TL;DR: A review matrix turns a pile of documents into a structured grid - each row a file, each column a question, every cell a cited answer. In Judicio you select multiple documents, build up to 25 questions (write them yourself, let the AI draft them from your files, or start from a template), choose an answer type for each, and run. Then you read per-cell confidence and citations, clear flagged cells in a review queue, scan the Insights view, and export to Excel, CSV, Word, or PDF. This guide walks the whole workflow with a worked lease-portfolio example.
If you have ever opened twenty leases to answer the same five questions about each one - who is the tenant, when does the term end, is there a renewal option, what is the rent, can the landlord assign - you already know the problem a review matrix solves. Done by hand, that work is slow, repetitive, and easy to get subtly wrong: a transposed date here, a missed clause there. A review matrix asks your questions once and applies them across every document at the same time, returning a grid you can scan in minutes with every answer cited to the page. This guide explains what a review matrix is, how to build one step by step in Judicio, and how to read, verify, and export the results.
What is a review matrix, and when should you use one?
A review matrix is a grid that crosses your documents against your questions. Each row represents one document; each column represents one question; and each cell holds the answer for that document, typed and cited. It is the right tool whenever you have many documents and the same set of questions for all of them - a portfolio of leases, a stack of NDAs, a folder of vendor contracts, or a set of employment agreements. Where a single-document review answers many questions about one file, a matrix answers the same questions about many files, which is exactly what comparison work demands.
The distinction that matters is repetition. If you are reading one complex agreement closely, document review is the better fit. If you are checking the same handful of points across a batch, the matrix is faster and more consistent, because it asks every question of every document in precisely the same way. Judicio's Review Matrix is built for that batch case, and it sits alongside Document Review in the same workspace, so you can move between the two without re-uploading anything.
How do you set up a review matrix step by step?
Setting up a matrix takes four steps: choose your documents, build your questions, set an answer type for each, and run. None of them requires technical skill - the work is deciding what you want to know.
Step 1: Select multiple documents
Start by uploading your files once into the File Library - drag and drop files, whole folders, or ZIPs, or import from Google Drive, OneDrive, SharePoint, or iManage. From there you select multiple documents for a single matrix run. If your set is larger, you split it into batches and run them in turn; the matrix is deliberately not a hundreds-at-once tool, and keeping each run focused keeps it fast and reviewable. Because every file is processed once on upload - with automatic OCR for scans - the documents are ready to query the moment you select them.
Step 2: Build up to 25 questions
A matrix holds up to 25 questions, and you have three ways to create them. You can write each question yourself; you can let the AI propose questions from the documents you selected or from a plain-English description of what you are looking for; or you can start from a template and adjust. A neat shortcut: type a column header such as Governing Law and the tool suggests a tuned question to match. Keep questions specific and answerable from the four corners of the document - asking for the annual rent is better than asking whether the rent is reasonable, because the first has an answer on the page and the second asks for judgment the matrix cannot supply.
Step 3: Choose the answer type for each column
Every question carries one of ten answer types, and choosing the right one keeps your results clean and sortable. Ask for a tenant name as Text, a lease end as a Date, the rent as Currency, a break option as Yes/No, an escalation as a Percentage, the permitted uses as a List, and a one-line description of an unusual clause as a Summary; Amount, Monetary Amount, and Tag round out the set. Typed answers mean a column of dates sorts as dates and a column of currencies adds up as money, rather than arriving as loose text you have to clean afterwards.
Step 4: Run the matrix
With documents, questions, and types in place, you run the matrix. Before it starts, Judicio shows an up-front time estimate so you know roughly how long the job will take, and the run continues server-side even if you close the tab - useful for a large document set. You can choose Deep Mode when you want the AI to read every page rather than just locate the relevant passages. When the run finishes, you have a complete grid: documents down the side, questions across the top, a cited answer in every cell.
A worked example: reviewing a lease portfolio
Suppose you are reviewing forty commercial leases ahead of a portfolio acquisition. You select them all - here, forty fit comfortably in one run - and build a compact matrix of the points that drive value and risk. The columns might look like this:
| Matrix column | Example question | Answer type |
|---|---|---|
| Tenant | Who is the named tenant? | Text |
| Lease term end | When does the current term expire? | Date |
| Annual rent | What is the annual rent payable? | Currency |
| Rent escalation | What is the annual escalation rate? | Percentage |
| Renewal option | Does the tenant have a renewal option? | Yes/No |
| Assignment | Can the tenant assign or sublet, and on what conditions? | Summary |
You run it, and in one pass you have a forty-row table that tells you which leases expire within two years, which carry above-market escalations, which restrict assignment in ways that complicate the deal, and which grant renewal rights you need to price. What would have been a day of opening files and transcribing terms into a spreadsheet becomes a grid you read in a sitting - and because every cell is cited, you can jump to the exact clause behind any figure before you rely on it. The same shape works for an NDA portfolio (term, definition of confidential information, permitted use, residuals, governing law) or any other batch of like documents.
How do you read per-cell confidence and citations?
A matrix is only trustworthy if it tells you how sure it is. Every cell carries a confidence signal shown as a colour: green for Clear, yellow for Ambiguous, red for Low confidence, and grey for Not addressed - meaning the document does not answer the question. That colour-coding is what lets you triage at a glance: green cells you can usually accept after a spot check, yellow and red cells you read closely, and grey cells tell you the term is simply absent from that document, which is often a finding in itself.
Alongside the confidence signal, each answered cell carries a citation to the exact page and the quoted passage it came from, with a section label. Clicking the citation opens the source document with the relevant region highlighted, so verification is a quick read rather than a re-search. The citation labels are deterministic - generated from the source, never invented by the model - so the same clause always carries the same reference. The American Bar Association's guidance on technology competence expects lawyers to understand the tools they use; reading the citation behind a cell is exactly the habit that satisfies it.
How do you work the review queue for flagged cells?
You do not have to check fifty cells one by one to find the ones that need attention. The Review Matrix collects uncertain cells into a review queue, so you can work straight through the answers the AI was least sure about - the ambiguous and low-confidence cells - and leave the clear ones for a lighter spot check. For any cell you can open the reasoning, re-extract a single cell if you want the AI to try again, or correct the answer yourself and mark it reviewed.
This is where the matrix earns its keep on accuracy. Rather than trusting the whole grid or distrusting all of it, you direct your attention to the small share of cells that are genuinely uncertain, confirm them against the cited source, and sign them off. A reviewed checkmark records that a human has looked, which gives you - and anyone reviewing your work later - a clear picture of what has been verified and what has not.
What does the Insights view tell you?
Beyond the grid itself, the Insights view summarises the whole matrix at a glance. Stat cards report how many files and questions you ran, how many cells were answered, how many need review, and whether any files failed to process. A confidence heatmap shows the grid coloured by certainty, so clusters of red or grey jump out - a column where many documents are silent on a point, or a document that produced weak answers across the board and may need a closer manual read.
Used well, Insights is a roadmap. A column full of grey Not addressed cells tells you which documents lack a term you expected; a band of low-confidence answers points to where the records are ambiguous and a follow-up is warranted. Instead of a flat table, you get a quick read on both what you have learned and where the gaps are.
How do you export the results?
When the grid is reviewed, you export it. The Review Matrix supports Excel, CSV, Word, and PDF, with a citations toggle so you can include or omit the page references depending on who the export is for. An Excel or CSV export is ideal when you want to sort, filter, or hand the data to a colleague or a financial model; a Word or PDF export with citations is better for a memo or a file note that needs to show its sources.
Because the typed answers are preserved, dates stay dates and currencies stay numbers in the spreadsheet, so the export is immediately usable rather than a block of text to clean up. For more on getting structured data out of documents, see our practical guide to legal data extraction, and for applying the matrix to transactions, our guide to AI for due diligence data rooms.
What are the limits, and what should you verify?
A review matrix is a powerful first pass, not a final answer. It can misread an oddly drafted clause, mistake one date for another in a poorly scanned document, or mark a cell Clear when the language is genuinely ambiguous. The discipline that keeps it safe is simple: read the cited passage behind any cell you intend to rely on, pay particular attention to the yellow and red cells the queue surfaces, and treat grey cells as a prompt to confirm the term really is absent rather than just missed. The matrix does the cross-referencing; you supply the legal judgment about what the answers mean.
It also will not tell you whether a rent is fair, a clause is enforceable, or a deal is wise - those are questions for you. Outputs are not legal advice. Kept in that lane, a review matrix removes the mechanical grind of comparison and frees your attention for the analysis only a lawyer can do. For the related batch single-document workflow, see our overview of bulk document review with AI.
Getting started with Judicio
Pick one batch you would normally compare by hand - a set of leases, a folder of NDAs, a stack of vendor contracts - and build a short matrix of the questions you ask every time. Upload the files once into the File Library, draft your questions or start from a template, set the answer types, and run. Compare how long the matrix takes against a manual read, verify the flagged cells against their citations, and let the result decide.
You can try it on your own documents with a 7-day free trial - 500 credits, no credit card required. Professional access is $200 per month for 5,000 credits, and you can explore the full feature set or contact us for a walkthrough. The matrix handles the comparison; the judgment stays with you.