# Mako, a PXL Labs product

> AI-assisted credit analysis for SME lenders. Documents in, traceable credit summary out.

Canonical page: https://pxl.services/labs/mako
Last updated: 14 Oct 2026

Status: Prototype
Public link: https://mako.pxl.services

## The problem

An SME loan application arrives as a stack of PDFs: months of bank statements, financial statements, management accounts. An analyst re-types the numbers into a spreadsheet before any actual analysis starts.

That work is slow, error-prone and the same every time. It's also where the risk signals hide: a customer that makes up half of all inflows, a run of returned debit orders, an overdraft that never comes down.

## What it does

1. **Ingest**: Takes the application pack as uploaded: scanned or digital statements, financials and supporting documents.
2. **Extract**: Structures transactions and financial line items, and keeps a link from every figure back to its page in the source.
3. **Analyse**: Computes the ratios an analyst would and flags risk signals, each with the evidence that triggered it.
4. **Report**: Drafts a credit summary with citations, written to be checked line by line, not trusted blindly.

## How it's built

### Decision-support, by design

Mako never approves or declines. It prepares the file; the analyst makes the call. Every claim in a summary carries a citation.

### Traceable extraction

Figures keep their source document and page, so a reviewer can verify any number in one click.

### Synthetic data only

The public demo runs on a synthetic borrower case. No real applicant data has been used.

### API-first

Each stage is designed to be callable on its own, so a lender can plug extraction or analysis into an existing credit workflow.

## Timeline

- 2026: Proof of concept completed, built end to end to the same scope as the lender engagement.
- 2026: Fully scoped with a prospective lender at the start of the year. The engagement didn't go ahead because of university time commitments, so the scope became the proof of concept.

## What's next

- An evaluation set to measure extraction accuracy per field.
- Conversations with lenders about piloting on their own document formats.

As a client build: [Lending and insurance](https://pxl.services/solutions/lending)