OffDeal consolidates 12+ agentic workflows into one Claude Agent SDK-based M&A advisor

Overview

OffDeal, an AI-native investment bank serving founder-led businesses with $1-20 million in annual profit, migrated from a long-context model to the Claude Agent SDK and consolidated over a dozen separate agentic workflows into a single general-purpose agent named Archie. The switch raised internal evaluation accuracy from 25% to 85%, eliminated context-overflow API failures, let each of its four bankers manage 5 to 8 concurrent deals, and produced a buyer-sourcing list that used to cost $12,000 and a week of analyst time for about $200 in compute. The firm closed 8 deals totaling $91 million in its first year with a team of two bankers.

The challenge

Before the Claude Agent SDK, each OffDeal workstream (seller lists, buyer lists, confidential investment memorandums, and more) ran on its own dedicated agentic workflow that did very poorly on problems outside its predefined instructions, and every time a standard operating procedure changed, the corresponding workflow broke. The deepest constraint was context: a single deal can involve more than a hundred million tokens of buyer profiles, past transactions, financial documents and market data, and the team's long-context model from another provider simply failed once API calls hit its million-token limit.

The solution

OffDeal consolidated nearly all of its AI workloads into Archie, a single general-purpose agent built on the Claude Agent SDK, connected to more than 20 data sources and handling work across every phase of a deal via modular, self-contained "skills" that Archie can combine and apply flexibly rather than hardcoded per-workflow logic. A buyer-sourcing skill encodes an experienced banker's roughly 10 sourcing approaches and runs autonomously for up to 4 hours, cross-referencing findings and recursively refining its search, while non-engineers, including a banker, can build new skills like branded presentation generation without engineering help.

Reported business value

Switching to the Claude Agent SDK and then refining MCP design and prompting raised OffDeal's internal eval accuracy from 25% to 85%, and context-overflow API failures dropped to zero overnight. Each banker now manages 5 to 8 concurrent deals and meets 2 to 3 new potential clients per day, a buyer list that used to take a team of analysts more than a week and upward of $12,000 now costs about $200 in compute, and a branded deck that used to take 30-40 hours now takes about an hour. The firm closed 8 deals totaling $91M in transaction value in its first year with a team of two bankers, and its average deal size climbed from $11M to over $20M in 2026.