Transparency Report

How We Measure Promtitude

A transparent, single source of truth for our performance claims, measurement methodology, and beta metrics.

Key Points

  • In beta: numbers may change as traffic grows
  • Labels: "target" = goal, "in beta" = observed last 30 days, "preview" = early estimate
  • Method: p95 latency, rolling 30-day windows, tenant-isolated data, no scraping

Quick Stats

Search Latency (p95)

0.80s

in betarolling 30d

Indexing Time per Resume

~1.8s

in betarolling 30d

Parse Success Rate

98.5%

in betarolling 30d

Active Recruiters

127

in betacurrent cohort

Resumes Searchable

45,892

in betatenant-scoped

Uptime

99.9%

targetlast 90d

* All figures exclude synthetic tests and internal traffic. See "Measurement Rules" below.

What We Measure (and Why)

Performance

Search latency (p50/p95)

Time from query received → top-k returned

Why: determines perceived speed

Indexing latency

Upload → searchable (median/p95)

Why: measures time-to-value for new data

API error rate

4xx/5xx % per endpoint

Why: reliability indicator

Quality

No-result rate

% of queries returning 0 candidates

Lower is better

Shortlist rate

% of searches that add ≥1 candidate to shortlist

Proxy for relevance/utility

AI↔Human score correlation

Pearson r between interviewer and AI scores

Per skill dimension

Search Latency Trend (30 days)

Day 1Day 30

Shortlist Rate % (30 days)

Day 1Day 30

How We Calculate

Measurement Rules

  • • Windowing: Rolling 30-day values unless stated; uptime uses 90-day
  • • Percentiles: Show p50 & p95 for latency; avoid averages for skewed data
  • • Scope: Exclude internal tenants, load tests, and bots
  • • Timezone: UTC for aggregation; visitor\'s local time for charts
  • • Weighting: Org-level deduping to prevent one power user from skewing metrics

Benchmarks Behind Claims

Scenario Set

Scripted tasks with 15 recruiters on identical 5k-resume corpus:

1

Find a senior Python + AWS candidate from corpus

2

Paste JD and shortlist top 5 matches

3

Interview (15-min segment) and produce scorecard

Baseline vs Promtitude

Baseline

ATS keyword search + manual scanning

12 min avg

Promtitude

Natural language + JD match + Copilot

3.5 min avg

Reported metric: time-to-first-shortlist (median)

Limitations

  • • Small sample size (15 recruiters)
  • • Tech roles only
  • • Identical resume corpus
  • • No outbound sourcing measured
  • • Results marked "target" until monthly study

Data Ethics & Privacy

Tenant Isolated

Analytics are tenant-isolated with no cross-customer model training

PII Protection

PII redaction for transcripts with configurable TTL policies

Your Data, Your Control

Opt-out available for analytics at tenant level

Important: We do not train shared foundation models on customer data. Your resumes and interview data are never used to improve models for other customers.

Changelog

2025-08-08

Published Metrics page; added quick-stats; defined p95/p50 rules

2025-08-01

Beta launch; initial metrics tracking enabled

2025-07-15

Added AI↔Human score correlation metrics

Frequently Asked Questions

Where do the numbers come from?

Production logs, vector DB metrics, and OpenAI usage reports (aggregated in our warehouse).

Why p95 instead of average?

Latency is skewed; p95 reflects worst-case user experience that 95% of users experience.

Do uploads train your models?

No. We don't train shared foundation models on customer data. Your data stays yours.

View all FAQs

See Promtitude in Action

Experience the speed and accuracy we measure. Join our beta to help shape these metrics.

Join BetaView Features