How it plugs in

Integration overview

No rip-and-replace. Lorikeet runs on your existing Twilio telephony as an intelligent IVR, takes a seat alongside your team in Zendesk, and is grounded in your public knowledge base. Both core connections are clean for us.

The picture

Architecture at a glance

Three layers: your existing channels on the left, Lorikeet in the middle, your systems of record on the right. Zendesk stays as the servicing system of record. Nothing leaves your data perimeter, and your team keeps human-in-the-loop control.

Channel layer

Voice — intelligent IVR (new)
Your Twilio number routes to Lorikeet instead of the five-option menu. Natural-language front door that solves or routes from the first sentence.
Chat (existing trial)
Your current AI messaging agent keeps running. The same Lorikeet brain can take it over later for one consistent experience across channels.
Email (future phase)
When you are ready, the same workflows extend to adviser email without a second build.

Lorikeet

Concierge agent
The adviser-facing AI. Operating procedures, knowledge base, guardrails, and tool calls. Identical behaviour across voice and chat.
Coach agent
QA and analytics. Reads every conversation, surfaces themes and drift, and runs simulations on workflow changes before they ship.
Guardrails + audit
No financial advice. Message checks for distress and fraud signals. Tamper-evident audit log on every action.

System layer

Zendesk (system of record)
Native integration. Tickets, conversations, contacts, macros. Lorikeet sits alongside your team, not in place of it.
Twilio (telephony)
Existing partner of ours. Inbound number, IVR routing, and call control run on the telephony you already operate.
Public KB · platform data
Public knowledge base grounds the agent today. Authenticated account and application data is a later, scoped phase.

Integrations by system

Each system mapped to its access pattern, ownership split, and whether it is ready for the first phase or scoped for later. We flag the open questions honestly rather than assume.

Twilio (telephony)

Partner

Twilio is an existing partner of ours, so the voice path is clean. Your inbound number points at Lorikeet, the five-option menu becomes an intelligent assistant, and complex calls route or hand off to the right team, on the telephony you already run.

Inbound: Answer the call, understand intent in natural language, solve or route
Routing: Mirror your existing five-option map; warm handoff to the right human queue
Setup: Repoint the number; no change to your underlying telephony estate
Lorikeet
Builds the intelligent IVR, the voice experience, and the routing logic against your call map.
Netwealth
Confirms the five current options and provisions the Twilio number and access.

Zendesk

Native

System of record for tickets, conversations, and contacts. Lorikeet operates inside Zendesk, taking a seat alongside your servicing team. Your existing inbox, audit trail, and human-in-the-loop tools stay exactly as they are.

Read: Contact, ticket history, custom fields, prior context
Write: Reply, internal note, ticket status, tags, assignment, escalation to a human agent
Handoff: Passes full context to the right queue so the agent hits the ground running
Lorikeet
Builds and maintains the Zendesk integration, the agent's actions, and the handoff behaviour.
Netwealth
Provisions Zendesk access and confirms the queue and routing model.

Public knowledge base

Scrape · read

Most of your support content is publicly available. We scrape it as part of our process and use it to ground the agent, so the demo answers "how do I open an account" and common how-to questions from your real material, not invented copy.

Read: Account opening steps, application how-to, product and platform guides
Refresh: Re-scraped on a schedule so answers track your published content
Private KB: Your internal articles can be added in a later phase under access controls
Lorikeet
Scrapes, structures, and keeps the knowledge base current; tunes answers to your terminology.
Netwealth
Points us at the canonical sources and flags anything that should not be surfaced.

Platform & account data

Scoping

Phase one is tier-0 and unauthenticated how-to, so it does not need account data. To answer adviser-specific or account-specific questions later, the agent would read from your platform. You flagged that surfacing this cleanly is a work in progress internally, so we scope it together rather than assume.

Open question: What account and application data can be exposed, and through what API?
Read (target): Adviser and account context, application status, common error states
Write (target): Lodge the action or task, record the outcome back in your systems
Lorikeet
Implements the integration once access and scope are agreed; sequences it after the tier-0 launch.
Netwealth
Confirms the data surface, identity model, and which authenticated actions are in scope.

Compliance & audit

Built in

Built for regulated financial services. Guardrails stop the agent giving financial advice or saying anything it shouldn't. Message checks route signals of financial distress or potential fraud straight to a human. Every action is logged for your risk and legal teams.

Guardrails: Hard rule against financial advice; decline and redirect, escalate if pressed
Message checks: Distress and fraud signals trigger immediate handoff to the right team
Audit: Full, reviewable trail of every conversation and action
Lorikeet
Ships the off-the-shelf controls and tunes them to your policies; presents them to your risk team.
Netwealth
Confirms the topics that must escalate and what risk and legal need to sign off on.

Existing AI messaging agent

Coexist

The messaging trial you launched in the last couple of months keeps running. Voice is the bigger near-term opportunity, so we start there. When it makes sense, the same Lorikeet brain can take over chat so advisers get one consistent experience across channels.

Now: Voice-first; the messaging trial continues untouched
Later: Consolidate chat onto Lorikeet for shared workflows, guardrails, and reporting
Benefit: One set of operating procedures behind every channel, not one per tool
Lorikeet
Runs voice now; ready to absorb chat when you choose to consolidate.
Netwealth
Decides if and when to move the messaging experience onto Lorikeet.
Worked examples

What a real call touches

Three inbound situations, showing every touchpoint end to end. These are the kinds of calls the intelligent IVR handles natively, with a clean handoff the moment a call needs a person.

Step How do I open an account Application error troubleshooting Investment question → human
Inbound Adviser calls the main number during an EOFY spike Adviser calls mid-application, hitting an error Adviser asks whether an asset is on platform / a returns question
Intelligent IVR "What can I help with?" — recognises an account-opening request Recognises a troubleshooting request, skips the menu Recognises the topic is moving toward investment specifics
Knowledge Pulls account-opening steps from the public knowledge base Matches the error to the known fix and walks through it Holds the line; does not attempt an advice-adjacent answer
Guardrails Stays on process; no advice on product suitability Stays on the documented fix; no workarounds it cannot stand behind Hard rule on financial advice; collects context for the handoff
Action Answers in full; logs the call and outcome to Zendesk Resolves the common case before it reaches a person Warm handoff to the right human queue with context attached
Result Adviser is unblocked without waiting in a phone queue Simple error cleared at tier-0; human time saved The person picks up already briefed, not starting cold