Petran

Broker workflows live in spreadsheets. That was the whole problem.

Australian equipment-finance brokers spend 90 minutes per deal reading a client's P&L, hand-calculating add-backs, checking lender policy PDFs, and cross-referencing rate cards. Every broker's method lives in their head. Every firm's competitive advantage is one lost laptop away from extinction.

Petran's brief was to collapse that workflow into a single product: upload statements, extract numbers, calculate DSCR, match against lender policies, and rank by rate. Deterministically, same inputs, same recommendation, every time. And deliver it as a production-ready React app a broker could open Monday morning and stop reaching for Excel.

A financial tool has one second to feel trustworthy.

Auth screens are the first thing a broker sees. We paired a calm left rail, logo, task, single action, with a generative glass visual on the right. The split-screen stays consistent across login, sign-up, password reset, and confirmation, so every state feels like the same product. Error states inherit the system red (#EF4444); success confirmations use #22C55E.

One client, one file, one source of truth.

Before the engine runs, the broker needs a place to drop business context. The client list uses a card-grid with status chips (Active · Pending · Rejected) so a broker scanning their pipeline at 8am sees exactly what's waiting on them. Each card surfaces ABN age, GST age, credit score, and annual turnover, the four numbers that determine which lender shortlist is even possible.

Opening a client drops into a two-column form: Business Information (ABN, GST, property ownership) and Contact details. Every field maps one-to-one to a criterion that shows up downstream in the recommendation engine. No orphan data.

Upload a P&L. Get a serviceability verdict.

The financials step accepts a drag-and-drop PDF. An LLM call (temperature 0, structured output schema) returns the trading income, operating profit, depreciation, interest expense, and non-trading income, each with a confidence score the broker can see and override. Every number is editable. The AI is a first draft; the broker is the audit.

Once extracted, the system computes Adjusted EBITDA by adding back depreciation and interest, excludes non-trading income, and applies the 25% tax rate (for turnover under $50M). That figure, divided by annual debt commitments, is the Debt Service Coverage Ratio, the number every lender cares about before anything else.

The shortlist is a direct function of the inputs.

Each lender is modeled as a set of thresholds: ABN age, GST age, minimum DSCR, Veda/comprehensive score, property-ownership requirement, asset-age-at-end-of-term cap, and loan-amount ceiling. Some lenders run two rate tiers, one with property ownership, one without, with a 20% deposit requirement on the latter.

The broker enters the loan scenario, amount, term, asset type, asset age, credit scores, ownership. The engine filters the rate card, ranks eligible lenders by rate, and explains why the rest were excluded. A 12-year-old truck on a 5-year loan silently removes any lender with a 15-year end-of-term cap. A 680 Veda score with a $250k ask unlocks the mid-tier products. No spreadsheet. No discretion. Same inputs, same answer.

Browse the full rate card. Know what's possible before you ask.

Outside the assessment flow, brokers need a read-only view of every lender's criteria, loan ranges, credit score requirements, asset-age limits, fees, ownership rules. The Lender Directory shows all 13 as cards with both rate tiers exposed (with property / without property), so a broker answering a client call can quote a ballpark in under 30 seconds.