Harbour Joinery Pty LtdWorking capital facility, due for renewal
ReferDeclared figures contradicted by the account
Decisions you can defend.
Agent-driven decision intelligence for financial institutions.
Discover what your current methods miss. Measure what improves the decision.
Book a conversation See what you can start now- Bank statement540 transactions, Aug 2025 to Aug 2026
txn 0001 to 0540 - Application and broker noteLargest customer about 20%; other debt: equipment rental only
application.json - PPSR searchSecurity over all present and after acquired property, two lenders
8 Sep 2026
Illustrative example. Harbour Joinery is a synthetic business.
- Without this
- The facility is renewed at the same limit on the declared figures: revenue counted from every inflow, the largest customer at about 20%, no other secured lenders. The risk is carried without being priced.
- With this
- The file is referred before the limit is renewed, with three questions the assessor can put to the customer. Files where the account agrees with the declaration are not referred, so the assessor's time goes where the evidence conflicts.
- What moves
- Bad debt on renewals, and assessor hours per renewal.
On research data on bank accounts: two years of statements arriving as four overlapping packs were reconciled into one history that matched the full account record; joined as they came, they overstated the average risk forecast by 54%. See the case
Personal account opened 12 days agoEveryday account, online onboarding
HoldPattern of a mule account, far outside its peers
- Inbound payments46 senders, none linked to the customer
pay in 2026-09-24 to 30 - Outbound paymentsTransfers to two other new accounts
pay out 2026-09-24 to 30 - Account opening recordOnline onboarding, 12 days ago
KYC 2026-09-18 - Peer groupPersonal accounts opened in the last month
3,210 accounts
Illustrative example.
- Without this
- $38,400 from 46 people leaves within two hours to two other new accounts. A payment level alert fires on one transfer after the money has already moved on twice.
- With this
- The account is held while the money is still in it. The team reviews one account instead of 46 payments, and the funds can be returned.
- What moves
- Money stopped per reviewer hour, and victims' funds recovered.
On synthetic bank transfer data: holding the 100 highest ranked accounts a day would have stopped $33.4m of the $78.4m they went on to move, against $14.9m ranking by the riskiest payment. See the case
One engine under every decision.
What you can start now
Every product starts as a fixed-fee pilot on your own deals, with agreed success criteria, measured against your own people or models. The winner is named even when it is not us.
For lenders and brokers
The underwriting pilot
The memo under your own credit policy, with bank statement analysis, the behavioural assessment and business verification inside it. Measured against your own underwriters.
The cashflow score pilot
A probability of default built on your own book in the pilot, run beside your scorecard and priced per score afterwards. A separate renewal score reads each customer's own history as well.
Take one or both. The two hand cases to each other: the score decides the straightforward cases under your policy, and the memo handles thin files, exceptions and referrals.
For banks and payments networks
The fraud pilot
One problem, scams, mule accounts or account takeover, tested beside your current controls at your review capacity. AI agents surface emerging patterns; named people approve.
The evaluation
What a vendor model, data feed or AI configuration adds to your baseline before you buy it, with the coverage table. The answer may be "keep what you have", and we say so.
Built for business lending: every signal, model, check and memo is engineered around how a business operates. All products
How the work gets done
Specialised agents propose the work. Each change is tested by our proprietary analytical engine, whose results repeat on the same inputs, against your own controls and data, at the same review capacity, and priced before adoption. Accountable people approve every consequential change.
What you get
More problems addressed
More decision problems are worked on with the same specialists.
No problem starts from zero
The lab retains what each investigation learns, so the next one begins further along.
Evidence you can approve
Every change arrives with the evidence that risk and compliance leaders need to approve it.
Your layer stays private
Your data, policies and findings stay in your private layer.
One engine, many applications
Each application is validated separately.
Underwriting
Understand the borrower.
- Underwriting memoYour policy, systems and underwriters' judgement in one decision-ready memo. See the memo.
- Bank statement analysisEvery account and transaction labelled by what it is to the business, anything unclear flagged. See the analysis.
- Behavioural assessmentAn applicant's statements rated on five behavioural measures. See the assessment.
- KYB business verificationAustralian registers, owners, address, web presence and related entities, every check with its evidence. See the checks.
Credit models
Predict the outcome.
- Cashflow scoreA cashflow credit score from bank statements, explained signal by signal and calibrated to your book. See the score.
- RenewalsWhat to do at renewal, ranked by risk and value. See renewals.
- Pricing and offersA price for each customer, and offers ranked within the rules you set. See pricing and offers.
Portfolio monitoring
Steer the book.
- Early warningCustomers heading for trouble before they miss a repayment. See early warning.
- Hardship and collectionsWhich actions cause payment, for which customers. See hardship and collections.
- PortfolioWhy the book moved, what happens if conditions change, and where capital should go. See the portfolio.
Fraud and scams
Find what controls miss.
- Fraud and scamsFind what existing controls miss and measure what a change would improve. See fraud and scams.
Five ways to use it
- Challenge existing controlsSee what your current rules and models miss.
- Build a decision use caseStart a new decision with the evidence behind it.
- Price a proposed changeKnow the effect of a change before it is adopted.
- Oversee models and feedsKeep watch on what is already in use.
- Evaluate a vendor or data sourceThe winner is named, even when it is not us.
Start with one pilot.
A fixed-fee pilot of one product on your own deals, with success criteria agreed first and no live changes. Then the hosted service for lenders and brokers, or your own environment for institutions. Terms are agreed after the pilot.
Your data, policies and findings stay under your control: hosted for lenders and brokers, in your environment for institutions. Security and governance
Luminnovate was founded in Sydney by Dr Parisa Milne, who holds a PhD in signal processing, pattern recognition and machine learning and has spent twenty years in applied machine learning across healthcare, retail and financial services.
Book a conversation