How One Independent Sponsor Ran 93 Deals With Zero Analysts

An independent sponsor screens dozens of CIMs a year but has no analyst pool: every scorecard, financial spread, revenue-quality read, SWOT, NDA markup, and capital-provider memo has to be produced by the principal, and each one takes hours of manual work that repeats deal after deal. Deal flow arrives faster than one person can evaluate it with institutional rigor, so screening decisions were being made on a partial read, and the diligence products a capital partner expects were only feasible on one or two deals at a time. The tooling that exists for this work is priced and designed for firms with junior staff, not for a sponsor running the whole pipeline alone.

TCG built a governed estate of 42 Claude skill units (22 standalone skills and two plugins) that runs the deal pipeline from CIM intake through the capital-provider package, inside Claude's Cowork desktop environment with the firm's HubSpot CRM, Microsoft 365, and deal folders connected. A single command processes every CIM in the intake folder: it scores the target against TCG's proprietary Intersection Framework, files the CIM and scorecard into a deal folder, renders a brand-conforming scorecard memo, and creates the broker, company, deal, and note records in HubSpot, with a manifest that makes the run idempotent and restartable. Downstream skills take the deal from there: verbatim CIM statement extraction, an MD&A with a formula-driven Excel workbook, revenue-quality and customer assessments, SWOT, deal structuring, population of TCG's deal framework and LBO templates with a three-step model audit, NDA redlines as tracked changes, broker-call prep, and a three-stage capital memo (source-tagged factbase, investment memo, AI-readable knowledge file). The estate is engineered rather than prompted: 115 Python scripts (about 34,600 lines) do the deterministic work, skills share vendored blocks and a common renderer for the TCG design system, and a build tool, regression suite (nine suites, 30 of 30 green on a golden fixture of a real completed deal), description-budget check, and file-checksum drift detector govern every release. A companion plugin, financial-databook, extracts QuickBooks, trial balance, and general ledger exports into a Quality of Earnings databook so red flags surface before a third-party QofE is engaged.

Jack Vander Leeuw runs Token Capital Group as an independent sponsor with no analyst pool, so he built the analyst layer himself. The system started with CIM intake: one command scores every CIM in the intake folder against TCG's Intersection Framework, files it, renders a branded scorecard memo and creates the broker, company, deal and note records in HubSpot, with a manifest that makes runs restartable. Downstream skills cover statement extraction, an MD&A with a formula-driven workbook, revenue-quality and customer reviews, SWOT, structuring, LBO template population with a model audit, NDA redlines, call prep and a three-stage capital memo. The hard part was catching output that looked right and was wrong: an EBITDA bridge that overwrote source values, a cash flow statement that double counted depreciation, liabilities mapped to equity. The answer was verification against the printed source, proof rows that must tie to zero, hard failures instead of warnings, and a regression suite run against a copy of a real completed deal. Anything that had to be identical on identical inputs moved out of the model and into code, about 34,600 lines of Python in all.

Independent sponsors, search funds, and lower middle market PE teams with more deal flow than analyst capacity; PE operating partners who want a governed, auditable pattern for deploying Claude skills across a portfolio company's back office rather than a pile of prompts.