Contract Clause Extraction at Scale: When AI Beats Manual Review

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-legal-AI-contract-clause-extraction-adults

1. A law firm plans to use AI to extract termination clauses from 3,000 vendor agreements. What is the MOST critical preparatory step before running the extraction?

   1. Define a precise taxonomy of what constitutes a termination clause, including edge cases
   2. Deploy the extraction immediately on the full portfolio to maximize time savings
   3. Train the AI model on the firm's own proprietary contract data first
   4. Hire additional paralegals to review the AI's output in real-time

2. Which outcome is MOST LIKELY if a firm attempts AI clause extraction without first establishing clear definitions for each clause type in their taxonomy?

   1. Extraction speed will increase but accuracy will vary unpredictably across clause types
   2. The firm will need fewer attorneys to review the results
   3. The extraction will fail because the contracts lack standard formatting
   4. The AI will generate new clauses to fill gaps in the contracts

3. A legal team uses AI to extract change-of-control clauses from 50 acquisition agreements. The AI returns 47 clauses with source quotes. What should the team do NEXT to validate the extraction?

   1. Run the extraction again with different prompt parameters
   2. Accept the results immediately since 94% extraction rate is acceptable
   3. Send the results directly to the deal team for strategic review
   4. Randomly sample and manually verify 5-10% of the extracted clauses against source documents

4. Why is it important for AI-extracted clauses to include the source-document quote from the original contract?

   1. It allows the AI to learn from its mistakes in future extractions
   2. It provides defensibility when the extraction is reviewed by opposing counsel
   3. It reduces the storage requirements for the extracted data
   4. It enables the AI to extract clauses from scanned PDFs

5. A junior attorney asks why the firm can't simply use AI to identify all material adverse change (MAC) clauses across the portfolio without human review. What is the MOST accurate response?

   1. The AI will generate new MAC clause language if existing ones are poorly drafted
   2. AI can reliably identify MAC clauses because they are standard provisions in all contracts
   3. The AI requires all contracts to be in PDF format before extraction can begin
   4. AI extraction will be inconsistent if the firm hasn't defined what constitutes a MAC clause in their taxonomy

6. When should a clause extraction protocol include an escalation workflow?

   1. After the 5-10% QA sample has been fully verified
   2. Only when the AI fails to extract any clauses from a document
   3. When extraction results include ambiguous or low-confidence clause identifications
   4. Whenever the contract portfolio exceeds 1,000 documents

7. A deal team is analyzing strategic provisions like MAC clauses and exclusivity agreements in a pending acquisition. Can AI fully replace their review of these clauses?

   1. Yes, AI can extract all strategic clauses with higher accuracy than attorneys
   2. No, AI cannot replace the deal team's interpretation of strategic clauses
   3. No, but only because AI cannot extract clauses from handwritten addenda
   4. Yes, but only if the AI model was trained on similar acquisition agreements

8. A firm wants to use few-shot examples in their AI extraction prompts. What is the PRIMARY purpose of including these examples?

   1. To enable the AI to extract clauses from contracts written in foreign languages
   2. To automatically format the extraction output into a table
   3. To reduce the amount of computing power the AI requires
   4. To anchor the extraction to the firm's specific interpretation of clause types

9. Which statement best describes what AI clause extraction can accomplish that manual review cannot match at scale?

   1. AI can generate new clause language to fill gaps in poorly drafted agreements
   2. AI can determine whether extracted clauses are favorable or unfavorable to the client
   3. AI can extract clauses from a 5,000-contract portfolio in hours rather than weeks
   4. AI can interpret the legal meaning of ambiguous contractual language

10. An AI extraction returns a clause labeled as 'force majeure' from a contract, but the source quote shows it is actually a 'impossibility' provision. What went wrong?

    1. The contract was written in an unsupported file format
    2. The extraction protocol lacked sufficient few-shot examples distinguishing force majeure from impossibility
    3. The AI hallucinated the clause because the contract was scanned poorly
    4. The AI correctly identified the clause but mislabeled it due to taxonomy confusion

11. Why is the deliverable format an important consideration in clause extraction protocol design?

    1. To meet billing requirements for time tracking
    2. To ensure the extracted data can be imported into the legal team's existing workflow systems
    3. To reduce the number of source quotes that must be stored
    4. To enable the AI to extract clauses from more document types

12. A regulatory inquiry requires a law firm to produce all contracts containing indemnification clauses from the past five years. The firm uses AI extraction. What is the appropriate workflow?

    1. Run extraction, sample-verify 5-10%, then have attorneys review for completeness and accuracy
    2. Deliver the AI results directly to the regulator without review
    3. Submit only contracts where the AI found indemnification clauses with 100% confidence
    4. Hire a third-party vendor to re-validate the AI extraction

13. What distinguishes clause extraction from other forms of contract analytics?

    1. Clause extraction maps contract language to a defined taxonomy of clause types
    2. Clause extraction focuses on structured data while other analytics focus on sentiment
    3. Clause extraction requires human reviewers while other analytics are fully automated
    4. Clause extraction only works on contracts written in English

14. An attorney claims the AI 'understood' that a particular clause was unfavorable to the client based on the extraction results. What is the flaw in this interpretation?

    1. AI cannot evaluate whether clauses are favorable or unfavorable—that requires legal judgment
    2. The source quote was missing, making verification impossible
    3. The extraction was run on the wrong clause taxonomy
    4. The AI should have flagged this as an ambiguous extraction requiring escalation

15. A firm has 10,000 contracts to review for a portfolio acquisition. Using AI clause extraction, how long should the initial extraction phase take compared to manual review?

    1. A few hours compared to weeks for manual review
    2. Longer than manual review due to setup requirements
    3. Approximately the same time as manual review
    4. It depends on the file formats of the contracts