The State of AI in Credit.

The primary evidence used to form this report were conversations with over 30 Australian credit funds. We spoke with managing directors, analysts, CIOs and more at firms ranging from approximately A$100 million to over A$50 billion in FUM.

These conversations are supplemented by:

• Current job advertisements and company careers page research
• ASX filings, results presentations and investor materials where relevant
• ASIC and APRA publications
• General AI and finance news

This is our first edition of the State of AI in Australian Private Credit and we aim to release a version of this report biannually. AI capability is changing quickly, so this is a point-in-time picture rather than a settled maturity ranking. In future volumes of our report we aim to include quantitative surveys to support our more qualitative interviews.

Australian private credit has transformed from a niche funding source to an estimated A$230 billion market in 2025 (the regulators are not actually sure what the size of the market is today). EY-Parthenon estimates it compounded at 21% a year from 2015 to 2025. This growth rate is nearly 4 times the growth rate of commercial bank lending (EY-Parthenon’s 2025 Australian Private Debt Market Overview).

“

We are aware of more than 240 private credit managers operating funds in Australia, with over 375 open-ended funds in market,” he said. “Around 70 per cent are focused on real estate rather than broader corporate lending.

Dugald Higgins, Head of Income Research at Zenith Group, quoted in The Australian.

For funds, competition is becoming more intense. A few basis points can be the difference between winning a deal and losing it.

“

Fund X is seen as Fund Y’s biggest competitive threat: larger and cheaper, putting pressure on our pricing advantage.

Analyst, A$2 billion FUM property fund

However the response cannot simply be to cut cost. ASIC has begun to turn its attention to the market making poor private credit practices an enforcement priority (ASIC’s 2026 enforcement priorities). ASIC’s concern has been around (ASIC’s June 2026 private credit update):

• credit deterioration,
• liquidity buffers,
• concentration-risk management,
• valuations that may lag economic reality

Funds are trying to do two things at once: run a leaner operating model and maintain and control risk.

Therefore the opportunities funds described come back to two outcomes:

1. Less manual work. Remove the repetitive assembly, checking and reformatting that absorbs analyst time.
2. More visibility and control. Give the fund a clearer view of what is happening, with source material, decisions and approvals easy to find and review.

Notably there was no real focus on utilising AI to generate alpha i.e. actions that give the funds a competitive advantage across risk mitigation or origination opportunities.

Automation of tedious workflows

Across the conversations, funds did not question the quality of their people. They questioned how their best people’s time was being allocated.

Their best analysts are often still assembling, checking and formatting information, often paper shuffling or moving sections around an excel or powerpoint. That work can be automated allowing the team to either:

• Focus on revenue generating work (meeting with prospective borrowers)
• Revenue protecting work (meeting and understanding existing borrowers)
• Handle larger volume of work (which naturally some funds hope for)

Below are the four workflows funds raised most often as opportunities for automation.

1. Loan monitoring and covenants

This was the clearest opportunity from our conversations. Monitoring is recurring, document-heavy and closely tied to a fund’s control environment. This is a particular area of focus for funds with:

• Broader landscape of high profile fund deterioration
• ASIC crackdown on valuations, of which loan monitoring is closely linked
• Growing funds looking to raise funds from institutional capital

Loan monitoring pulls together borrower reporting, covenant definitions, prior credit decisions, financial models and the portfolio manager’s current view.

“

Monitoring is the current #1 time sink. A huge amount of resources is getting monitoring up to speed.

CEO, ~$300M FUM Corporate Fund

This fund reported that each analyst spent roughly one business day per loan on a monthly monitoring cycle. With six to eight loans each, that equals approximately six to eight analyst-days a month, or roughly 30-40% of a 20-business-day month. This is a fund-reported baseline, not an independently audited measure.

The opportunity for monitoring is to collect the right inputs, at the right time, show the source material, flag what has changed and give the responsible analyst a better starting point for review. With that in place, the analysis can go deeper, drawing in more data and surfacing patterns across the book that would otherwise not be visible.

2. Deal intake and origination triage

“

The high volume of inbound deals is difficult to triage systematically.

Director, ~$600M FUM Corporate Lending Fund

This is a useful early workflow to automate because:

• It is bounded with a clear start and stop
• There are often clear criteria to quickly screen deals in and out for a team member to pick up and analyse
• Highly repetitive process occurring in some funds daily
• For funds focused on growing, origination is a bottleneck for growth
• It’s generally low risk

3. Credit papers and IC preparation

“

Our analysts spend too much time on formatting. How do we get this pitch deck to look like this in three minutes? So the team can spend their brain power on thinking through the deal

Head of Private Credit, ~$5 billion FUM, part of larger asset group

“

It takes two days for us to write an IC paper

Head of Property, ~$3 billion FUM, real estate and healthcare credit fund

A well automated IC process gives the analyst a stronger starting point and should be easier to trace and review. It should not create a polished recommendation that masks thin evidence.

We have seen interest in two uses:

• Turning a fund’s own prior papers into a structured reference set,
• Testing a new IC paper through defined reviewer personas

Both are well suited to AI systems, which can surface precedent, gaps and questions.

4. Valuations, reconciliations and investor reporting

These are often the largest manual blocks from our research.

“

It takes a week full time for each fund… We’d like to be communicating with our investors more, and just be a bit more systematic about that.

Analyst, ~$200M Boutique credit firm

Funds will often have internal or offshore teams to handle these operational processes. They are also more sensitive. A wrong valuation, a missed reconciliation item or an unsupported investor statement carries consequences far beyond a poor internal summary. These are not first-wave deployments.

For a fund at the start of its adoption curve, the sensible scope is speeding up preparation.

Seeing the fund as it is now.

Another large opportunity is simply being able to see the fund as it is now.

At one fund, portfolio reporting draws on around 15 separate touchpoints and is assembled manually each month. Its directors want one view of the book, showing what is on the watch list, what is expiring and where action is needed, without reviewing 75 to 100 loans one by one.

This is more than a reporting problem. To understand the current state of the fund, senior leaders often have to reconstruct the answer across multiple sources and people. When an analyst moves on, the fund can lose the history, judgement and exceptions that person has accumulated around a deal. The next person has to rebuild it from what they can find…

AI creates two opportunities:

1. Build shared fund context. Capture important information, decisions and rationale from everyday work, then retain it in the fund’s shared context rather than in one person’s inbox, files or memory.
2. Give each senior leader a realtime view of the information they care most about.
   • An analyst should see the information and next steps relevant to their deals.
   • A director should see the state of the book, the highest-priority risks and the actions that need attention

Every fund we spoke to was ‘doing something’ about AI. This appetite showed up in different forms.

Many of the funds we spoke with are still at the start of their AI journey.

“

Self-taught and ad hoc rather than systematically implemented.

MD of originations, ~$200M FUM corporate fund

“

Beginning of their AI adoption journey with no coordinated strategy currently in place.

COO, ~$750M Property Fund

“

Our adoption’s been slow.

Director of Investments, $2 billion FUM, part of larger asset group

These quotes show a market in motion. Tools are already in the hands of analysts, investment teams and operations staff. Formal capability has not caught up.

The question is not whether credit funds use AI. They already do. The question is whether the firm is getting value out of it…

So, is anyone getting real value yet? No, not really.

The demand for adoption is there, but the reality post implementation is very different. It starts with the fund struggling to articulate what it actually wants the outcome to be. Big ideas, no clear KPIs, and almost nothing invested in the execution.

The technology is genuinely revolutionary but that changes nothing when there is no genuine investment in change management, and clear metrics of what successful AI adoption looks like.

While each fund we spoke to had many hopes, aspirations, and dreams for what AI could do, not a single fund was happy with their progress. Below we list the most common failure points we observed from the funds we spoke to.

1. Poorly selected owner

You need someone (or even a team) accountable to make change happen. This person needs to have the time, focus, skill and authority to make things happen. We saw time and time again those selected to lead the fund’s AI transformation were set up to fail… or they did not exist…

“

They gave me the responsibility to implement AI as I’m the young guy in the team.

Analyst, $200M corporate lender

“

Whilst I am the most knowledgable on AI I appreciate that it needs to be someone that’s not me. I am already slammed

Director Private Investments, $300M NBFI lender

The below table highlights common patterns with AI ownership and common risks

We tested whether the market is beginning to hire for this role or AI skill set broadly. The data suggests not really and that’s not necessarily a bad thing.

On 17 August 2026, we reviewed live roles on Seek.com.au for credit funds. There were 86 live Credit roles and 87 Funds Management roles. We found no dedicated AI role advertised by an Australian private-credit fund manager. Nor was AI mentioned in the skillsets for any of these roles.

Despite this, there are some early signs that the largest managers are beginning to formalise the capability. Qualitas appointed a Chief AI Transformation Officer in 2025. While Balmain has an AI Adoption and Enablement Lead. These are not yet standard roles across the market. And even alone are not enough to see meaningful change across the organisation.

AI is not just a new technology, it’s a new way to do work. An expert in AI who knows how to use the tools won’t lead to organisation wide adoption. Instead you need tools that are capable of bridging the gap between the old and new ways.

For example, accountants did not adopt Excel because someone was hired who understood spreadsheets. They adopted it because it looked similar to the manual tooling they already used. Nobody had to learn a new way to think and the tool met them inside the job they already knew. Consequently a new way of working followed.

What this blocker contains: time, ownership, mandate, budget and accountability.

What it looks like in practice: a capable analyst building after hours; a director maintaining the tools; or a central programme that is not owned by the credit team.

2. Data is a blocker, but not in the way most think

Across the research set, origination data, post-settlement financials and supporting documents sat in separate systems. Teams rely on manual exports and reconciliation to join them, and some are part-way through migrations off legacy infrastructure. The data exists. Bringing it into a current, controlled view of a live position is the hard part. Ultimately, poorly accessible and fractured data will lead to poor quality AI output.

Almost every fund accurately identifies this problem, but they treat it as being more of a problem than it is in reality. And often it is the perception of the size of the problem that blocks any immediate progress.

The instinct is to clean everything first, move it all into one place, and only then start. That project rarely finishes, and it is not what the technology needs.

Think of it as a library rather than a data lake. You do not need every book rewritten, or even read. You need to know what is on each shelf, which edition is the current one, and where to send someone looking for a particular thing. A good catalogue makes the collection usable. Rewriting the collection does not.

The version problem is the real one. Nine drafts of the same IC memo sit in the folder and nothing in the file tells you which one went to committee. AI will confidently answer from the wrong one. Deciding which is authoritative, and keeping that true as new drafts land, is the work.

So the starting point is not cleaning all source data. It is mapping what exists, marking what is authoritative, and putting a process around keeping that map current.

The AI application that sits on top of that data is the value and can be realised much quicker than people think.

What this blocker contains: integration, data quality, systems age, security, access control, approvals and auditability.

What it looks like in practice: manual exports between systems; a tool that cannot access internal records; a useful experiment that cannot pass security review; or a platform replacement that keeps the organisation in transition.

3. IT and security slow deployment

Many funds described IT, security and management approval as a barrier to AI deployment.

“

We are very concerned about governance, privacy and data security given our client base.

IT team of an approximately A$1 billion property fund

“

Our AI rollout was deprioritised behind the organisation’s IT integration programme.

MD, portfolio monitoring at a multi-billion-dollar property fund

“

We were only allowed Copilot. You already have a hand behind your back.

Analyst, A$2 billion property fund

When approval feels slow or opaque, analysts abandon useful ideas or experiment outside approved processes (both were observed). One outcome leaves the fund unable to learn and at risk of losing good people. The other creates ungoverned use of sensitive information. The practical response is to involve IT and security early around a specific workflow and make the business case clear.

What this blocker contains: data classification and privacy, access permissions, vendor risk, data residency and retention, model-training terms, audit trails and incident response.

What it looks like in practice: a team cannot use a promising tool because it is unclear where client data is stored; an experiment stalls in a lengthy vendor-security review; analysts rely on consumer tools because no approved alternative exists; or access is granted without clear permissions, monitoring or accountability.

4. Adoption stalls between the individual and the organisation

Being good at AI yourself is not the same as your organisation being good at AI. We see three levels of AI fluency:

• Personal fluency: an individual knows how to use AI well in their own work.
• Team fluency: people share practical ways of using it, know when to trust it, and can hand work between each other.
• Organisational fluency: a workflow is owned, governed, measured and maintained. It works even when the original power user is busy or leaves.

Some funds in this research show personal AI fluency. Fewer have translated it into shared team practice. We spoke with only one fund taking deliberate steps towards organisation-wide capability.

That is why licences alone do not create adoption. A fund can buy access to a tool and still work exactly as before. It is not principally a technology problem. It is a people-and-process problem. Often these change management skills do not exist internally in the fund.

What this blocker contains: trust, workflow design, training, clear expectations, shared review, measurement, maintenance and human accountability.

What it looks like in practice: a few power users moving ahead of the team; a pilot that never becomes part of the process; or a high-performing workflow that depends on one person to keep it alive.

Our research points to six practical conditions that turn AI hopes and dreams into reality.

We believe the next step for AI in credit is a shift from individual use to fund-level change in how work gets done. Other regulated industries are already moving beyond individual experimentation…

Financial services is moving from pilots to operating systems

Large financial institutions show what operational AI looks like once the data, governance and internal capability exist.

• CBA reported about A$200 million in measured gross AI benefit in FY26, with roughly 80 per cent of staff actively using enterprise AI tools.
• Suncorp reported 3,900 staff-built agents.
• IAG has flagged about A$200 million of AI and AI-enablement investment for FY27, alongside more than 600 internal activators and over 90 published agents.

Specialist operating systems are emerging in adjacent professions

We are also seeing organisational AI capability be productised in law and banking. Law has moved beyond generic chat tools towards products that connect professional context, review and workflow. Currently there are two leading AI law products, Harvey and Legora. Both position themselves as domain-specific AI for legal and professional services, including contract analysis, diligence, compliance and litigation workflows.

Investment banking has comparable specialist platforms in Rogo and Hebbia. Both connect internal and external financial data with agents designed to produce work across PowerPoint, Excel and Word.

Private credit has the same ingredients: proprietary information, recurring documents, human accountability, a defined decision process and a need for source-level confidence. It has not yet developed the same operating layer.

Where Levercon sees the future of AI and credit

We do not think the future of AI in credit is a generic chat tool licensed to every employee, or one broad agent given access to the whole fund.

We think it is a fund operating layer: shared context, a small number of well-defined workflows, clear permissions, human sign-off and a feedback loop that keeps improving the work.

That is what we are building at Levercon.

If you would like to find out more please reach out.

Levercon is building customised AI operating systems for credit funds.

Credit teams already hold the information they need to make better decisions. It is spread across loan files, borrower reports, emails, models, policies and the people who know the portfolio best. Levercon connects that context, deploys workflow-specific agents and builds the review and feedback loops that allow the work to improve over time.

We start by embedding with your team to learn how the fund actually runs. The workflows you repeat, the systems they run across, and where the time actually goes. The operating system is then configured to your strategy, your team and your requirements, because no two funds run alike.

The aim is not to remove judgement. It is to give the people accountable for it a stronger way to work.

Further information

To discuss the findings, compare your fund’s starting point with the research, or explore a defined first workflow, contact the Levercon team.

Suggested citation

Davies, C. and Carp, J. (2026). The State of AI in Australian Credit. Levercon.