AMROAR Technologies

Amroar Technologies — a global IT company working across Salesforce, HubSpot, AI and full stack development
Book a Call
Lexara Group Case Study — Amroar Technologies
Case Studies › Agentic AI › Lexara Group
Agentic AI · Anthropic Claude · Legal

4,000 pages of
due diligence. Reviewed,
risk-scored, and summarised
in under three hours.

Lexara Group advises on M&A transactions across financial services and technology. Every deal required weeks of manual document review. Amroar built a four-agent Claude system that reads the full data room, flags risks, maps obligations, and delivers a structured legal brief — before the first associate opens a document.

View Agentic AI Work
Client
Lexara Group
Sector
Legal · M&A Advisory
AI Model
Anthropic Claude
Agents
4 autonomous agents
Context
200K token window
Scope
End-to-end due diligence
<3hrs
Full data room processed — previously 3–4 weeks of associate time
94%
Risk clause identification accuracy validated against manual review
200K
Token context window — entire contracts read in a single pass, no chunking
0
Failed deliverables across the full agentic AI engagement
The Challenge

M&A due diligence that takes weeks and costs a fortune — for every deal.

Lexara's associates were spending the majority of their billable hours reading documents that required attention but not judgment. Data rooms with thousands of pages of contracts, financial disclosures, employment agreements, and regulatory filings needed to be reviewed before a single strategic question could be asked. The cost and time were limiting how many deals Lexara could run simultaneously.

01
Data rooms with thousands of pages, reviewed manually
Every M&A transaction brought a data room of 2,000–6,000 pages — shareholder agreements, employment contracts, IP assignments, regulatory filings. Associates read every document looking for the same risk categories, duplicating effort across every deal.
02
Key obligations and risk clauses buried in long documents
Change-of-control provisions, restrictive covenants, penalty clauses, and liability caps were scattered across hundreds of documents. Missing one in a 400-page shareholder agreement carried material deal risk. Manual review under time pressure was inherently error-prone.
03
Associates spending billable time on document triage, not legal judgment
The partners at Lexara needed associates doing the work that requires legal expertise — structuring advice, negotiation preparation, risk framing. Instead, they were reading standard employment agreements and summarising boilerplate clauses.
04
Deal pace limited by review capacity, not deal pipeline
Lexara was turning away transactions because the due diligence team was at capacity. The bottleneck wasn't expertise or deal flow — it was document processing time. Every deal that couldn't be taken on was revenue left on the table.
Document Intelligence Architecture

How Claude processes an entire data room.

Documents enter the pipeline and four specialised Claude agents process them in sequence — each passing a structured output to the next. The full 200K token context window means entire contracts are read at once, not in fragments. No context is lost between sections.

Document Inputs
Shareholder Agreements
PDF · avg 180 pages
Employment Contracts
PDF · bulk upload
IP & Licensing Agreements
PDF · multiple parties
Regulatory Filings
Structured & unstructured
Financial Disclosures
Statements · notes · schedules
Document Intelligence Engine
Anthropic Claude
claude-3-5-sonnet-20241022
Four specialised agents coordinate via structured tool-use handoffs. Each reads documents in full — no chunking, no lost context.
Review Agent
Risk Agent
Obligation Agent
Brief Agent
200K
Token context window
Tool Use
Structured outputs
4-pass
Agent pipeline
Structured Outputs
Risk Register
Flagged clauses with severity score
Obligations Map
Every obligation, party, and deadline
Change-of-Control Summary
Provisions triggered by the transaction
IP Ownership Report
Rights, encumbrances, transferability
Executive Legal Brief
Partner-ready summary with recommendations
The Four Agents

Each agent has one job and does it completely.

No agent tries to do everything. Each one reads the documents through a specific lens, produces a structured output, and passes it to the next. The final brief is built from four expert passes, not one generalised attempt.

01
Document Review Agent
First-pass reading and classification
Claude · Tool Use
Reads every document in the data room in full, using Claude's 200K context window to process entire agreements without fragmentation. Classifies each document by type, extracts metadata, identifies parties and governing law, and creates a structured index that subsequent agents reference without re-reading source documents.
Processes
Reads complete documents — no chunking, no truncation
Extracts parties, dates, governing law, agreement type
Builds structured document index for downstream agents
Flags documents requiring specialist review (tax, regulatory)
claude-3-5-sonnet
Tool Use
200K Context
02
Risk Agent
Clause-level risk identification
Claude · Tool Use
Works from the document index produced by Agent 01 and reads each agreement specifically for risk-flagging. Identifies change-of-control clauses, termination provisions, liability caps, indemnity obligations, penalty clauses, and restrictive covenants — scoring each by materiality and flagging whether it is triggered by the proposed transaction.
Identifies
Change-of-control provisions triggered by the deal
Uncapped liability and indemnity clauses
Non-compete and non-solicitation obligations on key personnel
IP assignment gaps and licence termination rights
Output
Risk register with clause references and severity scores
Format
Structured JSON → readable risk table
claude-3-5-sonnet
Tool Use
Structured Output
03
Obligation Agent
Obligations, deadlines, and consent requirements
Claude · Tool Use
Maps every binding obligation in the data room to the party responsible, the deadline or trigger condition, and whether it requires third-party consent. Produces a complete obligations register — organised by counterparty and transaction phase — so the deal team knows exactly what needs to happen before completion and what consents must be obtained.
Maps
All third-party consent requirements (lender, landlord, supplier)
Regulatory notification and approval obligations
Post-completion obligations and conditions precedent
Deadline-critical obligations with date extraction
Output
Obligations map by party, trigger, and deadline
Format
Tabular output for deal management
claude-3-5-sonnet
Tool Use
Date Extraction
04
Brief Agent
Executive legal brief generation
Claude · Long-form
Synthesises the outputs of agents 01–03 into a structured, partner-ready legal brief. Formats findings into the firm's standard due diligence report template — executive summary, material risks by category, obligations register, recommended conditions precedent, and transaction-specific issues requiring partner attention. The brief is drafted for a legal audience and reviewed by an associate before delivery, not generated and sent raw.
Produces
Executive summary — deal overview and top 5 material risks
Risk register in standard Lexara template format
Obligations map with consent action items
Recommended conditions precedent based on identified risks
Issues list requiring partner-level strategic judgment
claude-3-5-sonnet
Long-form Generation
Template Formatting
The Outcome

Weeks of work
done in hours.

Measured across 14 live M&A transactions in the first six months of deployment. Every metric below was validated by comparing agent output against a parallel manual review on the first three deals before associates handed off the process.

<3hrs
Full data room processed
Previously 3–4 weeks of associate time. The same 4,000-page data room now has a complete first-pass brief before the first associate meeting.
94%
Risk clause identification accuracy
Validated against manual review across three pilot deals — Claude identified 94% of material risk clauses, with a 2% false-positive rate on standard boilerplate.
3×
Deals run simultaneously
Lexara's due diligence capacity tripled in the six months following deployment — without adding headcount. The constraint shifted from document review to deal origination.
100%
Associate review before delivery
Every brief is reviewed by an associate before it reaches a partner or client. Claude does the reading; the associates do the judgment. No AI output goes out raw.
200K tokens
Entire contracts read in a single pass — no fragmentation
Claude's 200,000-token context window means a 180-page shareholder agreement is read in full without splitting into chunks. This is the primary reason the risk identification accuracy is higher than RAG-based approaches — context is never lost between sections.
What the client said

The 200K context window is what made this work. We tried a RAG-based approach before and the risk identification wasn't reliable — clauses that referenced definitions in a different section were being missed because those sections were in different chunks. Claude reads the whole contract. That's the difference.

SB
S. Blackmore
Managing Partner · Lexara Group
Work With Us

Professionals spending billable time on document triage instead of judgment?

If your team is reading thousands of pages to extract the same information on every deal, every case, or every project — agentic AI changes that equation. Book a free discovery call.

Founders on every engagement. Not a junior consultant.

50+
Agentic AI implementations — Claude, OpenAI, and custom model stacks across industries
200+
Enterprise clients across US, UK, Australia and the Middle East
25+
Years combined experience — senior AI architects on every engagement
0
Failed builds across every agentic AI engagement we've delivered