Tutorial 10: Enterprise Deployment & Firm-Wide Adoption

What You'll Learn

This tutorial helps you plan firm-wide deployment of your AI assistant, Claude or ChatGPT: governance, build vs. buy decisions, training, and ROI measurement. IT or management involvement is helpful.

Learning Objectives

By the end of this tutorial, you will:

• Plan enterprise assistant deployment for legal organizations
• Compare build vs. buy decisions (general assistants vs. Harvey/Legora)
• Implement governance and compliance frameworks
• Measure ROI and optimize legal AI investments

Part 1: Deployment Models

Three tiers exist with either vendor: individual accounts, a team plan with shared workspaces and admin controls, and an enterprise plan with identity, retention and audit controls.

Comparison to Competitors

Part 2: Build vs. Buy Analysis

The Core Question

Should you build on a general assistant or buy Harvey/Legora?

Build on a General Assistant: Total Cost Analysis

Direct Costs (100 lawyers, example only):

Team/Business or Enterprise seats: verify current quote
Integration and enablement labor: estimate internal + partner time
MCP/connectors stack: estimate by provider
Training and governance rollout: estimate by cohort size

Buy Vendor Suite: Total Cost Analysis (example only):

Enterprise license quote: verify with vendor
Implementation package: verify included vs. separate SOW
Training/support tier: verify contracted scope
Expansion and add-ons: model in year-2+ forecast

Decision Framework

Choose a general assistant when:

• Cost sensitivity is high
• You want full customization control
• You have technical resources (even minimal)
• Your workflows are unique
• You want to iterate quickly
• Privacy/local processing is important

Choose Harvey/Legora When:

• Budget allows premium enterprise spend
• You want turnkey solution
• Vendor accountability is required
• Standard legal workflows suffice
• You lack any technical resources
• Enterprise support is critical

Hybrid Approach

Many firms are deploying both:

Harvey → High-value, standardized work
General assistant → Custom workflows, cost-sensitive matters

Part 3: Governance Framework

AI Acceptable Use Policy

# Legal AI Acceptable Use Policy
 
## Purpose
This policy governs the use of AI tools (Claude, ChatGPT, Harvey, etc.)
for legal work at [Firm Name].
 
## Scope
Applies to all attorneys, paralegals, and staff using AI
for client-related work.
 
## Permitted Uses
- Contract review and analysis
- Legal research acceleration
- Document drafting (with review)
- Administrative task automation
- Internal knowledge management
 
## Prohibited Uses
- Final legal advice without attorney review
- Sharing client confidential information without safeguards
- Using AI output without verification
- Uploading privileged documents to non-approved tools
- Making representations about AI accuracy to clients
 
## Required Practices
1. All AI output must be reviewed by licensed attorney
2. Citations must be verified in authoritative sources
3. Client data may only be used in approved, secured tools
4. Privilege designations must be maintained
5. AI use must be disclosed per client agreements
 
## Training Requirements
- Annual AI ethics training required
- Tool-specific training before use
- Ongoing education on capabilities and limitations
 
## Compliance
Violations may result in disciplinary action.
Questions: Contact [AI Governance Committee]

Data Classification Matrix

Approval Workflow

New AI Use Case Request
        ↓
IT Security Review (data classification)
        ↓
Legal Ethics Review (privilege, confidentiality)
        ↓
Risk Assessment (client consent, insurance)
        ↓
Approval/Denial + Conditions
        ↓
Implementation + Training
        ↓
Ongoing Monitoring

Part 4: Implementation Roadmap

Phase 1: Pilot (Months 1-3)

Objectives:

• Test the assistant with a select group
• Identify high-value use cases
• Develop initial playbooks
• Assess security requirements

Activities:

• Select 5-10 pilot users (mix of practice areas)
• Deploy Team or Business accounts
• Create 3-5 initial Project templates
• Document use cases and feedback
• Measure time savings

Success Metrics:

• User satisfaction >8/10
• Identified 3+ high-value workflows
• No security incidents
• 20%+ time savings on target tasks

Phase 2: Expansion (Months 4-6)

Objectives:

• Expand to full practice groups
• Build custom skills and playbooks
• Integrate with existing systems
• Develop training program

Activities:

• Roll out to 2-3 practice groups
• Develop firm-specific skills
• Implement MCP integrations
• Create training curriculum
• Establish support processes

Success Metrics:

• 50%+ adoption in target groups
• 3+ custom skills deployed
• Integration with DMS operational
• Training completion >90%

Phase 3: Enterprise (Months 7-12)

Objectives:

• Firm-wide deployment
• Full governance implementation
• Optimization and scaling
• ROI measurement

Activities:

• Migrate to Enterprise plan
• SSO/SCIM integration
• Full audit logging
• Advanced analytics
• Continuous improvement program

Success Metrics:

• 80%+ firm-wide adoption
• Measurable ROI documented
• Zero compliance incidents
• Established center of excellence

Part 5: Training Program

Curriculum Structure

Level 1: Fundamentals (All Users)

• What the assistant is and how it works
• Basic prompting for legal tasks
• Document upload and analysis
• Quality control requirements
• Ethics and compliance obligations

Level 2: Intermediate (Power Users)

• Advanced prompting techniques
• Using Projects effectively
• Legal Plugin commands
• Building personal playbooks
• Collaboration features

Level 3: Advanced (Champions)

• MCP integrations
• Custom skill development
• Workflow automation
• Training others
• Troubleshooting

Training Delivery

Certification Program

Legal AI Certification Path

Level 1: Certified User
- Complete fundamentals training
- Pass basic assessment
- Complete 10 supervised tasks

Level 2: Certified Practitioner
- Complete intermediate training
- Build and share a playbook
- Demonstrate 3+ use cases

Level 3: Certified Champion
- Complete advanced training
- Develop custom skill or integration
- Train 5+ colleagues

Part 6: Measuring ROI

Metrics Framework

Efficiency Metrics:

• Time saved per task type
• Tasks automated vs. manual
• Documents processed per hour
• Research time reduction

Quality Metrics:

• Error rate (before vs. after)
• Revision cycles reduced
• Client satisfaction scores
• Malpractice claims (long-term)

Financial Metrics:

• Cost per document reviewed
• Realization rate improvement
• Write-offs reduced
• Revenue per lawyer

ROI Calculation Template

ANNUAL ROI CALCULATION

COSTS:
Enterprise licenses: $________
Integration costs: $________
Training investment: $________
Internal time (implementation): $________
Total Costs: $________

BENEFITS:
Time savings (hours × blended rate): $________
Reduced outsourcing: $________
Error reduction value: $________
Faster turnaround premium: $________
Total Benefits: $________

NET ROI: (Benefits - Costs) / Costs × 100 = ____%

Benchmarking Data

Use pilot-measured data from your own firm before scaling:

• Baseline cycle time by task type
• Baseline error/rework rate
• Baseline effective hourly cost
• Post-pilot deltas after attorney validation

Example Calculation Framework (50-lawyer firm):

Baseline annual hours for target workflows: ______
Validated reduction after pilot (%): ______
Recovered capacity (hours): ______
Applied value per hour (blended): ______
Total program cost (licenses + implementation + training): ______

ROI = (Recovered value - Program cost) / Program cost

Part 7: Comparing to Harvey/Legora Enterprise

Feature Comparison

| Capability | Claude Enterprise | ChatGPT Enterprise (draft) | Harvey Enterprise | Legora Enterprise | |------------|-------------------|-------------------|-------------------| | Base Platform | | Natural language AI | Claude models | GPT models | Custom legal LLM | Multi-model | | Document processing | Verify current limits | Verify current limits | Vendor-managed | Vendor-managed | | Legal research | Via MCP | Via MCP | Built-in | Built-in | | Customization | | Custom playbooks | Full control | Full control | Limited | Moderate | | Custom workflows | Via skills/hooks | Via skills/hooks | Workflow builder | Workflow builder | | API access | Full | Full | Limited | Limited | | Integration | | DMS integration | Via MCP/connectors | Via MCP/connectors | Vendor-managed | Vendor-managed | | Research databases | Bring-your-own stack | Bring-your-own stack | Vendor-managed | Vendor-managed | | Microsoft 365 | Claude for Microsoft 365 (pricing page) | Verify | Office add-ins | Word add-in | | Security | | SSO/SAML | Yes | Verify | Yes | Yes | | SOC 2 | Yes | Verify | Type II | Type II | | Custom retention | Yes | Verify | Yes | Yes | | Audit logs | Yes | Verify | Yes | Yes | | Support | | Implementation | Self/assisted | Self/assisted | Managed | Managed | | Training | Self/partner | Self/partner | Included | Included | | Account team | Dedicated | Verify | Dedicated | Dedicated | | Pricing | | Model | Per user | Per user | Per user | Per user | | Typical cost | Quote-based; minimum 20 seats | Quote-based | Quote-based | Quote-based |

Decision Matrix

Score each factor 1-5, multiply by weight:

(Adjust weights based on your priorities)

Part 8: Future Considerations

Emerging Capabilities (Roadmaps Change Frequently)

Track these capability areas:

• Expanded platform availability and admin controls
• Better governance/observability tooling
• Deeper document and workflow automation
• Stronger ecosystem integrations
• Faster model and agent iteration cycles

Industry Trends:

• Agentic workflows becoming standard
• Small/specialized legal models
• Real-time collaboration features
• Deeper practice management integration
• AI-human handoff protocols

Preparing for the Future

1. Build flexible architecture: Choose solutions that can adapt
2. Invest in training: AI skills will be essential
3. Document learnings: Create institutional AI knowledge
4. Stay informed: Monitor legal AI developments
5. Engage ethically: Participate in standards development

Final Thoughts

Key Takeaways

1. General assistants support enterprise-grade legal workflows with configurable governance options
2. Customization is your advantage: Build exactly what you need
3. Start small, scale smart: Pilot → Expand → Enterprise
4. Governance is essential: Protect clients and firm
5. Measure and optimize: ROI justifies continued investment

Do This Now

• Complete deployment model assessment (individual vs. Team vs. Enterprise)
• Draft AI acceptable use policy for your firm
• Identify pilot users and 3–5 high-value use cases
• Develop implementation roadmap (pilot → expand → firm-wide)
• Create training plan and success metrics

Related

• How lawyers can use Codex
• Concepts

Tutorial Series Complete!

You've completed the Legal AI tutorial series.

What You've Learned

Next Steps

1. Apply what you've learned to real legal work
2. Share with colleagues and build internal expertise
3. Iterate and improve your playbooks and workflows
4. Engage with the community for new ideas
5. Stay current with Claude and OpenAI updates and legal AI trends

Resources

Claude

• Claude Documentation
• Claude Support
• Claude Pricing
• Claude Enterprise
• Claude Code Documentation
• Admin API Overview

OpenAI

• OpenAI API Documentation
• Codex Documentation

Sources

• Claude Pricing (Team/Enterprise)
• Claude Enterprise: security, governance and administration
• SSO Setup on Team and Enterprise
• SCIM Setup on Enterprise
• Custom Data Retention Controls (Enterprise)
• How to Access Audit Logs (Enterprise)
• Admin API Overview

Additional Reading

• Usage Analytics for Team and Enterprise Plans
• Anthropic Trust Center

Navigation:

• Previous: Skills & Hooks
• Next: Litigation & eDiscovery