EDMONTON, AB — February 28, 2026 — Napkin Deals, the AI-powered deal marketplace, today published Chapter 4 of its forthcoming book, “How I Built a Global AI Organization,” detailing the company's first real deal qualification: FlowSync, a $12M SaaS company with $2.2M in annual recurring revenue, 118% net revenue retention, and 76% gross margins.
The chapter documents what happened when Napkin Deals' AI infrastructure faced a live transaction for the first time. FlowSync represented an ideal test case — a well-run vertical SaaS business with strong unit economics and a motivated seller. The deal entered the pipeline with a $12M asking price, and the AI workforce was tasked with full qualification: financial analysis, buyer matching, risk assessment, and initial outreach.
Eleven Failures, Zero Reasoning Errors
Over the course of the FlowSync qualification, Napkin Deals recorded 11 distinct infrastructure failures in sequence. API timeouts, message queue drops, database connection pools exhausting under load, webhook delivery failures, and authentication token expirations all surfaced during the process. Every failure was a plumbing problem — a systems integration issue that could be diagnosed and fixed. Not one failure was attributable to AI reasoning.
The distinction proved foundational to the company's engineering philosophy. When AI agents failed, they failed because the infrastructure underneath them broke — not because the agents made poor decisions. This insight reshaped how the engineering team prioritized its roadmap, shifting resources from model fine-tuning to infrastructure reliability.
S1 Bot Identifies a $200K Revenue Gap
During the qualification process, the S1 analysis bot — an AI employee named Meera Krishnan — independently identified a $200K discrepancy between FlowSync's reported revenue and its bank deposit records. The gap was not flagged by the seller, nor was it visible in the summary financials. Meera surfaced it by cross-referencing monthly bank statements against the company's revenue recognition schedule, a task that would typically require several hours of manual analyst work.
The finding demonstrated that AI agents operating with access to source documents could perform substantive financial analysis, not just data summarization. It also validated the decision to give AI employees access to primary records rather than pre-processed summaries.
Two Bots, Two Models, One Conversation
The FlowSync deal also proved a critical architectural hypothesis: that two AI agents running on different large language models could communicate reliably through the company's NATS messaging layer and Supabase persistence layer. The CEO bot set five conditions for deal advancement and imposed a 48-hour deadline for the seller to respond. The S1 bot, operating on a separate model, received those conditions, validated them against its own analysis, and prepared the compliance documentation — all without human intervention.
“FlowSync taught us the most important lesson in AI operations: the intelligence works. It's the plumbing that breaks. Every failure we recorded was a timeout, a dropped message, a connection pool that ran dry. The AI never made a bad judgment call. Once we understood that, we stopped worrying about whether the bots could think and started making sure the pipes didn't leak.”
— Matthew Liberto, Founder & CEO, Napkin Deals
Implications for the AI Deal Model
The FlowSync qualification established the operational baseline for every subsequent deal in Napkin Deals' pipeline. The 11 infrastructure failures were catalogued, root-caused, and resolved. The bot-to-bot communication protocol was hardened. And the revenue gap identification became a standard capability built into every S1 analysis going forward.
Chapter 4 is part of “How I Built a Global AI Organization,” a forthcoming book by Matthew Liberto documenting the construction of an AI-native deal marketplace from first principles.
- Previous: Chapter 3 — Critical Gaps
- Next: Chapter 5 — The Money Layer
- Related: How an AI Sales Force Actually Works
About Napkin Deals
Napkin Deals is an AI-powered deal marketplace headquartered in Edmonton, Alberta with operations in Miami, Florida. The company maintains a live pipeline of 3,600+ qualified deals representing over $4.9B in enterprise value across 15+ countries. Napkin Deals operates with a hybrid workforce of 38 human professionals and 69 AI employees governed by its published AI Hiring Doctrine.
Media Contact
Press inquiries: [email protected]
Website: www.napkindeals.com