LegacyCore + Retell AI Integration

The problem

Final-expense insurance closes on the first call or it does not close at all. Traditional staffing models put a human agent on the line for every conversation — expensive, inconsistent, and hard to scale because the same agent who is great at opening rapport is not necessarily the one who is great at the verbal close. Worse, the warm-transfer model — qualifier hands the lead to a closer mid-call — is where the largest single block of leakage in the entire industry happens. Half the transferred calls drop, and the deal evaporates.

How the integration works

Retell AI runs the AI Closer — a single voice agent that takes the call from the opening hello through the verbal close, with no transfer. The Retell agent loads the lead context (carrier waterfall tier, household composition, prior contact history) from Supabase before the dial happens, then runs the call against a script tree authored to handle every common branch: tier pivots when a Tier 1 carrier knockout fires, callback scheduling when a verification code does not arrive within three minutes, and the post-close hold bridge that keeps the client on the line while NemoClaw submits the e-app.

Mid-conversation webhook injection is the single most useful feature we use in Retell. When the LegacyCore backend detects a 2FA failure on the carrier portal mid-call, it pushes an instruction into the Retell agent’s context — the AI then says, in its own natural voice, "Let me have you double-check that code, sometimes it takes a moment to arrive." The client never hears a transfer, never hears a beep, never realizes anything technical happened. This single feature is why the integration works.

Calls have no hard duration cap — long calls are frequently the closing calls. When the Closer is still progressing toward verbal yes, the call runs as long as it needs to. The TCPA disclosure window is honored within ten seconds of call start (now including explicit AI identification per Texas TRAIGA), and the AI Closer is never permitted to claim it is human.

Why Retell AI specifically

Retell ships exactly the primitives a closer needs that other voice agent platforms do not — mid-call webhook injection, fast first-token latency on warm sessions, and a transcript event stream that lets us observe every utterance in real time. The CDP-style live view also lets a human supervisor whisper-coach if the call goes off script, although in production that is rare. Pricing scales with talk time, which aligns the vendor’s incentives with ours.

Read the docs

Full Retell AI API documentation lives at docs.retellai.com. The agent configuration model and webhook contract we use are both documented there.