> ## Documentation Index > Fetch the complete documentation index at: https://docs.mavera.io/llms.txt > Use this file to discover all available pages before exploring further. # Cohort Analysis → Content Strategy > Compare retained vs. churned user behaviors from Mixpanel and generate retention content strategies with Mave Agent ### Scenario Retained users behave differently from churned users — they use different features, visit at different times, and engage with different content. You pull cohort data from Mixpanel comparing retained vs. churned user behaviors, then send the behavioral differences to Mave with the instruction to research strategies for converting churning behaviors into retained behaviors. ### Architecture ```mermaid theme={"dark"} flowchart LR A["Mixpanel cohort data (retained vs churned)"] --> B[Extract behavioral differences] --> C["POST /api/v1/mave/chat"] --> D[Retention strategies from behavioral gaps] ``` ### Code ```python Python theme={"dark"} import os, requests, time MP_SA = os.environ["MIXPANEL_SERVICE_ACCOUNT"] MP_SECRET = os.environ["MIXPANEL_SECRET"] MP_PROJECT = os.environ["MIXPANEL_PROJECT_ID"] MV = os.environ["MAVERA_API_KEY"] mp_auth = (MP_SA, MP_SECRET) def query_engage_segment(filter_expr, props): r = requests.post("https://mixpanel.com/api/query/engage", auth=mp_auth, json={ "project_id": MP_PROJECT, "filter_by_cohort": filter_expr, "output_properties": props, "page": 0, }) if r.status_code == 429: time.sleep(60) r.raise_for_status() return r.json().get("results", []) PROFILE_PROPS = [ "plan", "role", "company_size", "signup_source", "total_events", "sessions_count", "features_used", "days_active_last_30", "last_seen", ] retained = query_engage_segment( '(properties["days_active_last_30"]) >= 10', PROFILE_PROPS ) time.sleep(2) churned = query_engage_segment( '(properties["days_active_last_30"]) <= 1 and (properties["sessions_count"]) >= 5', PROFILE_PROPS ) def summarize(users, label): if not users: return f"{label}: No users found" props_list = [u.get("$properties", {}) for u in users] avg_events = sum(int(p.get("total_events", 0) or 0) for p in props_list) / len(props_list) avg_sessions = sum(int(p.get("sessions_count", 0) or 0) for p in props_list) / len(props_list) avg_days = sum(int(p.get("days_active_last_30", 0) or 0) for p in props_list) / len(props_list) from collections import Counter plans = Counter(p.get("plan") for p in props_list if p.get("plan")).most_common(3) roles = Counter(p.get("role") for p in props_list if p.get("role")).most_common(5) sources = Counter(p.get("signup_source") for p in props_list if p.get("signup_source")).most_common(3) features = Counter( f for p in props_list for f in (p.get("features_used", "").split(",") if isinstance(p.get("features_used"), str) else []) if f.strip() ).most_common(10) return ( f"**{label}** ({len(users)} users)\n" f" Avg events: {avg_events:.0f} | Avg sessions: {avg_sessions:.0f} | Avg days active (30d): {avg_days:.1f}\n" f" Plans: {', '.join(f'{p[0]} ({p[1]})' for p in plans)}\n" f" Roles: {', '.join(f'{r[0]} ({r[1]})' for r in roles)}\n" f" Sources: {', '.join(f'{s[0]} ({s[1]})' for s in sources)}\n" f" Top features: {', '.join(f'{f[0]} ({f[1]})' for f in features)}" ) retained_summary = summarize(retained, "Retained (10+ days active/30d)") churned_summary = summarize(churned, "Churned (≤1 day active, had 5+ sessions)") mave = requests.post( "https://app.mavera.io/api/v1/mave/chat", headers={"Authorization": f"Bearer {MV}", "Content-Type": "application/json"}, json={"message": f"""Research strategies to convert churning user behaviors into retained user behaviors. RETAINED USERS: {retained_summary} CHURNED USERS: {churned_summary} Analyze: 1. Key behavioral differences between retained and churned 2. Which features correlate with retention? Which are missing from churned? 3. Content strategies to drive feature adoption among at-risk users 4. Onboarding improvements to increase early activation 5. Re-engagement campaign ideas for users showing churn signals 6. Messaging recommendations: what to say at each stage of the retention curve"""}, ).json() print("--- Cohort Content Strategy ---") print(mave.get("content", "")[:3000]) ``` ```javascript JavaScript theme={"dark"} const MP_SA = process.env.MIXPANEL_SERVICE_ACCOUNT; const MP_SECRET = process.env.MIXPANEL_SECRET; const MP_PROJECT = process.env.MIXPANEL_PROJECT_ID; const MV = process.env.MAVERA_API_KEY; const mpAuth = "Basic " + Buffer.from(`${MP_SA}:${MP_SECRET}`).toString("base64"); const PROPS = ["plan", "role", "company_size", "signup_source", "total_events", "sessions_count", "features_used", "days_active_last_30"]; async function querySegment(filter) { const res = await fetch("https://mixpanel.com/api/query/engage", { method: "POST", headers: { Authorization: mpAuth, "Content-Type": "application/json" }, body: JSON.stringify({ project_id: MP_PROJECT, filter_by_cohort: filter, output_properties: PROPS, page: 0 }), }); if (res.status === 429) { await new Promise((r) => setTimeout(r, 60000)); return querySegment(filter); } return (await res.json()).results || []; } const retained = await querySegment('(properties["days_active_last_30"]) >= 10'); await new Promise((r) => setTimeout(r, 2000)); const churned = await querySegment('(properties["days_active_last_30"]) <= 1 and (properties["sessions_count"]) >= 5'); function summarize(users, label) { const props = users.map((u) => u.$properties || {}); const avg = (key) => props.reduce((s, p) => s + parseInt(p[key] || "0"), 0) / (props.length || 1); const topN = (key, n = 3) => { const c = {}; props.forEach((p) => { if (p[key]) c[p[key]] = (c[p[key]] || 0) + 1; }); return Object.entries(c).sort(([, a], [, b]) => b - a).slice(0, n).map(([k, v]) => `${k} (${v})`).join(", "); }; return `**${label}** (${users.length} users)\n Avg events: ${avg("total_events").toFixed(0)} | Sessions: ${avg("sessions_count").toFixed(0)} | Days active: ${avg("days_active_last_30").toFixed(1)}\n Plans: ${topN("plan")} | Roles: ${topN("role", 5)} | Sources: ${topN("signup_source")}`; } const mave = await fetch("https://app.mavera.io/api/v1/mave/chat", { method: "POST", headers: { Authorization: `Bearer ${MV}`, "Content-Type": "application/json" }, body: JSON.stringify({ message: `Research strategies to convert churning behaviors into retained behaviors.\n\nRETAINED:\n${summarize(retained, "Retained (10+ days/30d)")}\n\nCHURNED:\n${summarize(churned, "Churned (≤1 day, had 5+ sessions)")}\n\nAnalyze: 1) Behavioral differences 2) Feature correlations 3) Content strategies 4) Onboarding improvements 5) Re-engagement campaigns 6) Messaging at each retention stage`, }), }).then((r) => r.json()); console.log("--- Cohort Content Strategy ---"); console.log((mave.content || "").slice(0, 3000)); ``` ### Example Output ```text theme={"dark"} --- Cohort Content Strategy --- ## Key Behavioral Differences | Metric | Retained | Churned | |--------|----------|---------| | Avg events | 842 | 67 | | Days active (30d) | 18.4 | 0.6 | | Top features | Dashboard, API, Reports, Alerts | Dashboard only | | Plans | Pro (61%), Enterprise (24%) | Free (78%), Starter (18%) | ## Feature-Retention Correlation Users who activate **3+ features in the first 7 days** retain at 4.2x the rate of those who use only the dashboard. The "Aha features" are: 1. **API integration** — 89% retention when activated in week 1 2. **Alerts/notifications** — 72% retention 3. **Custom reports** — 68% retention ## Content Strategies for At-Risk Users 1. **Day 3 email:** "You've set up your dashboard — here's what power users do next" → tutorial link to API integration 2. **Day 7 in-app prompt:** "Set up your first alert in 2 minutes" → guided flow 3. **Day 14 re-engagement:** "Your data is waiting — 3 insights you're missing" ## Messaging by Stage - **Day 1-3:** "Quick win" focus — show immediate value - **Day 4-7:** "Level up" — introduce second and third features - **Day 8-14:** "Team value" — show collaboration features - **Day 15-30:** "ROI proof" — send usage stats and time-saved metrics ``` ### Error Handling Mixpanel's Engage API uses a specific filter expression syntax. Test filters in the Mixpanel UI's Explore section before using them in API calls. Common mistake: using `==` instead of `>=` for numeric comparisons. The code defines "retained" as 10+ days active in 30d and "churned" as ≤1 day active with 5+ historical sessions. Adjust thresholds to match your product's engagement model. Two Engage queries back-to-back count against the 60/hour limit. The 2s delay between queries is a courtesy; the actual constraint is 60/hour total. *** ## What's Next Back to Mixpanel integration overview Investigate funnel abandonment with focus groups Enrich personas with raw event patterns Full reference for POST /api/v1/mave/chat