Studio New Account Churn Risk

Studio New Account Churn Risk Report

Generates a live, interactive churn risk dashboard scoped to members who joined in the last 90 days, for any MarianaTek studio using the connected MCP server.

Step 1: Identify the MCP server

Check which MarianaTek studio MCP is connected and available. An example includes:

• Cousteau / Cousteau Admin

Use whichever studio MCP the user has connected and is asking about. If multiple are connected, ask the user which studio they want the report for.

Step 2: Pull live member data

Paginate through ALL members — studios can have thousands of accounts. Fetch in batches of 10 pages at a time (page_size: 100 per page), continuing until a page returns fewer than 100 results (indicating the last page).

Batch loop:

1. Fetch pages 1–10 concurrently
2. Check if any page returned fewer than 100 results → that's the last page, stop
3. If all 10 returned 100 results, fetch pages 11–20, then 21–30, etc.
4. Continue until the last page is found

Extract for each member:

• id
• first_name + last_initial (note: some MCPs use these separate fields instead of full_name)
• completed_class_count
• date_joined
• archived_at

Filter to:

1. Active members only (archived_at is null)
2. Joined within the last 90 days — compute today - 90 days and exclude anyone whose date_joined is older than that cutoff

Step 3: Classify risk tiers

Apply these rules in priority order — check from top to bottom and assign the first matching category:

1. New account — date_joined is within the last 7 days (recency always wins, regardless of class count)
2. Never attended — 0 classes completed
3. Critical — 1 class completed
4. High risk — 2–3 classes completed
5. Watch — 4–5 classes completed
6. Engaged — 6+ classes completed

Compute:

• Total new accounts (all members in the 90-day window)
• Count per tier
• % per tier (of total new accounts)

Step 4: Render the report widget

Use visualize:read_me with modules ["chart", "data_viz"] then visualize:show_widget to render the report.

Report layout (match this exactly)

Metric cards row (4 cards):

1. Total new accounts (last 90 days)
2. Never attended — count of Never attended tier — muted/secondary color
3. Critical — count of Critical tier — danger color
4. High risk — count of High risk tier — warning color

Two charts side by side:

Left — Doughnut chart (6 tiers):

• New account = #378ADD (blue)
• Never attended = #888780 (gray)
• Critical = #E24B4A (red)
• High risk = #EF9F27 (amber)
• Watch = #F5C842 (yellow)
• Engaged = #97C459 (green)
• Custom HTML legend above with tier name, count, and percentage

Right — Bar chart (member count by class bracket):

• Brackets: [0, 1, 2–3, 4–5, 6–10, 11+]
• Colors: gray for 0, red for 1, amber for 2–3, yellow for 4–5, blue/teal for 6+
• X axis label: "Classes attended"

Members table — show Critical, High risk, and Never attended only (exclude Watch, Engaged, and New account):

• Columns: Name | Joined | Classes | Risk level
• Sort order: Critical first, then High risk, then Never attended; within each tier sort by days_ago descending (longest since joining first)
• Risk level column width: 140px (wide enough so "Never attended" fits on one line without wrapping)
• No max-height cap — show ALL rows, do not truncate or scroll
• Row backgrounds:
  • Critical rows: background: var(--color-background-danger)
  • Never attended rows: background: var(--color-background-secondary)
  • High risk rows: alternating transparent / var(--color-background-secondary)
• Risk badges:
  • Critical: bg=#E24B4A, text=#FCEBEB
  • High risk: bg=#EF9F27, text=#412402
  • Never attended: bg=#888780, text=#F1EFE8
• "Joined" column: show as a relative date, e.g. "14 days ago" or "52 days ago"

Action buttons row (.no-print):

1. "Draft re-engagement email ↗" → sendPrompt('Draft a re-engagement email for my at-risk new members')
2. "Download PDF" → client-side via window.print()

Print stylesheet:

<style>
  @media print {
    .no-print { display: none !important; }
    body { -webkit-print-color-adjust: exact; print-color-adjust: exact; }
  }
</style>

Chart.js config notes

• Load from: https://cdnjs.cloudflare.com/ajax/libs/Chart.js/4.4.1/chart.umd.js
• Every canvas needs role="img", aria-label, and fallback text
• Disable default legend: plugins: { legend: { display: false } }
• Use responsive: true, maintainAspectRatio: false
• Canvas wrapper: position: relative; width: 100%; height: 220px
• All chart colors are hardcoded hex (canvas cannot resolve CSS variables)

Design system

• Metric cards: background: var(--color-background-secondary), border-radius: var(--border-radius-md), padding 1rem
• 4-card grid: display: grid; grid-template-columns: repeat(4, 1fr); gap: 12px
• Table: border: 0.5px solid var(--color-border-tertiary), border-radius: var(--border-radius-lg)
• No max-height or overflow-y on the table wrapper — all rows must be visible
• All text uses CSS variables for dark mode compatibility — never hardcode color: #333 etc.
• Buttons: plain <button> tags (pre-styled by the host)
• No emoji, no gradients, no drop shadows
• Sentence case everywhere

Step 5: Follow-up

After rendering, be ready to:

• Draft a re-engagement email targeting new at-risk members
• Drill into a specific member's profile
• Adjust the window (e.g. "show me just the last 30 days")
• Re-run with updated live data