Real customer. Real policy. Zero manual intervention. Here’s how it happened.
On 31 October 2025, at 4:11 PM, an Australian startup quietly crossed a threshold that the broader AI industry had been talking about for years: an AI agent completed a private health insurance sale — end to end — without a single human stepping in to close it.
The company was Konkrd, a Melbourne-based digital broker founded by former executives of iSelect, one of Australia’s largest insurance comparison platforms. The policy was an Extras product issued by Australian Unity. And the significance, at least for anyone watching how AI is moving inside regulated industries, was hard to overstate.
This wasn’t a sandbox test. It wasn’t a demo environment with synthetic data. It was a real customer, making a real financial decision, guided entirely by an AI system operating inside Australia’s licensed broker compliance rules.
The User Journey, Step by Step
To understand what Konkrd has actually built, it helps to see it in action. A demo of the live platform shows a conversational AI flow that is structured, guided, and notably different from a generic chatbot.
The session opens with a simple question: “What brings you here today?” — with preset options including saving money on premiums, reducing out-of-pocket costs, or changing cover. The customer selects one.
From there, the AI works through a short but precisely sequenced needs assessment:
- Who needs to be covered? (Single, couple, family, single parent)
- What type of cover? (Hospital and Extras combined, Hospital only, Extras only)
- Age bracket of the oldest person on the cover
- State or territory
- Continuous hospital cover history — notably, the AI asks whether the customer has held continuous hospital cover since their 31st birthday or for the last 10 continuous years, a compliance-relevant question tied to Australian Lifetime Health Cover loading rules
Before proceeding to recommendations, the system summarises everything collected — motivation, policy type, product type, age bracket, state, and continuous cover status — and asks the customer to confirm or correct it. Nothing moves forward until this is verified.
The recommendation output is detailed and explicitly reasoned. For a single, 31–40 year old in VIC seeking combined Hospital and Extras cover, the AI shortlisted two policies with a plain-language explanation of the trade-off between them: one offering stronger extras value for dental and physio users, the other a balanced all-rounder for those prioritising dental with fewer surprises.
The final screen shows a side-by-side policy comparison across multiple funds — ahm, Queensland Country Health Fund, Medibank — with itemised benefit coverage and monthly pricing, plus a “Compare all Plans” option for deeper analysis.
The entire flow from first question to policy shortlist runs in under two minutes.
What Konkrd Actually Built
Konkrd’s core product is its Autonomous Sales Infrastructure (ASI) — a governed workflow layer that sits above the AI models and dictates how every customer interaction must proceed.
The architecture is deliberately structured into four separate layers:
- Data — policy documents, fund comparisons, product schemas
- Reasoning — AI agents that interpret and apply rules in context
- Orchestration — the workflow engine that enforces sequence and compliance
- Model choice — the underlying LLMs, swappable without breaking the rest
This separation matters. Most AI deployments bolt governance on top after the fact. Konkrd built compliance into the execution layer first, then added the AI on top of it.
In the October sale, the ASI did the following without human intervention:
- Gathered the customer’s needs through a structured conversation
- Interpreted and compared policy fine print across multiple funds
- Generated a coverage-based recommendation (not just cheapest price)
- Completed the application and bound the policy via the insurer’s API
How Compliance Is Actually Enforced
Scott Wilson, Co-Founder and Executive Chair of Konkrd, describes the compliance model as a hybrid — part rules-based, part model-driven.
The system starts with the same compliance frameworks used to train human brokers: required questions, advice boundaries, disclosure obligations. These are encoded into the system and used to score AI agent behaviour in real time. On top of that, a separate compliance agent monitors every live conversation — not sampled conversations, every single one.
“Every conversation is monitored. Not sampled. That is a step change from traditional broking,” Wilson notes.
When regulations or partner requirements change, the compliance framework is updated, version-controlled, and re-tested against historical conversations before going live again.
Three things can trigger a human escalation:
- Customer-initiated — the customer can request a human at any point
- AI-detected — the system flags confusion, ambiguity, or out-of-scope requests
- Compliance-triggered — any potential breach or edge case
When escalation happens, the full conversation history passes to a human agent in real time. No restart. No loss of context. The human can complete the interaction or hand back to the AI.
The Audit Trail Problem — Solved Differently
One of the persistent concerns with AI in regulated environments is auditability. What exactly did the system decide, and why?
Konkrd’s audit trail logs everything:
- Full conversation transcripts
- Questions asked and responses given
- Recommendations generated and the reasoning behind them
- Source references back to specific policy document clauses
- Compliance scores and flags at each step
- Final outcome and policy selection
“It is more transparent than a human broker model,” says Wilson. “Nothing is lost or summarised. It is all recorded.”
Access is tiered — internal compliance and risk teams have full visibility, partners access relevant transaction records where required, and all data handling is in line with privacy and regulatory requirements.
The Hard Problems Before Launch
Two things had to be solved before Konkrd could go live, and Wilson is direct about what they were.
First: the AI had to beat human compliance scores, not just match them. Before launch, Konkrd set a hard internal bar — the AI must meet or exceed the compliance performance of human brokers. That required building a dedicated compliance agent layer, testing against real-world scenarios rather than synthetic ones, and running continuous scoring and replay tests.
Second: insurance data is a mess. Policy disclosure statements run 10–20 pages. Exclusions, waiting periods, and benefit limits are buried in inconsistent language across hundreds of fund documents. Konkrd had to ingest and normalise thousands of product documents, build a consistent policy schema across funds, and develop agents that reason over the data — not just retrieve keywords from it.
The ASI architecture — with its separation of data, reasoning, orchestration, and model — is the direct result of solving these two problems at scale.
What This Means for AI in Regulated Industries
The broader AI conversation tends to treat “agentic AI” as a future state — something that’s coming. Konkrd’s October sale is a data point that it’s already here, at least in specific, well-scoped domains.
The key architectural insight from their build is not the AI itself. It’s the sequence: workflow first, model second.
As Wilson puts it: “Most teams start with the model. The right starting point is the operating model. If the workflow is not compliant, the AI will not be either.”
For industries like financial services, insurance, healthcare, and legal — where compliance isn’t optional — this matters. Konkrd’s approach demonstrates that autonomy and governance aren’t in tension. They can be designed together, from the start, if the architecture separates concerns cleanly enough.
What’s Next for Konkrd
Health insurance is the first vertical. The broader roadmap covers energy, car, home, telco, travel, pet, and eventually mortgages, life insurance, and superannuation — essentially, the full landscape of complex household financial decisions that Australians routinely underpay attention to or get wrong.
Konkrd is also building a secure digital wallet where users can store all their policy documents and get AI-powered, plain-language answers about their cover.
The positioning is deliberate: not another comparison site (price-optimised, commission-driven), but a neutral, agentic, multi-provider digital broker — one that acts in the consumer’s interest across the full lifecycle of their financial products.
Konkrd is headquartered in Melbourne, Australia. For more information, visit konkrd.com.
Interview responses attributed to Scott Wilson, Co-Founder and Executive Chair, Konkrd.
