Trust & security

AI Trust & Security — 12 safety dimensions

Built to be trusted. Scored to prove it.

Every Sales Agent conversation is evaluated across 12 dimensions by an independent AI audit system. Here’s what we found — and how we built safety in from the start.

91.2Avg AI audit score
+79%Improvement since first release
10.0DNC compliance (perfect)
7,000+Conversations audited
The rubric

How we score every conversation

An independent AI auditor evaluates 12 safety and quality dimensions — from compliance rules to message personalization.

DNC & opt-out10.0

Never contacts DNC-registered leads.

Contact boundaries10.0

Respects hours, channels, and tags.

Speed to lead9.4

Median first response under 5 seconds across channels.

Message personalization8.8

References lead name, property, and behavior.

Intent detection9.2

Correctly identifies buying and selling signals.

Appointment booking8.5

Creates appointments when a lead confirms a time.

Scope recognition9.0

Routes out-of-scope questions to the agent.

Escalation & handover8.9

Escalates when human involvement is needed.

Self-check timing8.7

Post-showing follow-up fires on schedule.

Outcome capture8.6

Reads call logs and updates lead status.

Follow-up sequencing8.4

Varies message cadence appropriately.

Overall engagement8.9

Lead engagement rate across all channels.

Scoring methodology. Scores are generated by an independent large language model that evaluates each conversation turn against a 12-dimension rubric. All personally identifiable information — names, phone numbers, addresses, emails — is anonymized before evaluation. No human bias. No cherry-picking. Every conversation in the dataset is included.

Track record

Quality that keeps getting better

8 rounds of independent evaluation over 16 months (Jan 2025 → May 2026). Every release scores higher than the last.

AI Quality Score — Improvement Over Time
025507510050.9v1Jan ’2568.3v2Apr ’2572.1v3Jul ’2576.4v4Sep ’2580.2v5Nov ’2582.7v6Jan ’2685.1v7Mar ’2691.2v8 (latest)May ’26
First Release50.9

Early deployment. Timing gaps, message repetition, inconsistent scope handling.

+79% improvement
Latest Release91.2

Consistent compliance. Sharper personalization. Reliable outcome capture.

Score distribution

How quality distributes across all conversations

95–10012%Exceptional
90–9428%Excellent
85–8934%Strong
80–8418%Good
Below 808%Needs review
By design

Safety is not a feature. It’s the architecture.

Every guardrail is built into the core decision engine — not added as an afterthought.

Zero-tolerance DNC enforcement

Before any outreach, Sales Agent checks for AI:DNC, Y_DNC_REGISTRY_TRUE, AI:Wrong Number, and opt-out signals. If any tag is present, outreach is suppressed — no exceptions, no overrides.

10.0 / 10.0 — perfect compliance across all 7,000+ audited conversations

Human handover state machine

Sales Agent recognizes 6 escalation triggers: financial questions, legal concerns, heated tone, direct agent request, out-of-area leads, and schedule conflicts. On trigger: task created, agent notified, outreach paused.

8.9 avg — escalation accuracy

Ready to Accelerate your Business Growth?

Complete the form to schedule a demo and one of our solution specialists will reach out to confirm your appointment.

We’ll make sure you receive -

  • A personalized tour and guidance from one of our top solutions advisors
  • Guaranteed lowest price on new subscriptions and add-ons
  • Instant access to your new Lofty account upon sign-up
Schedule a demo with a Lofty solutions specialist