Cold outbound — the ethical framework

Nortinia Sales AI supports outbound voice calls via Telnyx. Up front: we do not cold-call. That is not a slogan — it is a technical limit of the system. This piece explains what the difference is, why it matters, and what four disclosures every call must include.

What "cold" calling is and why we do not do it

In the classic sense, cold calling means: a bought or scraped list, no prior contact, no opt-in, the recipient does not know who you are. In Hungary the GVH (Competition Authority) restricts this strictly; at the EU level the ePrivacy directive and the GDPR frame it. Beyond the legal minimum there is a practical problem too: these calls are bad. 0.8-1.5% pickup, 90%+ negative CSAT, and one in 30 calls generates a complaint or report.

Not worth it.

What permission-based outbound is

In our system, an outbound call can only go out if the recipient has at least one of:

1. Webform opt-in — explicitly ticked a box that they can be called
2. Content download — downloaded a document from us or the customer tenant, with phone number provided
3. Explicit sales-list confirmation — accepted, during a past sales-rep conversation, that they can be called back
4. Existing customer relationship — already a customer, with a meaningful reason for contact (account check-in, upgrade offer)

If none of these is present, the system will not put the number into the dialer. This is not tenant-configurable — hard limit.

The four mandatory disclosures

Every outbound call must, within the first 20 seconds, include:

1. Company identification — "calling from Nortinia / on behalf of [customer tenant]"
2. Purpose of the call — "about a specific offer" / "following up on a prior enquiry"
3. Opt-out option — "you can say at any point that we should not call again"
4. AI-assistance disclosure — if the call is AI-assisted (transcript / voice synthesis): "this conversation is being conducted with the help of an AI assistant"

The fourth is especially important and rarely required in the market. We include it anyway, because transparency is cheaper long-term than surprised customers.

How the system enforces it

The dialer lives in a state machine, and every call attempt generates an eng_voice_attempt row. Before start, a permission check runs on the lead. If no opt-in is found, the call ends with BLOCKED_NO_PERMISSION status before it rings.

The call itself is handled by the Telnyx Voice API; voice synthesis comes from OpenAI Realtime or ElevenLabs (tenant choice). The engine LLM answers the recipient's questions — but within strict script bounds. It can ask: "are you interested in the offer?", "does 3 PM today work?", "would you like to speak to a human rep?". It cannot close a deal, quote a price, or promise anything.

What happens to the call result

Every call ends with a structured call summary: pickup yes/no, conversation length, key points, lead next-step suggestion. This feeds back into lead scoring (engagement signal), and if the LLM detects an "interested in the offer" signal, the lead moves into a meeting-booking flow (Cal.com integration).

Outcome numbers

Six months in, ~12,000 calls:

• 18% pickup rate (vs. ~5% cold-calling industry baseline)
• 7% CSAT-positive (1-question SMS sent within 24h after the call)
• 0.3% formal complaints (vs. ~3% industry baseline)
• 0 GVH reports in six months

What we do not do

• We do not call bought lists
• We do not call on Sundays or after 19:00
• We do not mask the caller ID
• We do not repeat a call within 14 days after opt-out — forever blocked
• We do not call numbers flagged DNC (do-not-call) by enrichment

Two specific calls we are proud of

First: February 2026 — a customer asked us to make the fourth disclosure (AI assistant) optional. We declined. We would rather not ship the feature than ship it half-disclosed.

Second: April 2026 — voice synthesis temporarily became too human-like (a new ElevenLabs model), and recipients failed to notice they were talking to AI despite the disclosure. We dialled back one step so that the synthetic quality is audible. We do not boast about perfect-sounding voices — we boast about an honest comfort level.

The cold-call genre's bad reputation is not our fault. But if we contributed to it, it would be worse.