Campaign Event Analysis → Messaging Strategy

Pull Customer.io campaign metrics, identify winning message patterns, and generate new variations using Mave Agent and Mavera Generate

Scenario

Your Customer.io campaigns generate open, click, and conversion metrics across email, push, and in-app channels. But you never synthesize which message types actually perform — is it urgency-driven subject lines? Personalized value props? Social proof? This job pulls campaign metrics from the App API, identifies the highest-performing message patterns, then feeds those patterns to Mave Agent for analysis and Mavera Generate to produce new variations that extend the winning formula.

Flow: Customer.io App API /v1/campaigns/{id}/metrics → Rank campaigns by conversion → Extract winning copy patterns → Mave POST /api/v1/mave/chat (pattern analysis) → POST /api/v1/generations (new variations)

Architecture

Code

Python

import os, requests, base64, time

CIO_APP = os.environ["CIO_APP_KEY"]
MV = os.environ["MAVERA_API_KEY"]
APP_BASE = "https://api.customer.io/v1"
MB = "https://app.mavera.io/api/v1"
APP_H = {"Authorization": f"Bearer {CIO_APP}"}
MV_H = {"Authorization": f"Bearer {MV}", "Content-Type": "application/json"}

# 1. List campaigns
campaigns = requests.get(f"{APP_BASE}/campaigns",
    headers=APP_H).json().get("campaigns", [])

# 2. Pull metrics for each active campaign
ranked = []
for camp in campaigns:
    if camp.get("state") != "active":
        continue
    cid = camp["id"]
    metrics = requests.get(f"{APP_BASE}/campaigns/{cid}/metrics",
        headers=APP_H, params={"period": "days", "steps": 30})
    if metrics.status_code == 429:
        time.sleep(1)
        metrics = requests.get(f"{APP_BASE}/campaigns/{cid}/metrics",
            headers=APP_H, params={"period": "days", "steps": 30})
    metrics.raise_for_status()
    data = metrics.json()

    series = data.get("series", {})
    sent = sum(series.get("sent", [0]))
    clicked = sum(series.get("clicked", [0]))
    converted = sum(series.get("converted", [0]))
    ctr = clicked / max(sent, 1)

    ranked.append({
        "id": cid,
        "name": camp.get("name", "Untitled"),
        "type": camp.get("type", "email"),
        "sent": sent, "clicked": clicked, "converted": converted,
        "ctr": ctr,
    })
    time.sleep(0.12)

ranked.sort(key=lambda x: -x["ctr"])
top = ranked[:5]
bottom = ranked[-3:] if len(ranked) >= 8 else []

# 3. Pull message content for top campaigns
top_copy = []
for camp in top:
    actions = requests.get(f"{APP_BASE}/campaigns/{camp['id']}/actions",
        headers=APP_H).json().get("actions", [])
    for act in actions[:2]:
        body = act.get("body", "")[:500]
        subject = act.get("subject", act.get("name", ""))
        if subject or body:
            top_copy.append(
                f"Campaign: {camp['name']} (CTR: {camp['ctr']:.1%})\n"
                f"Subject: {subject}\n{body}"
            )

# 4. Mave pattern analysis
analysis = requests.post(f"{MB}/mave/chat", headers=MV_H, json={
    "message": f"""Analyze these top-performing Customer.io campaign messages.
Identify: 1) Common structural patterns 2) Tone and voice traits
3) Subject line formulas 4) CTA patterns 5) What the bottom performers lack

TOP PERFORMERS:
{chr(10).join(top_copy[:8])}

BOTTOM PERFORMERS (for contrast):
{chr(10).join(f"- {c['name']} (CTR: {c['ctr']:.1%})" for c in bottom)}"""
}).json()

print("=== Pattern Analysis ===")
print(analysis.get("content", "")[:1500])

# 5. Generate new variations
gen = requests.post(f"{MB}/generations", headers=MV_H, json={
    "prompt": (
        f"Based on these winning patterns from our email campaigns, "
        f"generate 5 new email subject line + 100-word body variations. "
        f"Topic: product feature announcement. "
        f"Patterns to follow: {analysis.get('content', '')[:800]}"
    ),
    "count": 5,
}).json()

print("\n=== New Variations ===")
for i, g in enumerate(gen.get("results", [gen]), 1):
    print(f"\nVariant {i}:\n{g.get('content', g.get('text', ''))[:300]}")

Example Output

=== Pattern Analysis ===
## Winning Patterns (5 campaigns, avg CTR 12.4%)

1. **Subject Line Formula:** [Number] + [Benefit] + [Timeframe]
   - "3 ways to cut onboarding time this week"
   - "Your Q2 metrics — 2 quick wins inside"

2. **Tone:** Direct, peer-to-peer, avoids marketing jargon.
   Uses "you/your" 3x more than bottom performers.

3. **CTA Pattern:** Single action, specific timeframe.
   Winners: "Reply with your top priority" (8.2% reply rate)
   Losers: "Learn more" (0.3% CTR)

4. **Structure:** <100 words, one idea per email, PS line with social proof.

5. **Bottom performers lack:** Personalization, specificity, urgency framing.

=== New Variations ===
Variant 1:
Subject: Your onboarding is 40% slower than peers — here's the fix
We analyzed 200 teams like yours. The gap? Manual steps between...

Variant 2:
Subject: 2 changes. 15 minutes. Measurable pipeline lift.
Last month, [Similar Co] made two adjustments to their workflow...

Error Handling

App API rate limit (10 req/sec)

The App API is strict at 10 req/sec. Code includes 120ms delay between requests. For workspaces with 50+ campaigns, batch in groups of 8 with 1s pauses between batches.

Empty metric series

New campaigns may return empty series arrays. The code defaults to [0] to avoid division errors. Filter out campaigns with fewer than 100 sends for meaningful CTR.

Campaign actions structure

The /actions endpoint returns message variants. Multi-variant campaigns have multiple actions. The code limits to 2 per campaign to stay within rate limits.

---

What’s Next

Customer.io Integration

Back to Customer.io integration overview

Customer Attribute Personas

Build attribute-clustered personas from Customer.io segments

Mave Agent

Full reference for POST /api/v1/mave/chat

Generate API

Full reference for POST /api/v1/generations