Finance judgment, built into systems that can be checked.

Retained to design and execute a simultaneous acquisition of four companies, two domestic and two international, with a post-close structure that could govern and integrate all of them. Led valuation, diligence management, term negotiation, and legal coordination across four concurrent workstreams, then presented the consolidated rationale to board and investors.

The risk in concurrent closings is not any one deal; it is the assumption that shared across them. One working-capital definition, applied uniformly, would have been wrong for the two international targets. Each got its own.

Architected the merger of two private defense-technology companies: entity architecture, bilateral due diligence, dual-entity valuation, a synergy case built into a combined operating plan, and the capital structure for the merged enterprise. Managed a compliance workstream with a specialized regulatory firm so that existing contract obligations and bidding eligibility survived the transaction.

Synergy cases in defense services fail on contract eligibility, not on cost. The regulatory workstream ran ahead of the financial one for that reason.

Authored the investor memoranda and decks for a managed-services roll-up and issued three simultaneous term sheets. The tuck-in carried a seller note whose principal fell pro rata if cumulative bookings missed target by more than a fifth, secured by a reversion of majority control on default. The cyber target's note adjusted to a bookings target inside a $6.57M floor and $8.03M capBasis: purchase-price adjustment band in the June 2023 term sheet: $7.3M headline, minus pro-rata shortfall to a $13.99M cumulative recurring-bookings target for 2023–24, floored at $6.57M and capped at $8.03M; first-year interest prepaid and reallocated on adjustment., with prepaid interest reallocated on adjustment.

Stand-alone and integration-enhanced projections were kept as separate cases throughout, with synergies shown as an increment. A model that bakes synergies into the base has no way to show an investor what happens if they do not arrive.

Led analysis on a proposed 85% acquisition with a language model as the analysis copilot: LOI issue extraction, lender-term framing, the tax trade-off for the counterparty, and a first draft of the investment memorandum. The draft was fluent and favorable. It treated a 49% stake in the acquirer as a 49% stake in the targetBasis: at 85% acquired, the investor's look-through interest in the target was 49% × 85% = 41.65%. In one review's recomputation the correction moved three-year MOIC from 3.8x to ~2.6x and five-year IRR from 56% to ~29%. The final memo states the 41.65% assumption beneath every returns table., quoted an entry multiple without its period, and computed leverage on the wrong denominator.

The memorandum that went out was the first I delivered as a navigable HTML document rather than a PDF: ten linked sections, animated value-creation waterfalls, every multiple carrying its basis in the sentence.

Nothing about the error was visible from reading the prose. It lived in the joints between documents, and it was found by rebuilding the ownership waterfall from the LOI rather than trusting the model's output tab.

The sponsor asked at the Q4 board meeting; the deadline was the following Tuesday; the ERP-to-CRM integration was incomplete. Claude took the board transcript and extracted the real question, which was the cash-positive crossover rather than growth. A connector failure under deadline was answered with a saved-search export rather than debugging. Three model versions were reconciled in four days and the cover note was rewritten to a register the sponsor would accept.

The model accepted a prior version's numbers as baseline and carried a project the CEO had already killed. Caught by re-reading the actual state of the file, not the narrative around it.

Designed the complete fund economics and an unusual structure in which limited partners co-capitalize a dedicated management company as part of their commitment, so that portfolio monitoring and value creation are funded by the fund itself rather than by fee arbitrage. Full architecture delivered in under twelve months; the GP executed the raise.

The structure works only if the ManageCo's budget is fixed before the first close. Left floating, it becomes a second fee.

Took a private-credit concept to a defined business in about a week, with models doing the drafting and me holding the design: no fixed investor or borrower terms until a specific opportunity is underwritten, so return, tenor, pricing, and draw timing match each asset and the fund never carries return risk it has not priced. Delivered as a deck, a scrollable HTML launch plan, and a standalone HTML financial model.

Six presentation iterations were rejected before the visuals met the standard. A model's first deck is a text dump with headings; the discipline is refusing it.

Policies drafted with the model from the standards themselves, then decided by me. The model proposed a "contracts receivable" contra structure for deferred revenue; I rejected it for standard contract-asset treatment, because a plausible structure an auditor has never seen is not a structure. On the systems side: invoicing migrated off the CRM, a SuiteScript consolidation specified for a contractor when the model's confidence fell below the floor I ship at.

The rule that came out of this: code written by a model ships only above 95% confidence. At roughly 60% you write a specification for a human instead.