Scenario
You’ve built a 1% Lookalike from your best customers, but you’re missing adjacent segments that could convert. This job takes your Lookalike source data, creates seed audience personas in Mavera, then uses Mave to identify adjacent demographic and psychographic segments you haven’t targeted. The output is new expansion personas you can test with incremental budget.Architecture
Code
import os, requests, time
META = os.environ["META_ACCESS_TOKEN"]
ACCT = os.environ["META_AD_ACCOUNT_ID"]
MV = os.environ["MAVERA_API_KEY"]
GRAPH = "https://graph.facebook.com/v24.0"
MB = "https://app.mavera.io/api/v1"
MH = {"Authorization": f"Bearer {MV}", "Content-Type": "application/json"}
# 1. Find Lookalike Audiences and their seeds
audiences = requests.get(
f"{GRAPH}/{ACCT}/customaudiences",
params={
"access_token": META,
"fields": "id,name,subtype,approximate_count,lookalike_spec,description",
"limit": 50,
},
).json().get("data", [])
lookalikes = [a for a in audiences if a.get("subtype") == "LOOKALIKE"]
print(f"Lookalike audiences: {len(lookalikes)}")
# 2. Get seed audience details
seeds = {}
for la in lookalikes[:5]:
spec = la.get("lookalike_spec", {})
origin_id = spec.get("origin", [{}])[0].get("id") if isinstance(spec.get("origin"), list) else None
if not origin_id:
continue
seed_info = requests.get(
f"{GRAPH}/{origin_id}",
params={"access_token": META, "fields": "id,name,approximate_count,description,subtype"},
).json()
seeds[la["id"]] = {
"lookalike_name": la["name"],
"seed_name": seed_info.get("name", "Unknown"),
"seed_size": seed_info.get("approximate_count", 0),
"seed_description": seed_info.get("description", ""),
"ratio": spec.get("ratio", 0.01),
"country": spec.get("country", "US"),
}
# 3. Get demographic insights on ads using these lookalikes
for la_id, seed in seeds.items():
insights = requests.get(
f"{GRAPH}/{ACCT}/insights",
params={
"access_token": META,
"fields": "impressions,clicks,ctr,actions",
"breakdowns": "age,gender",
"filtering": f'[{{"field":"audience","operator":"CONTAIN","value":"{la_id}"}}]',
"date_preset": "last_90d",
"limit": 50,
},
).json().get("data", [])
top_demos = sorted(insights, key=lambda x: int(x.get("clicks", 0)), reverse=True)[:5]
seed["top_demographics"] = [
{"age": d.get("age"), "gender": d.get("gender"), "clicks": d.get("clicks")}
for d in top_demos
]
time.sleep(0.5)
# 4. Create seed personas
seed_personas = []
for la_id, seed in seeds.items():
demo_desc = ", ".join(
f"{d['gender']} {d['age']} ({d['clicks']} clicks)"
for d in seed.get("top_demographics", [])[:3]
)
persona = requests.post(f"{MB}/personas", headers=MH, json={
"name": f"Seed: {seed['seed_name'][:40]}",
"description": (
f"Seed audience for lookalike '{seed['lookalike_name']}'. "
f"Size: {seed['seed_size']:,}. Country: {seed['country']}. "
f"Top converting demographics: {demo_desc}. "
f"{seed['seed_description']}"
),
"demographic": {
"top_segments": seed.get("top_demographics", []),
"country": seed["country"],
},
}).json()
seed_personas.append({
"persona_id": persona["id"],
"seed_name": seed["seed_name"],
"lookalike_name": seed["lookalike_name"],
})
time.sleep(0.3)
# 5. Mave: identify adjacent expansion segments
seed_summary = "\n".join(
f"- Seed \"{sp['seed_name']}\" → Lookalike \"{sp['lookalike_name']}\""
for sp in seed_personas
)
demo_summary = "\n".join(
f" {s['seed_name']}: {', '.join(f\"{d['gender']} {d['age']}\" for d in s.get('top_demographics',[])[:3])}"
for s in seeds.values()
)
expansion = requests.post(f"{MB}/mave/chat", headers=MH, json={
"message": f"""Analyze these Meta Ads seed audiences and identify adjacent segments for expansion.
SEED AUDIENCES:
{seed_summary}
CONVERTING DEMOGRAPHICS:
{demo_summary}
For each seed audience:
1. What adjacent demographic segments are likely to convert but aren't in the current seed?
2. What psychographic traits unite the converting demographics?
3. Suggest 3-4 expansion personas (new segments to test) with:
- Name, age range, gender, interests
- Why they're adjacent to the seed audience
- Estimated overlap with current targeting (low/medium/high)
- Recommended ad angle for this segment
4. Which expansion segment should be tested first and why?"""
}).json()
print("\n=== Expansion Analysis ===")
print(expansion.get("content", "")[:2000])
# 6. Create expansion personas from Mave's recommendations
expansion_prompt = requests.post(f"{MB}/mave/chat", headers=MH, json={
"message": f"""Based on your expansion analysis, output exactly 4 expansion personas in this format:
Name: [persona name]
Age: [range]
Gender: [any/male/female]
Interests: [comma-separated]
Description: [2 sentences]
Only output the 4 personas, no other text."""
}).json()
expansion_text = expansion_prompt.get("content", "")
personas_blocks = [b.strip() for b in expansion_text.split("\n\n") if b.strip().startswith("Name:")]
expansion_personas = []
for block in personas_blocks[:4]:
lines = {l.split(":")[0].strip(): ":".join(l.split(":")[1:]).strip()
for l in block.split("\n") if ":" in l}
if not lines.get("Name"):
continue
p = requests.post(f"{MB}/personas", headers=MH, json={
"name": f"Expansion: {lines['Name']}",
"description": lines.get("Description", "Expansion segment from Mave analysis."),
"demographic": {
"age_range": lines.get("Age", "25-54"),
"gender": lines.get("Gender", "any"),
"interests": [i.strip() for i in lines.get("Interests", "").split(",")],
},
}).json()
expansion_personas.append({"name": lines["Name"], "id": p["id"]})
time.sleep(0.3)
print(f"\nCreated {len(expansion_personas)} expansion personas:")
for ep in expansion_personas:
print(f" {ep['name']} → {ep['id']}")
const META = process.env.META_ACCESS_TOKEN;
const ACCT = process.env.META_AD_ACCOUNT_ID;
const MV = process.env.MAVERA_API_KEY;
const GRAPH = "https://graph.facebook.com/v24.0";
const MB = "https://app.mavera.io/api/v1";
const MH = { Authorization: `Bearer ${MV}`, "Content-Type": "application/json" };
// 1. Lookalike Audiences
const audiences = await fetch(
`${GRAPH}/${ACCT}/customaudiences?access_token=${META}&fields=id,name,subtype,approximate_count,lookalike_spec,description&limit=50`
).then(r => r.json()).then(d => d.data || []);
const lookalikes = audiences.filter(a => a.subtype === "LOOKALIKE").slice(0, 5);
console.log(`Lookalikes: ${lookalikes.length}`);
// 2. Seed details
const seeds = {};
for (const la of lookalikes) {
const originId = Array.isArray(la.lookalike_spec?.origin) ? la.lookalike_spec.origin[0]?.id : null;
if (!originId) continue;
const seedInfo = await fetch(
`${GRAPH}/${originId}?access_token=${META}&fields=id,name,approximate_count,description`
).then(r => r.json());
seeds[la.id] = {
lookalike_name: la.name, seed_name: seedInfo.name || "Unknown",
seed_size: seedInfo.approximate_count || 0, seed_description: seedInfo.description || "",
ratio: la.lookalike_spec?.ratio || 0.01, country: la.lookalike_spec?.country || "US",
};
}
// 3. Demographic insights
for (const [laId, seed] of Object.entries(seeds)) {
const insights = await fetch(
`${GRAPH}/${ACCT}/insights?access_token=${META}&fields=impressions,clicks,ctr,actions&breakdowns=age,gender&filtering=[{"field":"audience","operator":"CONTAIN","value":"${laId}"}]&date_preset=last_90d&limit=50`
).then(r => r.json()).then(d => d.data || []);
seed.top_demographics = insights
.sort((a, b) => parseInt(b.clicks || "0") - parseInt(a.clicks || "0"))
.slice(0, 5).map(d => ({ age: d.age, gender: d.gender, clicks: d.clicks }));
await new Promise(r => setTimeout(r, 500));
}
// 4. Seed personas
const seedPersonas = [];
for (const [laId, seed] of Object.entries(seeds)) {
const demoDesc = (seed.top_demographics || []).slice(0, 3)
.map(d => `${d.gender} ${d.age} (${d.clicks} clicks)`).join(", ");
const persona = await fetch(`${MB}/personas`, {
method: "POST", headers: MH,
body: JSON.stringify({
name: `Seed: ${(seed.seed_name || "").slice(0, 40)}`,
description: `Seed for '${seed.lookalike_name}'. Size: ${seed.seed_size.toLocaleString()}. Top demos: ${demoDesc}.`,
demographic: { top_segments: seed.top_demographics, country: seed.country },
}),
}).then(r => r.json());
seedPersonas.push({ persona_id: persona.id, seed_name: seed.seed_name, lookalike_name: seed.lookalike_name });
await new Promise(r => setTimeout(r, 300));
}
// 5. Mave expansion analysis
const seedSummary = seedPersonas.map(sp => `- "${sp.seed_name}" → "${sp.lookalike_name}"`).join("\n");
const demoSummary = Object.values(seeds).map(s =>
` ${s.seed_name}: ${(s.top_demographics || []).slice(0, 3).map(d => `${d.gender} ${d.age}`).join(", ")}`
).join("\n");
const expansion = await fetch(`${MB}/mave/chat`, {
method: "POST", headers: MH,
body: JSON.stringify({
message: `Analyze seeds and find expansion segments:\n\nSEEDS:\n${seedSummary}\n\nDEMOS:\n${demoSummary}\n\nFor each: 1) Adjacent segments 2) Unifying psychographics 3) 3-4 expansion personas 4) Test priority`,
}),
}).then(r => r.json());
console.log("\n=== Expansion Analysis ===");
console.log((expansion.content || "").slice(0, 2000));
// 6. Create expansion personas
const expPrompt = await fetch(`${MB}/mave/chat`, {
method: "POST", headers: MH,
body: JSON.stringify({
message: "Output exactly 4 expansion personas with Name, Age, Gender, Interests, Description. No other text.",
}),
}).then(r => r.json());
const blocks = (expPrompt.content || "").split("\n\n").filter(b => b.trim().startsWith("Name:"));
const expansionPersonas = [];
for (const block of blocks.slice(0, 4)) {
const lines = Object.fromEntries(
block.split("\n").filter(l => l.includes(":")).map(l => {
const [k, ...v] = l.split(":");
return [k.trim(), v.join(":").trim()];
})
);
if (!lines.Name) continue;
const p = await fetch(`${MB}/personas`, {
method: "POST", headers: MH,
body: JSON.stringify({
name: `Expansion: ${lines.Name}`, description: lines.Description || "Expansion segment.",
demographic: {
age_range: lines.Age || "25-54", gender: lines.Gender || "any",
interests: (lines.Interests || "").split(",").map(i => i.trim()),
},
}),
}).then(r => r.json());
expansionPersonas.push({ name: lines.Name, id: p.id });
await new Promise(r => setTimeout(r, 300));
}
console.log(`\nCreated ${expansionPersonas.length} expansion personas:`);
expansionPersonas.forEach(ep => console.log(` ${ep.name} → ${ep.id}`));
Example Output
=== Expansion Analysis ===
## Seed: "High-Value Purchasers" → 1% Lookalike
**Converting demographics:** Female 25-34 (42%), Male 35-44 (28%), Female 35-44 (18%)
**Unifying psychographics:** Career-driven, value efficiency, early adopters, active on Instagram
### Expansion Personas
1. **"Career-Switching Millennials"** (M/F, 28-36)
Adjacent because: Share career-driven traits but are in transition — higher urgency
Overlap: Low — different life stage signals
Ad angle: "Made for people building what's next"
2. **"Gen X Decision Makers"** (M/F, 45-54)
Adjacent because: Have budget authority your 35-44 segment aspires to
Overlap: Medium — some already in 2% lookalike
Ad angle: "Your team will thank you" (delegation framing)
3. **"Side-Hustle Creators"** (M/F, 22-30)
Adjacent because: Early adopter behavior matches, but lower purchase power
Overlap: Low — different income tier
Ad angle: "Start free. Scale when you're ready."
4. **"Remote-First Team Leads"** (M/F, 30-42)
Adjacent because: Efficiency-oriented, digital-native workflow preferences
Overlap: Medium — behavioral overlap, different job title signals
Ad angle: "Built for teams that never meet in person"
**Test first:** "Career-Switching Millennials" — lowest overlap means most incremental
reach. High urgency drives faster conversion cycles.
Created 4 expansion personas:
Career-Switching Millennials → per_exp_csm_1
Gen X Decision Makers → per_exp_gx_2
Side-Hustle Creators → per_exp_shc_3
Remote-First Team Leads → per_exp_rft_4
Error Handling
Lookalike spec structure varies
Lookalike spec structure varies
The
lookalike_spec.origin field can be an array of objects or a single ID depending on API version. Always check if it’s an array first.Filtering by audience ID in Insights
Filtering by audience ID in Insights
The
filtering parameter uses a JSON array string. Encoding issues cause silent failures — validate the JSON structure before sending.Mave persona extraction is best-effort
Mave persona extraction is best-effort
The structured output from Mave’s second call may not perfectly match the expected format. Add fallback parsing for variations in line formatting.
Expansion personas need validation
Expansion personas need validation
These are hypothetical segments. Test with small budgets ($50-100/day) before scaling. Track CPA against your seed audience as the benchmark.
All Meta Ads Jobs
Browse all Meta Ads integration jobs
Personas
Creating and managing personas