% → 24% over the past 5 days.
• Likely cause: deploy on Apr 24 changed the onboarding flow.
• Recommend: rollback or fix before more cohort flows through.
Top action item: investigate activation drop before more users hit it.
Pairing
This skill is the analyst piece of a 3-agent data crew. Pair with a dashboard-builder skill (coming soon) and a market-context skill (coming soon) for full coverage.
For the full crew: see crewclaw.com/use-cases/data-analytics-team.
Failure modes to watch
- API rate limits: Mixpanel raw export rate-limits at 60 req/hour. The skill batches sensibly, but for very high event volumes, set
mixpanel.use_jql: true(faster) or run from a VPS in the same region. - Time-zone confusion: all metrics computed in the timezone you configure. If your Stripe data is UTC and your Slack channel expects local time, the digest will note the mismatch but won't auto-convert.
- Cohort definition drift: if you change
activation_eventafter running for 2+ weeks, historical baselines will look wrong. Reset the baseline window after any definition change.