An executive summary of Lorikeet's response to TAL's External AI Member Concierge RFP.
TAL has been protecting Australians for 150 years across individual, group, and superannuation cover. Today TAL's members interact at the moments that matter most — lodging a claim after illness, injury, or loss; checking the claim that's holding up their financial security; or applying for cover to protect their family. Many of those interactions still run through phone, email, and manual processes; members wait, repeat themselves, and navigate complexity when they need simplicity and empathy.
TAL is looking for a no-code agentic AI platform to deliver a member-facing concierge across digital properties at enterprise scale — covering claims lodgement and check-ins on long-running claims, application and onboarding, policy servicing, and general member support — with full compliance for the regulated Australian insurance industry.
The intent is not to remove the human element. It is to make every interaction faster, more informed, and available when the member needs it.
Lorikeet is an Australian AI customer interaction platform built for complex, regulated businesses. We were founded by Steve Hind - formerly product at Stripe and Watershed - and Jamie Hall - formerly a named author on Google's Meena and LaMDA papers, the world's foundational research on conversational AI.
Since 2023, Lorikeet has been in production with AI concierges for both Australian and global regulated businesses, predominantly in finance and healthcare verticals. Our core product offering is a conversational agent that works across channels (chat, voice, email, and SMS) to facilitate member experiences throughout the member lifecycle - from application assistance, to account support, through to claims management, and proactive re-engagement.
Lorikeet is purpose built for the core intent reflected in TAL's requirements, apparent across 5 pillars:
We describe our capabilities across these 5 pillars in the cards that follow.
The default success metric in AI customer support is containment — tickets resolved without a human. For low-stakes commerce, that's a fair proxy for value. For TAL members, who interact with TAL at moments of injury, illness, bereavement, and major life decisions, it is the wrong metric. A member deflected from the claim they need help with is not a saved ticket. They are a complaint, a churn risk, and a compliance event in the making.
Lorikeet was founded on a disagreement with that standard. The concierge serves the full member journey — multi-turn, context-aware, transactional, across every channel, with a human always one step away. The metric we optimise for is end-to-end resolution. Containment is a side effect of doing that well.
The concierge transacts on members' behalf — lodging claims, validating against policy, capturing supporting documents, completing identity verification — live during the conversation, never bouncing the member to a separate page. Human handoff is native: the human inherits the full transcript, reasoning trail, collected data, tool results, and risk flags. The conversation can return to AI when the moment passes.
End-to-end resolution at concierge quality is what Lorikeet was built to deliver. Every other choice in the platform flows from that one.
End-to-end resolution sounds clinical. In life insurance, resolution touches conversations about illness, injury, bereavement, and the decisions families make to protect each other. Lorikeet is built to use precisely the right amount of AI, with precisely the right tone, to meet members at those moments.
Sensitive intents are not handled by prompt instructions alone. Each routes into a dedicated sub-workflow whose pacing, language, transactional access, and escalation thresholds are configured independently: a bereavement sub-workflow can withhold transactional prompts; a financial-hardship sub-workflow can lower the sentiment-escalation threshold. Sentiment is scored per message against an absolute threshold and a trajectory, so a drift toward distress across three turns fires before any single message would. The vulnerable-customer flag is first-class structured state with TAL-configurable triggers, audit-logged at every decision. AFCA-bound topics and vulnerable-customer indicators hard-route to designated human teams; the handoff payload includes transcript, reasoning trail, slot values, tool results, citations, sentiment trajectory, and the trigger that fired.
Long-running claims persist state across sessions and channels. The concierge can trigger outbound — SMS, voice, or email — for a wellness check-in, a missing-documents follow-up, or a rehabilitation-progress prompt. Claim type and status are typed state (TPD, income protection, terminal illness, death; in-flight, awaiting documentation, in assessment, paid, declined), available to every sub-agent without re-asking. Status enquiries — where is my claim, what is the next step, is anything outstanding — are answered against that state with live system data.
Life insurance is not generic support. The pacing, the routing, the escalation thresholds, and the moments we hand the member to a human are all configurable in Lorikeet's no-code platform.
The no-code platform that makes Pillar 2 work did not come together by accident. It sits on top of integrated infrastructure that is hard to assemble piece by piece, and a specialist team that is hard to staff.
Each major capability — Router, sub-agent runtime, knowledge retrieval, tool execution, guardrails, voice pipeline — scales independently. LLM inference runs across multiple foundation model providers with automatic failover, and every invocation is logged with the specific model used. Layered on top is a regulated-interaction defence: knowledge grounding for factual answers, deterministic Choice nodes for compliance-mandated paths, configurable guardrails with five actions, and Coach scoring 100% of tickets against TAL's quality rubric. A simulation framework runs prompt variants against synthetic scenarios before changes reach members. Rollback executes in under five minutes. The audit trail is immutable and queryable — regulator-grade for CPS 230, CPS 234, and ASIC investigation.
Each of those components is a multi-team engineering effort. The stitched-together alternative — foundation model from one vendor, voice from another, RAG from a third, guardrails from a fourth, observability from a fifth — leaves TAL holding the integration, upgrade, and regulator-readiness risk. Lorikeet has built and integrated all of them. TAL keeps its proprietary logic; Lorikeet runs the experience layer. New connectors land at speed; a new sub-agent can go live as soon as TAL's business logic is ready; and improvements made for one regulated subscriber flow to the rest.
Lorikeet is the runtime. The integration work underneath is the buy decision.
Buying the runtime is the right decision, but a platform integrated against TAL's real systems is a different artefact from a platform demonstrated in a sandbox. A platform that maintains TAL's quality bar over time is different again. AI customer interaction is not set-and-forget: models change, member language drifts, regulatory positions shift, edge cases emerge that no simulation surfaced, and the quality work that began at launch never stops.
Lorikeet's delivery model puts a forward-deployed AI engineer and a forward-deployed product manager alongside TAL's team for the duration of your subscription. The PM works with TAL business owners on workflow authoring, knowledge structure, escalation policy, and quality calibration. The engineer partners with your teams on each integration, authentication pattern, and connection. The global SaaS pattern — kickoff, configuration handed to the customer in a black box, vendor disappears — is not the pattern here.
Australia is the operating environment, not just a hosting decision. Lorikeet's product and engineering organisation work from Sydney.
TAL will be working with a team in its own city, in its own time zone, against its own regulator framework.
Working with a regulated Australian life insurer is not a customisation — it is the centre of what Lorikeet is built for. Governance, audit, and compliance posture are decisions Lorikeet made before TAL was a customer, not features added when TAL asked.
Every member-facing decision is logged at the granularity APRA, ASIC, and TAL's internal governance teams will look for: every routing decision with confidence score, every workflow node executed, every tool call with retry counts, every guardrail evaluation, every knowledge passage retrieved, every model invocation tagged with the specific model used. The audit trail is immutable, queryable, and exportable — built to support APRA CPS 230 outsourcing reviews, CPS 234 audit obligations, ASIC investigation requests, AFCA complaint legal-hold workflows, and TAL's incident response and quality forensics.
Certifications are current and mapped to the obligations TAL carries: SOC 2 Type II (Oct 2025), ISO 27001:2022 (surveillance Jul 2025), HIPAA (Mar 2025), GDPR, and a penetration test completed Aug 2025. ISO 27001 and SOC 2 controls map directly to APRA CPS 234. The Business Continuity and Disaster Recovery Plan, incident response, and operational evidence support TAL's CPS 230 material outsourcing obligations, and the engagement will be formally assessed as a material service provider arrangement under CPS 230 where TAL determines the threshold applies. Tenant data lives in Lorikeet's VPC in the Australia region, encrypted at rest with AES-256, with AU egress IPs available for firewalling.
Lorikeet doesn't ask you to figure out your governance requirements, then bring them to our platform. Your governance requirements are the foundation of our platform.
Customers we've already done this for at scale.
AFSL-licensed Australian payments platform. Live with Zendesk integration 11 days from contract signature in January 2025.
47× volume increase with no service degradation · 55% of regulated KYC / compliance queries fully resolved by AI · 17-second median response (down from 2.9 hours).
Rent payment platform. 12+ months in production. Agent operates across 40+ customer attributes with real-time API calls into Flex's backend.
2× CSAT improvement vs prior AI tool · 50% reduction in median conversation duration · 4× peak volume spikes absorbed during rent cycles.
US fintech with 1M+ customers — credit-builder loans, secured Visa cards. Lorikeet's voice agent runs 24/7 across Self's full inbound voice estate.
~75,000 calls/month at 99% accuracy and 95% performance.
Regulated AU health platform handling sensitive member interactions across a fast-scaling subscriber base.
3× ticket volume growth absorbed while CSAT lifted 10 points.
Australian unicorn creative platform serving millions of paid customers globally.
90% CSAT from paid customer interactions · 90% of subscription and billing reviews independently resolved by AI.
Selected Lorikeet following an extensive head-to-head evaluation on live calls versus a conversational AI competitor.
Won on claims management phone call accuracy — correct intent classification, accurate policy compliance, reliable tool calling.
Lorikeet's pricing is built around one principle: we only succeed when TAL's members are served. The commercial model reflects that directly.
No platform licence fees · no implementation fees · no setup costs · no per-seat costs for Builder users · no LLM inference pass-through · no professional services day rates.
Every dollar TAL pays Lorikeet maps to a member who was successfully served. That is the entire commercial alignment.