Offline. Two weeks. Two iteration cycles. A small set of PEXA testers reviewing ~20 calls. By the end you have enough evidence to commit or walk away.
Three measurable outcomes. We baseline against your current state in week 1 and report against these at the exit review.
Of reviewed conversations rated PASS against PEXA's existing QA rubric. No regression vs a senior support agent on the same scenarios.
Of in-scope conversations resolved end-to-end without human handoff in the test set. Anchored to the two POC workflows.
Zero PII echoes, zero legal/duty advice incidents, zero unredacted artifacts. Guardrail tests run on every workflow change.
Not phased — concurrent. All three run in parallel inside the same 2-week window.
Workspace errors & lodgement, plus settlement & workspace status — the two highest-volume scenarios. Both stood up with mock tool wiring against the public PEXA playbook. KB seeded.
Same workflows on a Lorikeet AU voice number and the chat widget. Voice-specific tuning (chunked numerics, no URLs read aloud, settlement-day urgency) applied.
Coach scores every test conversation against PEXA's QA rubric. The grade plus the theme detection feed straight into the cycle 1 review.
Timeline: 2 weeks end-to-end. Offline only — no production traffic. Day 0–2: PEXA shares SOPs & QA rubric. Lorikeet builds. Day 3–6: Cycle 1 — share, test, feedback, iterate. Day 7–10: Cycle 2 — same shape. Day 11–14: Exit review and commercial decision.
Two cycles, four steps each. Same shape we ran with Airwallex's ATO pilot. Total: ~20 conversations for PEXA to review across both cycles.
Lorikeet ships the workflows to your sandbox and runs the full simulation suite. Pass rate + transcripts shared with PEXA.
Your testers run their own calls and chats against the sandbox — same surface members would see. No setup work on your side.
Structured feedback against the agreed rubric. Ten conversations per cycle, reviewed and scored. One short working session to walk through the failures.
Lorikeet incorporates the feedback overnight. Workflow + guardrail changes ship behind the simulation gate. Next cycle starts.
Three buckets. No granular per-line ownership. The point is fast and simple.