COHARBOUR

AI that speaks human, writes legal.

Comment

COHARBOUR — AI that speaks human, but writes legal.

Andrew Wilkinson on Podcast

Inspiration

"Legal is a huge inefficiency... when lawyers get involved, it overly complexifies everything. What I'm excited about is when I can get on the phone with the founder and say, hey, let's both open Claude over the next three hours. Let's just like figure this out." - Andrew Wilkinson, Tiny.com

Traditional M&A deals frequently take 6+ months with substantial legal fees because lawyers create adversarial processes. As Tiny.com states: "Selling your company can be brutal... After months of negotiation you're usually left with a bunch of legal fees and no deal."

I was inspired to build the solution Andrew described - a platform where buyers and sellers can "figure it out" through AI-facilitated conversation instead of adversarial legal complexity. The future of M&A should be collaborative, not combative.

What it does

COHARBOUR is voice AI for M&A deals—from messy human conversations into clean legal documents for a $2+ trillion market.

In a space where 70% of deals fail due to communication breakdowns and legal complexity, sophisticated buyers like Andrew Wilkinson and founders deserve better than adversarial legal warfare. They want dream exits, not legal nightmares.

COHARBOUR enables honest voice conversations with AI advisors that transform emotional concerns into structured legal terms. The result: professional Stock Purchase Agreements generated from natural conversations, drastically reducing legal fees while closing deals in days instead of months.

Core Features:

  • Voice-Enabled AI Advisors: Each party gets their own conversational AI that understands M&A terminology and responds through natural voice interactions
  • Full Deal Lifecycle Context: AI advisors maintain complete memory of every conversation, concern, and strategic decision while keeping all information completely confidential
  • Trust Through Separation: Complete data isolation ensures private conversations never cross between parties
  • Intelligent Mediation Engine: Neutral AI powered by legal benchmark leading LLM analyzes sanitized positions from both sides and generates smart recommendations
  • Real-time Collaboration: Live deal dashboard showing progress across 8 core M&A categories
  • Professional Legal Generation: Top-tier Stock Purchase Agreements generated from agreed terms, ready for lawyer review not lawyer creation

How I built it

Three-AI Architecture with Voice Integration

Built with Bolt.new:

  • Guided Conversation Interface: React-based system with assignable conversation prompts and smart recommendation cards for structured deal progression
  • Real-Time Deal Dashboard: Live collaboration across 8 M&A categories showing progress without revealing sensitive information
  • Desktop-Optimized UX: Async, approachable design patterns built for serious business negotiations (desktop only at the moment)

ElevenLabs Voice AI Integration:

  • Natural Voice Processing: Context-aware AI advisors that understand M&A terminology and respond through voice interactions
  • Speech-to-Structure Pipeline: Real-time conversion of complex business discussions into structured legal positions
  • Persistent Deal Memory: Each advisor maintains complete context and conversation history throughout the entire deal lifecycle

Supabase Backend:

  • Trust Through Separation: Secure data isolation ensuring private conversations never cross between parties
  • Real-Time Synchronization: Live updates between buyer and seller dashboards with sanitized position sharing
  • Legal Document Pipeline: Automated generation mapping business agreements to professional contract terms

AI Intelligence Architecture:

  • Advanced Prompt Engineering: Sophisticated conversation analysis that transforms basic discussions into strategic M&A intelligence
  • Legal Expertise Integration: Deep M&A knowledge that identifies negotiable positions and potential deal conflicts
  • Semantic Business Analysis: Context-aware processing that extracts actionable insights from natural language

Technology Stack: React, TypeScript, ElevenLabs Voice AI, Supabase, Tailwind CSS, all deployed through Bolt.new's integrated workflow

Voice Conversation Flow

Challenges I ran into

Trust Problem: How do you enable collaboration while maintaining confidentiality? M&A requires honest discussion but parties can't reveal sensitive information.

  • Solution: Complete data isolation architecture - the mediation engine never sees raw conversations, only clean business summaries generated by private AI advisors. No prompt hacking, no data spillover.

Voice Integration Complexity: Making ElevenLabs feel natural in high-stakes M&A context, not robotic or superficial

  • Solution: Legal-focused prompts that understand M&A terminology and deal structures, plus conversation memory that builds strategic context over time

Real-time Synchronization: Keeping both parties aligned as deal evolves without revealing private thought processes

  • Solution: Supabase real-time updates with careful data flow design that shows progress without exposing sensitive details

Legal Document Generation: Converting messy conversational agreements into professional legal format

  • Solution: Structured mapping system that translates human business concerns into proper Stock Purchase Agreement language

Emotional Processing: Traditional M&A forms can't capture "I'm scared about my team" - but that's often the real deal blocker

  • Solution: Voice AI that processes business concerns and helps transform them into negotiable terms

Accomplishments that I'm proud of

Solved M&A's Fundamental Problem: I feel like the trust dilemma that prevents direct founder collaboration may have been solved - enabling honest conversations while maintaining complete confidentiality through the novel three-AI architecture.

Voice AI Innovation: Most AI M&A tools are "wrappers" - I think this is a genuine business intelligence that processes the human side of transactions. Founders can express concerns verbally and receive strategic counsel that transforms worries into negotiable terms.

Andrew Wilkinson Validation: I admire Andrew as a fellow Canadian, and wanted to build his vision for Tiny. Every feature addresses the specific pain points he described about legal complexity preventing direct dealmaking. It's still early, and this is the worst version.

Technical Architecture: Three-AI system with complete data isolation is genuinely novel - not just another AI interface but a sophisticated trust infrastructure that enables new forms of business collaboration. We can extend this into divorce negotiations, union disputes, etc.

User Experience Design: Conversation-first interface creates satisfying completion experience while making complex M&A accessible to any founder. We also have chat for when voice just doesn't feel right.

Real Business Impact: Drastically reduced legal fees, deals structured in days instead of months, professional documents ready for lawyer review rather than lawyer creation.

What I learned

Voice AI Changes Everything: Traditional M&A forms can't capture much nuance - but voice conversations with business-focused AI can transform that fear into specific retention structures, earnout terms, or cultural preservation agreements.

Trust Through Separation Works: Complete data isolation doesn't prevent collaboration - it enables it. When parties know their private thoughts stay private, they're willing to be more honest about real concerns and priorities.

Founders Need Different Tools: M&A tools are built for lawyers and bankers, not founders. Conversation-based interfaces with business intelligence serve the actual decision-makers better than traditional form-heavy applications.

AI Architecture Matters: The three-AI system isn't just clever - it's the only way to maintain trust while enabling collaboration. Simple AI interfaces can't solve complex human coordination problems.

Market Timing is Perfect: Andrew Wilkinson's quotes prove the market is ready for this approach. Sophisticated buyers want direct founder collaboration but lack the tools to make it trustworthy.

What's next for COHARBOUR

Immediate Focus: Complete the technical foundation at a higher level - polish the voice integration, create a proper onboarding experience, speed up document generation, get authentication going etc.

Personal Goal: Honestly, I had this idea just two weeks ago and built it solo. I want to get some kind of validation from Andrew Wilkinson himself or from the judges that this approach actually solves the problem he described.

Technical Completion: Finish the ElevenLabs voice integration for production deployment, add proper error handling and loading states, and create guided demo flows that help new users understand the three-AI concept.

UX Enhancement: Build better interaction patterns for the smart recommendations - make it easier for users to understand, modify, and negotiate the AI-generated suggestions with intuitive controls and clear feedback.

Market Validation: Test with actual M&A professionals to see if the "trust through separation" architecture actually works in real negotiations, and refine based on feedback from people who do deals for a living.

Long-term Vision: If this gets validation, expand beyond M&A to other high-stakes business negotiations where trust and collaboration currently conflict. Transform adversarial business culture into collaborative problem-solving facilitated by AI.

Live Demo: coharbour.com Repository: Built entirely in Bolt.new - see project URL below Video Demo: https://youtu.be/scDLNDtahjw

Real Scenario: Example Company → Tiny.com Acquisition

The Setup: Sarah built a profitable SaaS company with 50 employees. Andrew Wilkinson wants to acquire it for Tiny.com's portfolio. Instead of 6 months of lawyer emails, they use COHARBOUR.

The Process:

  1. Private Conversations: Sarah tells her AI advisor "I'm terrified they'll fire my entire team after closing - these people trusted me." Andrew tells his AI "I need this team intact, but I can't commit to keeping everyone forever if the business changes."

  2. Position Generation: Sarah's AI creates sanitized summary: "Requires employment protection guarantees." Andrew's AI creates: "Wants team retention with business flexibility."

  3. Smart Recommendations: The Intelligent Mediation Engine suggests: "18-month employment guarantees with performance-based extensions + $2M retention bonus pool distributed quarterly based on team goals."

  4. Negotiation: Both parties can see and refine this specific proposal without revealing their underlying fears or constraints.

  5. Legal Output: Professional Stock Purchase Agreement includes the negotiated retention structure, plus all other agreed terms, ready for lawyer review.

Result: Deal structured in hours instead of months. Legal fees drastically reduced. Both parties got what they actually needed instead of what they thought they had to demand. Coharbour is AI that speaks human, but writes legal.

Built With

Share this project:

Updates