Post-call analytics · Vapi & Retell

Your voice agent “resolves” 71% of calls.
Only 48% actually got helped.

Vapi and Retell tell you when a call ends. Neither tells you if the customer got what they needed. ConvoMetrics reads every transcript, classifies intent and outcome, and pushes structured events to Slack, PostHog, or Mixpanel. Works with chat too—Intercom, Zendesk, or any webhook.

Connects to: Vapi Retell Intercom Zendesk Custom webhook
Pushes to: Slack PostHog Mixpanel Zapier Make
Platform Completion Rate
0%
what Vapi / Retell reports
Actual Resolution Rate
0%
after transcript analysis
↑ 23-point gap
Vapi and Retell have no post-call resolution analytics. This is the gap.
4,847 calls + chats analyzed · Mar 31 – Apr 6, 2026

You’re flying blind.

Vapi and Retell report call completion. Intercom reports ticket closure. Nobody reports whether the customer actually got what they needed.

71%
Completion ≠ Resolution
Your voice agent marks a call “done” when the caller hangs up. Your chat agent marks it “done” when the user stops typing. Neither knows if the issue was resolved. That’s the number you report.
96%
The polite churner problem
96% of customers who hit a dead end never complain. They say “ok thanks” and your agent logs positive sentiment. Resolved. Satisfied. 47 of them cancelled within a week. This pattern is invisible without transcript-level analysis.
1%
You can’t review every conversation
At 500 calls and chats a week, you spot-check maybe five. You’re shipping prompt updates based on 1% of conversations. Hamming and Coval test before launch. Nobody analyzes what happens after.

Three steps. One briefing. The truth.

01
Connect your agent
Paste a Vapi or Retell API key. Or connect Intercom, Zendesk, anything—via Zapier, Make, or webhook. Five minutes.
vapi · retell · zapier · make · webhook
02
Every conversation classified
LLM reads every transcript. Intent, outcome, failure pattern. Voice or chat. No manual tagging.
automatic · no labeling
03
Structured events, not dashboards
Events push to PostHog, Mixpanel, or anywhere via Zapier. Weekly briefing lands in Slack every Monday.
slack · posthog · mixpanel · zapier

What lands in your Slack every Monday.

convometrics — weekly briefing
⚡ WEEKLY BRIEFING — Vapi Voice Agent + Intercom Chat
period: Mar 31 – Apr 6, 2026
transcripts analyzed: 4,847 (3,102 calls · 1,745 chats)
platform completion rate: 71%
actual resolution rate: 48% ← 23pt gap
neither Vapi nor Intercom reports this number.
top failing intents:
  billing_dispute_with_context — 412 transcripts, 14% resolved
  cancel_subscription_flow — 289 transcripts, 22% resolved
  reschedule_appointment — 847 transcripts, 91% resolved
🔍 hidden pattern: polite_churner
  283 customers said “ok thanks” after a failed response.
  agent flagged all as resolved. positive sentiment.
  47 of them cancelled within 7 days.
PUSHED TO: Slack ✓  PostHog ✓  Zapier ✓
Every number is computed from real transcripts, not estimated.

Fix it, not just find it.

ConvoMetrics doesn’t just tell you what’s failing. It shows you exactly how to fix your agent’s prompts—with a before/after rewrite based on real failure patterns.

billing_dispute_with_context — 412 transcripts/week, 14% first-contact resolution.
Your agent answers the question customers didn’t ask.
Before — current prompt
When a user asks about billing, retrieve their current plan details and explain the features and pricing of their active subscription. Direct them to the help center for further billing questions.
After — suggested rewrite
When a user asks about billing, first check if they are referencing a past charge or historical event (e.g. “I was charged”, “last week”, “I downgraded”). If so, retrieve their billing history and recent plan changes—not just current plan state. Acknowledge the specific charge before explaining. If a refund may be owed, initiate the refund flow directly.
A customer says “I was charged $49 but I downgraded last week.” Your agent explains what the Free plan includes. The customer says “I KNOW I’m on free. I’m asking about the CHARGE.” Your agent sends them to the help center. They hang up. Your dashboard marks it resolved.
~200 fewer escalations/week +0.5pt human CSAT recovery 412 transcripts/week affected
Every suggestion is generated from clustered failure transcripts—not templates.

Stop guessing what your agent gets wrong.

Paste an API key. Get your first briefing. Free to start. No credit card.

Vapi · Retell · Intercom · Zendesk · Zapier · Make · Any webhook