Doc review

Ask the same questions of every document.
Check every answer against a page.

The review grid for twelve tenancy leases, every cell answered: 36 of 36, with assignment clauses, rents and expiry dates in columns

Build a review from a matter's documents or a standalone set, add your questions as columns, and run. Every document and question pair is answered on its own, each answer carries the passage it was read from, and the grid fills in front of you.

Book a demo See all features
Encrypted in transit Documents and AI in Australia No lock-in contracts Guided setup available Your data exportable on request

Last reviewed

How does bulk document review work for law firms?

A review project is a set of documents and a set of questions. MatterFirst answers every question against every document, in parallel, and renders the results as a grid: one row per document, one column per question. Every answer carries the verbatim passage it was read from, and its page, so each square can be checked.

How it works

From document set to answered grid

1

Add documents

2

Ask questions

3

One cell per pair

4

Answered with an excerpt

5

Grid fills live

The grid takes shape before the first answer lands

An excerpt that cannot be verified renders blank, never as an answer

A failed cell retries alone; the run is never lost

Each cell is its own job: answered independently, retried independently, and never shown without a passage that was found in the source document.

Review projects

One row per document. One column per question.

Add documents from a matter, or build a standalone set, then add your questions as columns. Running the project creates one cell for every document and question pair, and each cell is answered on its own, in parallel, rather than as one long job that can fail as a whole.

  • Rows are documents, columns are questions, and each cell is answered independently
  • The grid takes shape immediately and fills in live as answers land
  • Runs across a matter's documents or a standalone set that belongs to none
  • Progress counted as it runs: answered, not found and errors shown separately

Capability 1 of 4

The grid appears first. The answers arrive into it.

Typed questions

Columns with types keep answers comparable

Every column has an answer type: text, date, money, number, yes or no, a list, a party, or a fixed set of options. The type shapes the schema the model must answer into, so a date column comes back as a date and an options column can only answer with the options you set.

  • Eight answer types, enforced through the model's answer schema rather than the prompt
  • Option columns constrained to your own list, so one concept gets one spelling
  • Practice-area templates seed the standard question set in one step
  • Applying a template twice skips questions already on the project

Capability 2 of 4

The type is a constraint, not a suggestion.

Verifiable answers

Every square points at a page

An answered cell carries the verbatim passage it was read from and the page it appears on. That excerpt is checked against the document's own text: if it cannot be found there, the cell records an error and renders blank, because a square that cannot be checked is worth less than an empty one.

  • A verbatim source excerpt and page reference on every answered cell
  • Excerpts verified against the document text before an answer is shown
  • A document that does not answer the question is recorded as not found, which is itself a finding
  • CSV and XLSX export, with a second sheet carrying the excerpt and page behind every answer

Capability 3 of 4

Nothing rendered that cannot be checked.

Human review

Verify once. Keep the work.

Each cell can be verified or corrected by a person, and a correction is kept beside the extraction rather than written over it. Where a column is mapped to the matter's fact record, verifying a cell also lands the value there with its source attached, so the review's work outlives the review.

  • One-click verify, or correct a value with the original extraction kept alongside
  • Verified answers can land on the matter as facts, with the excerpt and page as their source
  • A failed cell retries on its own; the rest of the grid is never redone
  • Re-running a whole column is a separate, deliberate act, so verified work is not lost to a stray click

Capability 4 of 4

The grid's work compounds into the matter.

MatterFirst is an AI-native legal practice management platform for Australian law firms, focused on small and boutique firms from solo practitioners up to about 20 users, built by NorthCape Technology, covering matters, AI document work, configurable workflows, automatic time capture for review, Matti, built-in signing, trust accounting, billing and a client portal, with firm and matter visibility of MatterFirst AI usage charges.

This suits work where the same questions must be asked of many documents: due diligence, a lease portfolio, a book of employment contracts, a discovery set. It is not the tool for understanding one document deeply, which is what document extraction does, and a set small enough to read in an afternoon is usually better read. The grid does not replace judgement: it puts every answer next to its evidence so a lawyer can apply some.

FAQ

Frequently asked questions

How is this different from MatterFirst's document extraction?
Document extraction reads one document and feeds structured values into that matter's fields. Bulk review asks one set of questions of many documents at once and renders the answers as a grid you can compare across the set. Use extraction for the contract in front of you, and review for the pile.
Can the AI just make an answer up?
It cannot show an answer without evidence. Every answer must quote a verbatim passage, and that passage is checked against the document's actual text before the cell is shown. If the quoted passage cannot be found, the cell records an error and renders blank rather than as an answer. A model can still misread a real passage, which is why the excerpt sits beside the answer and each cell can be verified or corrected by a person.
What happens when a document does not answer a question?
The cell is recorded as not found and shows a dash, which is different from an error. Not found is a real result: a set of leases where none contains a make-good clause is a finding, and the grid records it as one.
How is a review priced?
Per cell, against your firm's AI balance in Australian dollars. The cost of a run is quoted before it starts, cells that fail are not charged, and re-running a partly answered project only pays for the cells it actually answers. Firms can also set a daily spend ceiling that stops work rather than reporting it afterwards.
Where is my data stored?
The application runs on Fly.io in Sydney. Database and backups are hosted on Fly.io Managed high-availability PostgreSQL clusters in Sydney. Documents are stored in Amazon Web Services S3 in Sydney. Document extraction and matter questions run on Amazon Web Services Bedrock in Australia, and reading a scanned document runs in Microsoft Azure's Australia East region. These locations describe the application, document storage, database, backups and document AI services. The trust evidence page at /trust records the scope and dated evidence for each published control.
How much does MatterFirst cost?
Pricing is per workspace rather than per seat, starting at $199 per month for Solo. Each plan includes a set number of users, a monthly AI balance in Australian dollars, the client portal and automations. Extra users are billed only past the included seats. Full plan inclusions are on the pricing page.
Can I get my data out if I leave?
Yes. Your matters, contacts, documents and time records are exportable on request, and there is no lock-in contract. We would rather you were able to leave and chose not to.
Is there a free trial?
Yes, a 14 day guided trial on Solo, Practice or Firm. It is guided rather than self-serve because a practice management system tells you very little until it has your real matters in it, and we would rather help you load them than leave you looking at an empty workspace.
Who builds MatterFirst?
NorthCape Technology, a Melbourne-based software company. MatterFirst is built in Australia for Australian legal practice, with trust accounting and Xero as first-class capabilities. Available through coordinated setup: PEXA, InfoTrack. Coordinated manual setup is arranged with MatterFirst and the provider.
Is MatterFirst suitable for a firm of my size?
It is designed for solo practitioners up to firms of roughly twenty seats. Below that, Solo carries the full platform for one user. Above it, the platform works but you should expect to talk to us about migration and rollout rather than signing up online.

Reviewed by Nick Hallam

Co-founder and Operations Director, NorthCape Technology

Aerospace engineer with a background in mission-critical software and complex systems integration, and the one who talks to firms about how they actually run. Keeps MatterFirst grounded in how firms operate, and reviews the practice-area and workflow pages on this site.

Last reviewed

Ready to put a document set in a grid?

Book a demo and bring the questions from your last due diligence or lease review. We will build the columns with you, so you can judge the answers against your own documents.

Book a demo Talk to us