Legal services
The document shows the position. Revisions show the judgment.
Partner revisions explain why a clause changed. Attorney annotations connect arguments to rulings. Firm-authored playbooks record the reasoning behind fallback positions. Those decision histories give AI teams structured examples for drafting and reasoning evaluation. Expert-authored synthetic matters add hypothetical scenarios with expert scoring.
01Systems you already run
The raw material is already being recorded.
An assessment starts from the software you run today. It asks what those systems hold, never for the records themselves.
- Document management
- iManage, NetDocuments
- Practice management and billing
- Clio, Aderant, Elite 3E, Smokeball
- E-discovery and litigation support
- Relativity, Everlaw, DISCO
- Drafting and contract management
- Litera, Ironclad, Agiloft
- Research and court records
- Westlaw, Lexis, PACER and state court portals
02Dataset opportunities
Where the learnable data sits.
Each opportunity follows a decision from input to outcome. That chain, not the number of records, is what makes data useful for training and evaluating AI.
01
Firm-owned know-how and its revision history
Precedent forms, annotated clause banks, negotiation playbooks with fallback positions, and practice notes the firm authored and owns, with the history of how each changed.
- Practice question
- Partner guidance
- Template and drafting notes
- Revision after a new rule or ruling
- Current house position
Why it is hard to reproduce
Reasoning notes, fallback positions, and revision history can show why a firm's guidance changed. A current form alone does not capture that sequence of expert decisions.
AI use cases
- Drafting and redlining evaluation
- Fine-tuning on expert-authored explanations
- Rubrics for contract-review agents
02
Attorney annotations on public court records
Practicing lawyers reading public dockets, filings, and opinions and recording structured judgments: issues, argument quality, predicted outcomes, and why.
- Public filing
- Issue identification
- Attorney assessment
- Predicted outcome
- Actual ruling
Why it is hard to reproduce
Public records alone do not supply a lawyer's assessment. Creating reasoned predictions and checking them against actual rulings requires expert review and a consistent annotation process.
AI use cases
- Legal reasoning evaluation sets
- Outcome-prediction benchmarks with expert baselines
- Preference data comparing argument quality
03
Draft and revision histories
The path from intake through research, drafts, and senior revisions to resolution, including the reasons for each change.
- Intake
- Issue analysis
- First draft
- Senior revisions
- Resolution
Why it is hard to reproduce
Senior revisions and their reasons capture judgments that a final draft omits. Linking them to a matter's resolution captures the full drafting workflow.
AI use cases
- Post-training on draft and revision pairs
- Long-horizon agent tasks across research and drafting
- Evaluation of legal-assistant work product
04
Expert-authored synthetic matters
Realistic disputes and transactions written from scratch by practicing lawyers, argued on both sides, and decided by attorneys or arbitrators with reasoned decisions.
- Fact pattern drafted
- Submissions on each side
- Neutral review
- Reasoned decision
- Expert scoring
Why it is hard to reproduce
Independently authored scenarios require legal expertise to create realistic facts, competing arguments, and reasoned decisions. Expert grading supplies a rubric and reasoned scores for each hypothetical outcome.
AI use cases
- Agent environments for negotiation and dispute resolution
- Evaluation sets with expert-graded rubrics
- Preference data comparing competing arguments
03Volume is not the test
What is usually not valuable on its own.
Lots of data is not the point. A record of an expert deciding, and of what happened next, is.
Public case law and statutes on their own
Available from other sources. Expert analysis linked to rulings can add a decision and outcome trail that the text alone lacks.
Time entries and billing records
They describe activity and duration rather than the reasoning behind a decision.
Unannotated document collections
Folders of executed agreements or pleadings, without context on what was negotiated, why, or how it turned out, provide little evidence of the decision process.
Publisher forms
Standard forms provide a drafting baseline. Original reasoning notes, revisions, and fallback positions capture the firm's judgment.
04Rights and compliance
The sector rules for the dataset.
Rights reviewed before any outreach.
Last reviewed .
How we handle rights and privacy- 01
Client confidentiality
ABA Model Rule 1.6 is a model; the applicable jurisdiction's adopted rules control. Its confidentiality duty covers information relating to a representation, including information from public sources, beyond privileged communications. Commercial licensing generally requires purpose-specific informed client consent and ethics review.
- 02
Privilege and work product
Attorney-client privilege and work-product protection are distinct from professional confidentiality and have different waiver rules. Work-product protection generally concerns material prepared in anticipation of litigation. Third-party disclosure may waive protection depending on the material, recipient, and jurisdiction. Client consent and de-identification do not automatically preserve either protection.
- 03
Ownership of legal deliverables
Client agreements and outside counsel guidelines sometimes assign ownership of deliverables to the client or restrict their reuse. Check before treating templates derived from client work as firm-owned. Publisher licenses may restrict redistribution and derivatives; original annotations do not automatically confer rights to license the underlying forms or revised versions.
- 04
Professional conduct review
The arrangement needs ethics review of consent wording, supervision of lawyers doing annotation work, and revenue sharing.
- 05
Court record restrictions
A docket may list sealed filings that are not publicly accessible, and accessible documents may contain redactions. Court orders, protective orders, access terms, privacy duties, and third-party rights may limit reuse. Public access alone is not permission to license every document.
05AI use cases
Training examples and evaluation tasks.
Attorney assessments checked against rulings create reasoning tasks with observed outcomes. Draft and revision pairs give drafting assistants examples of expert corrections. Independently authored synthetic matters provide controlled hypothetical scenarios with expert-graded responses.
06Questions
What owners in this sector ask first.
- Which record structures should we assess first?
- Start with firm-authored clause banks and their revision history, attorney assessments on public records linked to rulings, and independently authored synthetic matters with expert scoring.
- What is an expert-authored synthetic matter?
- A hypothetical dispute or transaction written independently of confidential client material and worked by practicing lawyers: facts, submissions from each side, and a reasoned decision by a neutral. Expert grading makes it a structured evaluation task.
- What makes attorney annotations useful?
- Issue classifications, argument assessments, reasoned predictions, and the ruling that followed connect expert judgment to an observed outcome. The annotation process and grading rubric make those judgments comparable.
- What do we need for the first assessment?
- Describe the record types, how expert judgments are captured, the period covered, and whether revisions or outcomes are linked. Start with a dataset description rather than matter documents.
Find the AI opportunity in your records.
Start with a free assessment of your systems, the decisions recorded, and the outcomes linked to them. You keep ownership.