A read of PEXA's support operation today, what you're asking an AI agent to solve, and where Lorikeet fits. Built from your team's framing across the conversations we've had.
PEXA's support team handles a regulated, voice-led operation for ~50,000 lawyers, conveyancers, and financial institutions across all Australian states. The team is small, in-house, and high-trust — but the volume and complexity push against the limits of a manual operating model.
Six things that make this more than a nice-to-have for PEXA — the reasons an AI agent is on the roadmap, not just on a list.
Members call because the platform is hard, not because the FAQ is missing. ~850 paths through Exchange × state-by-state quirks means a knowledge-base chatbot doesn't move the needle. Voice has to actually resolve, with workspace context built in.
25,000 help centre visits a month, but limited visibility into how many of those visitors end up calling anyway. The self-serve to phone path is invisible — the team needs digital-first containment that actually holds.
Manual QA today reviews a small sample of calls per agent. Coverage is partial, theme detection is slow, and coaching sits on team-leader bandwidth. The team wants visibility into every conversation for sentiment, theme, and quality drift.
The bar is one consistent AI agent across voice, chat, and any future channel — same workflows, same guardrails, same compliance posture. Native AI add-ons in existing platforms are on the comparison list precisely because point tools would sprawl.
Regulated industry. No legal advice, no duty quotes, no PII echo. Redaction has to be built into audit logs and stored artifacts, not bolted on. Strict line between platform support and practitioner advice.
The voice POC is on the product roadmap with budget allocated. This is an evaluation, not an exploratory exercise — PEXA is comparing real options against a real spend decision.
Listed in priority order.
Six things we bring to this specifically.
The Concierge agent runs identically on voice, chat, email, and SMS. Workflows authored once, guardrails enforced everywhere — same workspace lookup, same compliance posture, same redaction across every channel.
No legal/duty advice, no PII echo, no inventing state-specific rules — enforced at the platform layer, not in prompts. When a member asks for a duty figure, the agent declines and points at the state Revenue Office.
A second agent grades every ticket against your criteria, surfaces themes, and runs simulations against persona variations before a workflow change ships. Replaces the small-sample manual QA cycle.
Read and write directly into Salesforce Service Cloud. Inbound and outbound voice through Genesys Cloud. Identity verification via the same security questions you use today — no second auth flow for members.
Account numbers, BSBs, and third-party representative identifiers are redacted from agent replies and from stored transcripts and audit logs. The agent doesn't echo, doesn't store.
Not tied to a proprietary model. We use the best LLM for each step in the workflow and migrate as the frontier moves. PEXA keeps the choice.