Call Recording Analysis

Transcribe Close CRM call recordings and feed them into Mave Agent for structured meeting analysis — decisions, objections, next steps, and coaching insights

Scenario

Close stores call recordings with metadata (duration, direction, user). You want to transcribe recordings and feed them into Mave Agent for structured meeting analysis: decisions made, objections raised, next steps promised, and coaching insights.

Architecture

Code

Python

import os, requests
from requests.auth import HTTPBasicAuth
from openai import OpenAI

CLOSE_KEY = os.environ["CLOSE_API_KEY"]
CLOSE_AUTH = HTTPBasicAuth(CLOSE_KEY, "")
MAVERA_KEY = os.environ["MAVERA_API_KEY"]

def close_get(path, params=None):
    r = requests.get(f"https://api.close.com/api/v1{path}", auth=CLOSE_AUTH, params=params or {})
    r.raise_for_status()
    return r.json()

# 1. Pull calls with recordings
calls = [c for c in close_get("/activity/call/", {"_limit": 50}).get("data", []) if c.get("recording_url")]
if not calls:
    print("No calls with recordings found.")
    exit()

call = calls[0]

# 2. Download and transcribe via Whisper
audio = requests.get(call["recording_url"], auth=CLOSE_AUTH)
audio.raise_for_status()
audio_path = "/tmp/close_call.mp3"
with open(audio_path, "wb") as f:
    f.write(audio.content)

whisper = OpenAI(api_key=os.environ.get("OPENAI_API_KEY", ""))
with open(audio_path, "rb") as af:
    transcription = whisper.audio.transcriptions.create(model="whisper-1", file=af, response_format="text")
transcript = transcription if isinstance(transcription, str) else transcription.text

# 3. Feed to Mave Agent
mave = requests.post(
    "https://app.mavera.io/api/v1/mave/chat",
    headers={"Authorization": f"Bearer {MAVERA_KEY}"},
    json={"message": (
        f"Analyze this sales call transcript. Extract:\n"
        f"1. Key decisions made\n2. Objections raised\n"
        f"3. Next steps and commitments\n4. Coaching insights for the rep\n"
        f"5. Call quality score (1-10)\n\n"
        f"Duration: {call.get('duration', 'unknown')}s | Direction: {call.get('direction', 'unknown')}\n\n"
        f"Transcript:\n{transcript[:8000]}"
    )},
)
mave.raise_for_status()
print(f"Call Analysis for {call.get('phone', 'unknown')}")
print(f"Duration: {call.get('duration', 0)}s\n")
print(mave.json().get("content", "")[:1000])

Example Output

Call Analysis for [phone] | 1847s

**Decisions:** Prospect agreed to technical deep-dive (next Thursday).
Budget deferred to Q2 planning.

**Objections:** (1) "Locked into 2-year contract" — handled well.
(2) "No bandwidth for implementation" — offered onboarding but
didn't quantify time savings.

**Next Steps:** Send case study by EOD Friday. Schedule 30-min demo
with CTO + 2 engineers. Research competitor contract clause.

**Coaching:** Great discovery first 10 min. Missed ROI quantification.
Talk-to-listen 60/40; aim 40/60. **Quality: 7/10**

Error Handling

Error	Cause	Fix
recording_url returns 403	Recording expired	Check Close recording retention settings
Whisper timeout	File > 25 MB	Split audio or use Deepgram streaming
Garbled transcript	Poor audio, multiple speakers	Use Deepgram with diarization
Mave truncation	Transcript too long	Send first 8,000 chars or pre-summarize