Products & Tools

Australia's loan AI talks. Humans still make the decisions.

Five Australian loan AI tools can speed up research and support, but ASIC and lender data show that decisions still rest with rules, brokers and accountable humans.

Five AI tools now inside the loan process, and what each still can't do
Five AI tools now inside the loan process, and what each still can't doAI-generated
By Elena Vasquez6 min read

Why it matters

  • ASIC's Report 798, published in October 2024, covered 23 licensees and 624 AI use cases.
  • On November 14, 2026, eight of 10 lender, broker and lending-technology sites reviewed had no visitor-facing chat assistant.
  • Lendi Group's policy agent draws on about 10,000 pages and was being extended to all 1,300 brokers.
  • Experian research found 72% of 102 Australian lending decision makers used agentic AI; 2% supported fully autonomous decisions at scale.
  • Separate surveys found 6% of 1,500 Australians were most comfortable with an AI assistant, versus 32% for a mortgage broker; 59% of 481 credit-active Australians were comfortable with an application-handling AI agent.

By Elena Vasquez

On November 14, 2026, a review of 10 Australian lender, broker and lending-technology websites found that eight exposed no visitor-facing chat assistant.

One bot asked for a name, email and phone number before offering help. One remaining site answered a typed question.

That narrow result matters. AI has entered the front of the loan process, but customer-facing use still trails the headlines. Across the five tools examined, software reads, sorts, answers and drafts. Rules or people retain authority over eligibility, advice and approval.

The November 2026 review describes a common design. Each product separates conversation from consequence. The software can guide, while a rule set or person controls the result.

What did ASIC find?

In October 2024, ASIC published Report 798, its first review of AI in financial services. The report covered 23 licensees and 624 AI use cases, including systems in use or development as of December 2023.

ASIC found adoption relatively cautious. Most systems supported human decisions or improved efficiency. About six in ten licensees planned to do more.

Two years later, that pattern still runs through public loan products, where conversational software handles information while people retain the decisive role.

What can the five public loan tools actually do?

The five products cover business finance and mortgages. Each occupies a different point in the process. None presents itself as the final decision-maker.

Switchboard AI: business-finance triage

The November 2026 commentary identifies Switchboard AI as its own product and asks readers to weigh that section accordingly.

Switchboard AI by Switchboard Finance asks a business owner what is going on and shows where they stand for business finance before it asks who they are.

It cannot approve or decline a loan. Eligibility comes from a fixed check, not the language model’s opinion. A licensed broker reviews every scenario before anything reaches a lender.

Arthr: a mortgage broker with an assistant

Arthr calls itself “Australia's first AI native mortgage broker”. A borrower starts by text or voice. By the company’s own description, a human broker then picks up.

The assistant opens the conversation. It does not finish it.

Lendi Group: policy help for brokers

Lendi Group told IT Brief Australia in August that its policy agent, a manager and 28 agents draw on about 10,000 pages of lender policy. The company was extending the system to all of its 1,300 brokers.

The agent answers questions such as how a lender treats casual or overseas income. It links to the source documents.

Lendi said anything regulatory, or anything that has to be certain, runs through deterministic processes. The AI presents options before handing over to a broker.

Yellow Brick Road: a platform behind the broker

Yellow Brick Road unveiled Nectaria in September. Australian Broker reported that it records client calls and maps the conversation against the customer’s profile.

The platform flags compliance considerations. It creates follow-up tasks for missing information and lets brokers search policy across the lender panel.

It works for the broker, who still makes the calls. The borrower may never see it.

TomoBroker: a white-label assistant for broker websites

TomoBroker sells AI tools to mortgage brokers by invitation. Its lineup includes a chat assistant that sits on a broker’s own website.

On TomoBroker’s published governance list, the model never writes the figures. Fixed calculations produce the numbers. A filter removes advice wording. Every model call is logged, so a conversation can be replayed.

Five builders make five different products. They keep the same line in the same place: the language model stays away from the number and the yes.

Why have lenders kept people in charge?

Experian research, reported in August, surveyed 102 senior decision makers in Australian lending. Seventy-two per cent already used agentic AI to support underwriters.

About half said they were comfortable letting AI decide without human review only in low-risk cases, a figure of 51 per cent. Just 2 per cent said they were comfortable with fully autonomous decisions at scale.

Those figures describe support, not surrender. Tools can accelerate work that follows a fixed policy. They do not carry the same licence or answer for the same outcome.

The commentary’s case for caution is practical. AI can read a bank statement in seconds. It still cannot assess a business it has never seen.

A quarter that looks like a collapse may signal a planned shutdown for a refit. A director’s guarantee is a judgment about a person. A self-employed borrower’s file contains facts that are true and still need explaining.

The result is speed at the front of the process. It is not displaced judgment.

How comfortable are borrowers with loan AI?

In mid-2025, Agile Market Intelligence asked 1,500 Australians which route they would be most comfortable with when researching a home loan. Six per cent chose an AI assistant, against 32 per cent for a mortgage broker.

In July 2026, Forrester surveyed 481 credit-active Australians for Experian. It found 59 per cent comfortable with an AI agent handling a credit application for them.

The two results do not measure the same choice. One tested research preferences; the other tested application handling. The samples, questions and dates also differ.

Together, they show comfort moving. It has not settled.

What should a borrower ask before trusting a loan assistant?

Borrowers can test three things before they trust a loan assistant: the scope of its authority, the human oversight behind it and the destination of their data.

  • What does it decide, and what does it only explain? A tool that explains can be wrong politely. A tool that decides needs a licence holder standing behind it.
  • Who checks it? Look for a person, a rules engine, or both. Look for a record of what was said.
  • Where does what you type go? Ask where it is sent, how long it is kept, and who can read it. A tool worth trusting should answer in plain terms.

What has to change next?

The commentary labels itself general commentary from the lending side. It says product descriptions come from company statements and the named trade reports as of November 2026. It is not advice on which tool or lender to use.

The tools show why product-level details matter. Lendi points brokers to source documents. TomoBroker logs calls for replay. Switchboard uses fixed eligibility rather than a model’s guess.

Those controls reduce ambiguity. They do not remove the need for a visible answer about who remains responsible.

The commentary’s closing test is blunt: “A good assistant gets a borrower to the right question faster, at an hour when no broker is awake.” The work that follows still belongs to a person.

As these tools reach more borrowers, each provider will have to make three things plain: what the model may decide, who checks its work, and where borrower information goes. The practical test is visible accountability, not conversational polish.

Source: itbrief.com.au

Share this article:

More from Elena Vasquez

Elena Vasquez

Show full bio

Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.

708 articles

Related articles

  1. AI Answers Now Rank Australia's Lenders — and Challengers Are Winning
  2. Nearly Half of US AI Users Admit Lying to Chatbots to Reach a Human
  3. Splendid Finance launches AI business loan finder covering 175 Australian lenders
  4. Ansarada pushes AI from documents to data with OS Bondi 3.0
  5. Uptiq launches GovernAI to govern AI agents in banking at no extra cost

« Previous articleNext article »