Source Effectiveness → Recruitment Marketing

Analyze Lever postings and source data with Mave and Generate for source-optimized recruitment content

Scenario

Your Lever postings attract candidates from different sources — LinkedIn, Indeed, referrals, your careers page, agencies. But which sources produce the best candidates? And how should your recruitment marketing differ by source? You pull posting performance data with source attribution, analyze source effectiveness with Mave, then generate source-optimized recruitment content.

Flow: Lever GET /postings + GET /opportunities (source data) → Mave POST /mave/chat (source analysis) → POST /generations (source-optimized content)

Architecture

Code

Python

import os, requests, time, base64
from collections import defaultdict

LV_KEY = os.environ["LEVER_API_KEY"]
MV = os.environ["MAVERA_API_KEY"]
LV_BASE = "https://api.lever.co/v1"
MV_BASE = "https://app.mavera.io/api/v1"
MV_H = {"Authorization": f"Bearer {MV}", "Content-Type": "application/json"}

lv_auth = base64.b64encode(f"{LV_KEY}:".encode()).decode()
LV_H = {"Authorization": f"Basic {lv_auth}"}

def lv_get(path, params=None):
    r = requests.get(f"{LV_BASE}{path}", headers=LV_H, params=params or {})
    if r.status_code == 429:
        time.sleep(1)
        return lv_get(path, params)
    r.raise_for_status()
    return r.json()

# 1. Pull active postings
postings = lv_get("/postings", {"state": "published"}).get("data", [])

# 2. Pull opportunities with source data
opps = []
offset = None
while len(opps) < 800:
    params = {"limit": 100}
    if offset:
        params["offset"] = offset
    resp = lv_get("/opportunities", params)
    opps.extend(resp.get("data", []))
    offset = resp.get("next")
    if not offset:
        break
    time.sleep(0.15)

# 3. Aggregate source performance
source_stats = defaultdict(lambda: {"applied": 0, "screened": 0, "onsited": 0, "offered": 0, "hired": 0})

stage_resp = lv_get("/stages")
stage_names = {s["id"]: s["text"].lower() for s in stage_resp.get("data", [])}

for opp in opps:
    sources = opp.get("sources", [])
    source = sources[0] if sources else "Direct"
    stage_name = stage_names.get(opp.get("stage", ""), "unknown")

    source_stats[source]["applied"] += 1
    if any(kw in stage_name for kw in ["screen", "phone"]):
        source_stats[source]["screened"] += 1
    if "onsite" in stage_name or "final" in stage_name:
        source_stats[source]["onsited"] += 1
    if "offer" in stage_name:
        source_stats[source]["offered"] += 1
    if opp.get("isArchived") and "hired" in stage_name:
        source_stats[source]["hired"] += 1

# 4. Mave analysis
source_block = "\n".join(
    f"- {src}: Applied={s['applied']} Screen={s['screened']} Onsite={s['onsited']} Offer={s['offered']} Hired={s['hired']}"
    for src, s in sorted(source_stats.items(), key=lambda x: -x[1]["applied"])[:10]
)

analysis = requests.post(f"{MV_BASE}/mave/chat", headers=MV_H, json={
    "message": f"""Analyze recruitment source effectiveness from this Lever data:

{source_block}

For each source:
1. Conversion rate through funnel stages
2. Quality indicator (hired/applied ratio)
3. Where candidates drop off
4. Recommended recruitment marketing strategy for this channel
5. What messaging resonates on this channel vs others"""
}).json()

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

# 5. Generate source-optimized content
for source in list(source_stats.keys())[:3]:
    stats = source_stats[source]
    gen = requests.post(f"{MV_BASE}/generations", headers=MV_H, json={
        "prompt": (
            f"Write a recruitment marketing post optimized for {source}. "
            f"This source has {stats['applied']} applicants with {stats['hired']} hires "
            f"({stats['hired']/max(stats['applied'],1)*100:.0f}% conversion). "
            f"Match the tone and format conventions of {source}. 150-200 words."
        ),
    }).json()
    print(f"\n--- {source} Content ---")
    print(gen.get("output", gen.get("content", gen.get("text", "")))[:500]
    time.sleep(0.5)

Example Output

=== Source Effectiveness Analysis ===

## Source Performance Ranking

| Source | Applied | Hired | Conversion | Quality |
|--------|---------|-------|-----------|---------|
| Referral | 89 | 23 | 25.8% | ★★★★★ |
| LinkedIn | 312 | 18 | 5.8% | ★★★ |
| Careers Page | 201 | 12 | 6.0% | ★★★ |
| Indeed | 187 | 4 | 2.1% | ★★ |

## Key Insights
- **Referrals** convert at 4.5x the rate of LinkedIn but generate 3.5x fewer applicants.
  → Invest in referral bonus program and make sharing easier.
- **Indeed** has highest volume but lowest conversion — candidates drop at Screen stage.
  → Tighten job description requirements to pre-filter.
- **Careers Page** converts well but lacks volume.
  → SEO + content marketing to drive organic traffic.

--- Referral Content ---
Know someone who'd thrive here? Our best hires come from people like you.
We're looking for [role] — someone who [specific trait]. Your referral gets
fast-tracked to a hiring manager call within 48 hours...

Error Handling

Source field is an array

Lever stores sources as an array (candidates can have multiple). The code uses the first source. For multi-source attribution, count each source separately.

Archived vs active

Hired candidates are often isArchived: true with their last stage containing “hired”. Check both fields for accurate hire counts.

Custom stage names

Stage names are fully customizable in Lever. The keyword matching (screen, onsite, offer) may miss custom names. Audit your stages with GET /stages first.