Legal Aid AI Procurement: A 2026 Reference Process

Procurement is often the bottleneck for legal aid clinics adopting AI. This article summarizes a working procurement process based on observed 2024-2025 cycles in several state court systems and legal aid clinics.

Procurement phases

1. Need assessment
2. Vendor landscape review
3. Pilot definition
4. RFP / pilot agreement
5. Pilot execution
6. Evaluation
7. Scale-up or end

Each phase has observable milestones; this article walks through each.

Phase 1: Need assessment

Before shopping, the clinic should answer:

• What specific use case (research memo template, motion draft, document review)?
• What is the volume (queries per month, documents per week)?
• Who are users (lawyers, paralegals, clients)?
• What data sensitivity level (privileged, HIPAA-covered, trade secret)?
• What's the budget?
• What's the timeline?

Without these answers, vendor demos are wasted.

Phase 2: Vendor landscape review

Common AI vendors for legal in 2026:

• Thomson Reuters CoCounsel
• LexisNexis Protege / Lexis+ AI
• Westlaw AI
• Spellbook
• Harvey AI
• Ironclad (contracts)
• CourtGPT.ai
• Custom solutions on Postgres/pgvector

Note: not endorsing any specific product. Verify with current vendor capability assessments before procurement.

Phase 3: Pilot definition

Define a 60-90 day pilot:

• Specific use case
• Specific users
• Specific eval framework
• Specific exit criteria
• Pilot budget cap

Common exit criteria: citation accuracy >95%, latency p95 <2s, hallucination rate <2%, lawyer satisfaction survey >3.5/5.

Phase 4: Pilot agreement

A working pilot agreement should cover:

• Data ownership (clinic's, not vendor's)
• Audit log access for the clinic
• HIPAA / BAA terms if applicable
• Termination for cause
• Performance benchmarks
• Liability for AI errors
• SLA (uptime, response time)

Phase 5: Pilot execution

• Pre-deployment: data audit, bias audit, security review
• Daily: log all queries, monitor latency
• Weekly: eval framework scores
• Monthly: lawyer satisfaction survey

Phase 6: Evaluation

At end of pilot:

• Did the system meet exit criteria?
• What did lawyers learn?
• What broke?
• What would scale?

If met → proceed to scale-up. If not → end (extend with new vendor or alternative solution).

Phase 7: Scale-up or end

Scale-up:

• Negotiate production agreement
• Deploy to all relevant users
• Update privacy / security policies
• Train staff
• Document for future audits

End:

• Document why not
• Retain vendor relationship for occasional use
• Plan alternative path

Procurement pitfalls

Common 2024 failures observed:

• Skipping need assessment, jumping to "AI hype"
• Selecting on demo not empirical eval
• No audit log access for the clinic
• No data ownership clarity
• Underestimating integration costs
• Underestimating change management costs
• Underestimating citation accuracy requirements

Recommended timeline

Total procurement cycle: 4-8 months from need assessment to scale-up.

• Months 1-2: Needs assessment, vendor review
• Months 2-3: Pilot agreement
• Months 3-5: Pilot execution
• Months 5-6: Evaluation and decision

For 50+ lawyer firms or courts, allow 6-9 months.

Acknowledgments

This article summarizes observed procurement cycles as of early 2026.

Dillon Deutsch has worked with state courts and legal aid clinics on AI procurement. https://courtgpt.ai