Contact us
Live demo

AI customer experience for Arlo Money

Money's a design problem dressed up as a tech problem. We built an AI support agent for Arlo — a fictional Australian bundled fintech — that handles failed direct debits, hardship requests, insurance claims, transaction disputes, and account access. End to end. Across chat and voice. With the empathy and tone a designed brand deserves.

Try the Arlo Assist agent

Give the agent any name, email, transaction reference, claim ID, or payment amount — it rolls with whatever you provide and returns realistic mock data for that identity. No login, no signup.

Click any prompt to copy it, then paste into the chat (or just call):

Make up any name, email, payment amount or transaction reference you like — the agent will roll with whatever you give it. The mock backend returns the same realistic shape regardless of input.

Call 0480 847 879

Same brain, voice or chat.

The agent uses Arlo's voice — warm, plain English, empathy-first on hardship. AU Emma voice. 0.9× pace. Spells emails back before any write action. Reads dollar amounts in whole numbers, never digit-by-digit. Calm under crisis: surfaces Lifeline 13 11 14, the National Debt Helpline, and warm-hands to a Care Team specialist when a customer is in real distress.

0480 847 879

Live demo line. AU number. Try the four scenarios above by phone.

Built for regulated financial services

Six things this agent does that a typical bank chatbot doesn't — and that the contact-centre teams behind Australia's financial brands keep telling us they need.

Hardship done with empathy first

The agent leads with one short, real acknowledgement — never with policy. Opens a hardship case under NCCP s72, pauses interest, books a Care Team callback. Surfaces the National Debt Helpline (1800 007 007) when the conversation needs more than a product can give.

Calm under crisis, on a regulated channel

Real STEER guardrail for acute distress and self-harm. Surfaces Lifeline 13 11 14 and Beyond Blue, warm-hands to a human, never says "Oh my god" or "Of course" in a crisis moment. The exact tone-failure mode most AI agents ship with — designed out from the first call.

Resolves, doesn't deflect

Retry the failed DD. Lodge the claim. File the dispute. Send the verification link. The agent IS the support team — never deflects to "call our 1300 number" or "log into the app." Every promise has a real tool wired behind it.

Voice that respects voice

One question per turn. Spells emails back letter-by-letter before any write. Reads dollar amounts in whole numbers. Doesn't read ABNs or transaction refs aloud. The pace, the silences, the tone — designed for ear, not for screen.

Grounded in real regulation, no fabrication

Every fact in the knowledge base is grounded in the actual Australian framework — NCCP s72 for hardship, ePayments Code for disputes, ASIC RG271 for IDR, AFCA for external review. No invented rates, no fabricated policies, no made-up clauses.

Designed experience at scale

Coach — Lorikeet's QA layer — scores every conversation against brand voice, empathy, compliance, and resolution. The experience you design once, measured the same way across millions of interactions. The feedback loop a CX designer at a bank can't get from human agents today.

How we built this

A working AI support agent for a fictional regulated fintech — built in one session in Sydney.

1

Grounded the KB in real regulation

Eight knowledge articles written against the actual Australian framework: NCCP s72 hardship, ePayments Code disputes, the Insurance Code of Practice, AFCA, IDCare. Every claim is cited and verifiable.

2

Built six topic-matched workflows

Failed direct debit, hardship support, claim FNOL, transaction dispute, account access, and an openConversation router. Each one channel-aware — voice rules differ from chat rules and they're written into the workflow.

3

Wired 11 mock tools, demo-permissive

Lookup credit, lookup cover, recent transactions, claim status. Plus the write actions: retry DD, lodge hardship, lodge claim, lodge dispute, send verification link, escalate. Every tool accepts any input and returns a realistic mock — no specific identity required.

4

Layered safety, then went live

13 brand guidelines and 5 STEER guardrails — including a high-priority crisis indicator that surfaces Lifeline and the National Debt Helpline. Chat widget + AU phone live the moment the workflows were published.

What's next?

From this demo to production in weeks, not months.

1

Train on your real data

Replace the demo customer data with your actual member records, product configs, and brand voice. Wire the agent into your core banking and policy systems — read-only to start.

2

Phase 1 — Read-only access

The agent answers questions and triages requests using lookups only. No write actions on production. Your contact-centre lead reviews the first cohort of conversations daily, signs off on tone, scope, and edge cases.

3

Phase 2 — Write-back actions

Retry DDs, lodge hardship cases, file disputes, send verification links — for real. Tight blast-radius controls, full audit trail, dollar limits, and automatic Care Team handover for anything outside the agent's scope.

4

Scale across channels

Voice and SMS layer on top of chat. Coach measures every conversation against the brand voice and compliance standard. Proactive outbound for adoption nudges and renewal reminders once the foundation is settled.

What this means for a regulated FS brand

The contact centre stops being the experience floor

Today the contact centre is where most banks' brand experience goes to die — outsourced, scripted, queue-times. AI doesn't just reduce cost there; it lets you design the experience for the first time at scale.

Hardship feels like care, not a process

Hardship volume goes up in every cycle. Doing it well — with empathy first and policy second — is what separates a regulated FS brand customers stay with from one they leave. The agent's been built to do exactly that, by default.

Designed once, delivered at scale

You spec brand voice, tone, escalation rules, and crisis handling in one place. Coach measures the actual delivery against the spec on every interaction. Drift gets surfaced before it becomes a CX incident.

Compliance posture built in, not bolted on

STEER guardrails for personal financial advice, insurance advice, off-product topics, crisis indicators. The agent stays on safe ground by design — no "as an AI" disclaimers, no clinical-adjacent overreach, no rate quotes from memory.

AU data residency, AU phone, AU regulation

Lorikeet hosts in AU for AU customers. The voice is AU Emma. The regulatory grounding is AU-specific (NCCP, ePayments Code, AFCA, ASIC). No quiet US-first surprises in your stack.

Built in days, not quarters

This Arlo Money demo — sandbox, 11 tools, 6 workflows, 8 KB articles, 5 guardrails, chat widget and AU voice — was built in a single Sydney session. A real-data POC for your subscriber is ~2–4 weeks.

Want one of these for your brand?

We'll build it for your contact-centre realities — your products, your policies, your tone — and walk you through it live.

Talk to the team