New ebook |

Research

Voice AI benchmark methodology

Help Genie is building an original, citable benchmark from real production voice conversations — after-hours share, call duration, handoff rates, and industry patterns. We have not published the dataset yet. This page explains how we will do it honestly when aggregate access is available.

Status (July 2026): Call-level tables are RLS-gated from this marketing repo. Publishing “what N thousand calls reveal” without a secure aggregate view would be misleading. See data/seo/pr-data-asset-findings.md in the repository.

Publication rules

  • Aggregate stats only — no transcripts, names, numbers, or tenant IDs
  • Minimum 50 conversations per published metric
  • Exclude demo genies, marketplace seeds, and internal test accounts
  • Date-stamped cohorts with documented instrumentation versions
  • Privacy reviewer sign-off before any public chart ships

Metrics we plan to report

  • After-hours and weekend conversation share (local business hours)
  • Median call duration and interquartile range
  • Human handoff rate where handoff events are instrumented
  • Lead capture and booking rates where those goals are enabled

What you can use today

While the benchmark ships, use our live proof assets: Hear It Work, the ROI calculator, and permissioned or illustrative case studies. For press enquiries, contact us.

FAQ

Is the benchmark dataset published yet?
No. Call-level aggregates are not yet exposed through a privacy-safe public view. This page documents the methodology only.
Will you publish individual call transcripts?
Never. Only aggregate metrics with minimum sample sizes and no customer-identifying fields.
What counts as a publishable stat?
Any metric must have at least 50 data points in the cohort, exclude demo/test traffic, and pass privacy review.
Where can I see Help Genie working today?
Try the live demo at /hear-it-work or browse industry genies at /discover.
How do I cite this page?
Cite the methodology URL and retrieval date. Do not cite placeholder metrics — none are published here.