Finance judgment, built into systems that can be checked.

I am Dovid (Zusha) Gurevich. Over twenty yearsBasis: engineering program management from 2000; investment platform from 2006; family-office CFO/CIO from 2009; operating CFO roles 2018–2026; advisory practice concurrent since 2009. I have moved capital as a banker, run it as a CFO, and lately taught machines to do the analyst's share of both. $750M+Basis: aggregate transacted capital across 15+ M&A advisory mandates, 2009–present, per the confidential transaction addendum. Includes a four-company simultaneous acquisition ($50–100M) and a $250M+ defense-sector merger. transacted, $250–500MBasis: capital raised or structured across 10+ engagements: equity (seed through growth), convertibles, venture debt, credit facilities, fund formation, PIPEs, and one reverse-merger IPO. Range reflects instruments counted at commitment versus close. raised, and a $2BBasis: single-family office where I served as CFO and CIO, 2009–2017; legacy direct investments redeployed over a 5–7 year horizon into a vertically integrated healthcare and medical real-estate platform. family office rebuilt into an operating platform. Since mid-2023 the work has been done with AI as the analyst layer, under written rules, and I keep the record of where it failed.

Every figure on this page carries its basis. Hover or tap one.

Career

The arc runs engineering to institutional investing to operating finance, with an advisory practice alongside for sixteen years. Each stop added a layer the next one used.

  1. National Technical Systems

    Program manager, telecommunications compliance and R&D

    A $300MBasis: revenue of the publicly traded parent, per the CV; testing and engineering services across aerospace, defense, telecom, and automotive. public engineering-services company. Ran four project managers and up to seventy engineers on $10M+Basis: annual project volume of the division, per the CV. of annual project volume across the US, Korea, and Vietnam. Raised division margins ten points through process automation. The engineering habit of writing the spec before the build never left.

  2. Knowledge Universe Holdings

    Vice President, M&A, strategy and operations

    The Milken and Ellison investment platform: $500MBasis: initial capital of the platform, per the CV; 50+ portfolio companies, $1.6B combined revenue, $2.25B peak valuation. initial capital, fifty-plus portfolio companies. Recruited as the technology diligence specialist, ended up leading K-12 deal origination, structuring, and portfolio monitoring. Instrumental to $500M+Basis: direct investments across minority and change-of-control EdTech transactions in which I authored memoranda, built valuation models, and managed diligence, per the CV; separately contributed to $350M in portfolio exits. of direct investments and $350MBasis: portfolio liquidations and exits I contributed to during the period, per the CV. of exits.

  3. Single-family office

    Chief Financial Officer and Chief Investment Officer

    Retained to turn a portfolio of underperforming direct investments into a generational platform. Liquidated legacy positions, redeployed over five to seven years into a vertically integrated healthcare and medical real-estate enterprise across fifteen-plus direct investments, and built it to $1.7B in revenue at 22% EBITDABasis: revenue and EBITDA margin of the integrated healthcare and real-estate platform at the end of my tenure, per the CV. Entity is a private family office; not named.. Created a captive credit fund, designed the ownership and entity architecture for asset protection and wealth transfer, and consolidated finance, IT, and HR into a shared-services center that cut functional spend 25%Basis: annual reduction in cumulative functional spend across portfolio companies after consolidation, per the CV. a year.

  4. Mursion

    Chief Financial Officer

    An AI and VR enterprise training platform, joined pre-institutional. Raised $56M+Basis: Series A $2M, A-1 $7.5M, Series B primary $20M and secondary $26M, per the CV; valuation step-up of 10x across the rounds. Plus $1.5M venture debt and an AR facility converted to a $2.5M term loan. across four rounds with a 10xBasis: valuation step-up from Series A to Series B, per the CV. step-up, moved the revenue model from managed services to SaaS, and took the company through three consecutive clean audits from a standing start.

  5. Tynrose

    Board member and Chief Financial Officer

    A PE-backed managed-services platform formed by carve-out and two acquisitions. Built finance from zero across three entities, ran the deal side from the first year, and delivered the board-mandated turnaround model that took the levered cash gap from ($1.3M) to ($293K)Basis: cumulative levered cash gap in the full-year cash-basis model, before and after the Cost Optimization and Profitability Improvement Plan ($2.434M annualized savings across five initiatives), FY2025–26. Company scale at the time: ~$17.5M revenue, ~$11M ARR, 257 accounts.. This is where AI became the analyst layer; several of the cases below come from it.

  6. Holding Advisory

    Principal, M&A advisory and fractional CFO

    Buy-side, sell-side, mergers, capital formation, and interim finance leadership for PE and VC funds, family offices, and corporates across technology, healthcare, government services, real estate, and education. 40+Basis: 15+ operating and governance engagements, 15+ transaction advisory engagements, 10+ capital-formation engagements, per the CV. engagements. The practice is also where STELL, the finance operating system below, was designed.

Selected work

Counterparties are unnamed; structures and numbers are real. Each entry ends with what would have gone wrong without a second look, because that is the part that transfers.

Buy-side M&AFour companies acquired at once, two of them abroad
Client
Mid-market SaaS company
Size
$50–100M aggregate
Instruments
Stock purchases, earn-outs, convertible preferred
Outcome
All four closed

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.

Strategic mergerTwo defense-technology firms, one regulatory workstream
Size
$250M+
Scope
Merger structure, bilateral diligence, dual valuation, synergy plan, shared-services design
Outcome
Closed with full regulatory approval

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.

Platform formationThree term sheets, one financing plan, first year as CFO
Targets
$4.4M to $25.5M purchase price
Structures
Rollover, seller notes, performance-contingent price adjustment inside floor and cap
Financing
$59.5M tranched equity plan; secured-debt alternative modeled alongside

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.

Diligence with AIAn 85% acquisition at $42.5M, and a memo whose numbers did not tie
Target
Cybersecurity-services company
Debt
$10.7M term loan, SOFR + 3.25%, 1.35x DSCR
Tax
F-reorganization with $1.4–2.0M estimated state-tax cost
Tools
GPT for synthesis and drafting; recomputation by hand

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.

TurnaroundA zero-growth cash model, due Tuesday
Company
PE-backed MSP, ~$17.5M revenue, three entities
Ask
Three-month cash projection at zero new-customer acquisition, actions mapped to cash
Output
$2.434M annualized savings; levered cash-positive crossover modeled

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.

Fund formationA research fund whose LPs also capitalize the manager
Vehicle
Multi-vehicle investment platform, therapeutics and devices
Size
$50–100M
Scope
Fee and carry, call and distribution waterfall, LP documents, GP operating model

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.

Private creditA lending platform that sets investor terms after it finds the loan
Client
UK defense-sector group
Design
Luxembourg fund; 2% management, ~2% origination, 0.75% servicing, 15% over 8% hurdle
Operating model
One payroll line, 90% parent support, break-even near $41M AUM

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.

Operating financeAccounting policy drafted from the standards, and a structure I rejected
Scope
ASC 606 deferred revenue; ASC 705-20 vendor consideration; commission clawback; AP cadence
Systems
NetSuite from zero under carve-out transition terms

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.

Systems I built

Twenty-five repositories, most of them private, most of them TypeScript and Python written with Claude Code to specifications I author. Three matter here.

STELL, a finance operating system

General-purpose AI is not a finance tool by default. STELL is the layer that defines the deliverable first, a cash model, a lender diligence package, a board note, a variance bridge, and routes reasoning, data access, and verification toward it. A planning engine holds the modeling methodology as executable rules; an intelligence layer holds the market corpus; agent routing decides which model does which step and what may leave the perimeter. Built for my own practice. Module one, STELL-Finance, is in active build.

A decision method that refuses to advance on silence

Packaged as a Claude skill and used to build STELL itself. Before any solution: name the entities, who owns each decision, what changes on create, update, and delete. Work moves through gated phases; a check that could not run must say so, and a missing result is a failure, not a pass. Every decision has one owning record, every derived claim traces to a source that exists, and every deviation during the build is logged and reconciled at the end. It is the discipline of tying a number to a ledger entry, applied to reasoning.

The evaluation tooling around them

A financial-modeling skill with a calibration workbook and a named failure-mode list, tiered from projection through valuation. An adversarial review that argues five fixed positions against a decision before synthesis. A 137-companyBasis: HA-AIE-Corpus, a competitive-intelligence system over the AI-enablement market: ingestion, data model, and scoring built with Claude Code; deployed on a VPS behind access control, gated to two users. competitive-intelligence corpus behind access control. A knowledge vault that captures every working session automatically and indexes it into projects, decisions, and reusable patterns, so the next session starts where the last one ended. And the practice's own surface: holdingadvisory.com, designed and shipped as a static site the same way this one is, with HAPE, an engagement engine that drafts and posts the practice's content to LinkedIn and X on zero ad spend, running behind it.

How I work with models

Written rules, applied on every task. Each exists because something went wrong without it.

Research

Foundations of quantum mechanics, pursued as an independent researcher with the same verification discipline. Three papers; the first is public.

Contact

[email protected]
linkedin.com/in/dovidgurevich
github.com/zusha4ever
holdingadvisory.com
Cleveland, Ohio

Selected transaction detail is available under NDA. Case studies on this page omit counterparty names by design. Elsewhere: a small horology venture, and a slow habit of writing things down.