# AI Delivery Policy
**[BRAND_NAME] is an AI-first consulting firm.** Most of our services are delivered through autonomous AI agents with human quality review. This page explains exactly what that means and the safeguards we maintain.
## What we use AI for
- **Outbound prospecting:** AI personalizes cold emails using publicly available enrichment data
- **Voice qualification:** AI voice agents call inbound prospects and qualify fit
- **Discovery interviews:** AI voice agents conduct 20-30 minute structured interviews with you and your team during the audit phase
- **Deliverable generation:** AI synthesizes interview transcripts, uploaded artifacts, and public data into your AI Opportunity Map
- **Implementation:** Implementations we deploy for you (voice agents, chatbots, workflow automations) are themselves AI
## What we always disclose
- **AI identity:** Every AI call opens with: "I'm an AI assistant calling on behalf of [BRAND_NAME]." No exceptions.
- **Recording:** Every AI call discloses recording at the start. Recordings are retained 30 days unless engagement is active.
- **Opt-out:** You can stop or transfer to a human teammate at any time during any AI interaction.
## What humans do
- Quality review of every audit deliverable before it reaches you
- Strategy and judgment calls that AI is not yet reliable for
- Final scope and pricing on implementation engagements
- Direct response to escalations, complaints, and edge cases
- Compliance review on every engagement involving regulated data
## What AI does NOT do without human review
- Sign or modify legal agreements
- Make pricing decisions outside published menu
- Promise specific revenue or performance outcomes
- Process Protected Health Information (PHI) without a signed BAA and additional human gating
- Take destructive actions on your systems
## How we handle errors
AI agents may occasionally produce inaccurate information ("hallucinations"). When discovered:
- We correct the error and re-deliver any affected deliverable at no charge
- We flag the failure mode in our internal logs to improve prompts and pipelines
- For material errors that affect your decision making, we'll proactively notify you
## Compliance
- **FCC AI-voice rule (2024):** Mandatory AI disclosure on every call.
- **TCPA:** B2B landlines only on cold outbound dial. Cells and consumer numbers require prior express written consent.
- **A2P 10DLC:** Brand and campaign registered before any SMS.
- **CAN-SPAM:** Physical address and one-click unsubscribe on every commercial email.
- **GDPR / CCPA:** Data export and deletion endpoints in client portal.
- **HIPAA:** Engagements involving PHI require a signed BAA and use a separate, BAA-eligible LLM provider tier.
- **SOC 2:** We are pursuing SOC 2 Type II certification.
## Data handling
- All client data encrypted at rest (AES-256) and in transit (TLS 1.3).
- Per-client data isolation via row-level security.
- Zero credential handover required for any standard engagement. We integrate via OAuth-scoped tokens or service accounts you provision.
- No co-mingled training: your data is not used to train any model.
## Questions
[BRAND_EMAIL]