AI-powered support for Stake
An AI agent that resolves the tickets that need a human today — looking up the actual FX rate on your Apple trade, cancelling a Stake Black subscription, diagnosing a deposit that hasn't shown up, sending a password reset, explaining a dividend you're waiting on. Not deflection. Action. On both chat and voice.
Try the Stake AI agent
Open the chat and ask anything below. The agent looks up real data, explains the why, and actions what it can — without dumping you into a help article.
Click any prompt to copy it.
What the agent handles
Five action-led workflows plus a knowledge fallback. The differentiator vs Maven: every flow looks up real data and actions what it can, rather than handing the customer a help article.
Explain an FX rate / trade
Looks up the actual trade — fill price, USD total, base AUD/USD rate, Stake spread, effective rate, AUD charged. Explains the math, doesn't deflect to an FX article.
Cancel Stake Black
Confirms the plan, offers a light retention (annual switch), executes the cancellation, confirms period-end and downgrade dates. Real action — the subscription is actually cancelled.
Check a deposit
Looks up the deposit by amount / method (PayID, OSKO, BPay), tells you the actual status — pending normal with ETA, cleared, or failed with the specific reason and next step.
Account lockout — password or 2FA
Forgotten password? Reset link sent on the spot. Lost 2FA device? Logs the manual ID-verification case and routes to the Account/Identity team with full context. Skill-matched, not generic.
Dividend explainer
Looks up the dividend status for a holding — ex-date, record date, payment date, when Stake credits your wallet, US withholding for AU customers on US dividends, franking for ASX. Plain language, real timeline.
Stays in scope
Never gives investment advice or market predictions. Never quotes personal tax outcomes. Recognises off-product questions and deflects cleanly. The agent's job is resolution, not opining.
Voice channel — same agent, on a real phone line
Same workflows, same lookups, same actions — running on voice. Brooke voice, English STT tuned for Stake terminology (ASX tickers, PayID, W-8BEN, franking). The phone number's being provisioned; we'll attach it before your demo session.
How we built this
Wired up specifically for Stake's product and customer questions.
Modelled your real product
Mapped the actual things Stake customers ask about — trades and FX math, Stake Black, AU funding (PayID / OSKO / BPay), 2FA, dividends and franking.
Built the workflows
Six topic-matched workflows plus a router. Each one looks up the relevant data first, then explains or actions. Hard rules prevent deflection — "I don't have access to that" is banned.
Mocked your systems
Twelve mock tools — customer / trade / charge / deposit / dividend / subscription lookups, plus write actions for cancellation, password reset, 2FA reset, refund, and skill-matched escalation. Swap for your APIs in production.
Went live on chat and voice
Same agent, same brand, both channels. Identity-free — type whatever email or name in the chat, the agent rolls with it.
What's next
Four steps from demo to production.
Train on your real data
Point Lorikeet at the Stake KB you already have (Confluence, Maven's training set, help centre articles). The agent's grounded in your actual content, not approximations.
Phase 1 — read-only
Replace mock lookups with real API access. Agent answers FX, trade, deposit, dividend, and subscription questions using actual customer data. Zero write actions yet — pure deflection upgrade.
Phase 2 — write-back
Enable actions one at a time — cancel Stake Black, send password reset, initiate 2FA reset case, process small refunds. Confidence-gated rollout; the agent earns each action.
Orchestration layer
Skill-matched escalation to the right human team (Trading, Account/Identity, Compliance, Stake Black billing, Tax). Urgency triage for market-hours-sensitive cases. Coach reviews 100% of conversations.
What this means for your team
Built for regulated AU fintech. Identity-conscious, action-led, orchestration-ready.
AU data residency
Multi-region. Australian customer data stays in AU if you want. Pick the jurisdiction that fits your privacy posture.
Phased API permissions
Start read-only. Enable write-back per action as confidence builds. The agent only sees and actions what you've explicitly authorised.
No investment or tax advice
Brand guidelines prevent the agent from recommending what to buy, predicting markets, or quoting personal tax outcomes. Anything regulated routes to the right specialist team.
Skill-matched escalation
When the agent hands off, the human picks up with the full context — what was looked up, what was tried, what's outstanding. Trading / Account-Identity / Compliance / Stake Black billing / Tax routing built in.
Action-led, not deflection
The agent doesn't just say "here's the help article" — it cancels the subscription, sends the reset, explains the actual charge. That's the gap vs deflection-only tools.
Coach — 100% QA review
Every conversation is reviewed post-hoc. Patterns surface for your team — what the AI's resolving cleanly, where it's escalating, where the KB is thin. Self-improving loop.
Want to see this on your real data?
Happy to walk through how this lands alongside your current Maven setup, and what Phase 1 would look like with your APIs.
Talk to the team