Blog
Notes from building NaS_OS.
Workflow economics, AI in sell-side advisory, and the changing shape of corporate finance work.
June 20, 2026
What "verified" should mean when AI touches a deal
In a transaction, "the AI generated it" is not a standard. Here is a concrete bar for what verified AI output should mean in finance: traceable to source, arithmetically checked, and auditable end to end.
June 16, 2026
From data room to working model in an afternoon: rethinking where the analyst adds value
The traditional financial-analysis workflow front-loads days of extraction before any analysis begins. Compressing that front end doesn't replace the analyst — it moves them to the part of the work that was always the point.
June 11, 2026
Why reading the numbers is the hard part of AI in finance
Generating fluent text is the part AI makes look easy. Pulling one reliable figure out of a messy P&L is the part that's genuinely hard — and it's the part that decides whether an AI tool belongs anywhere near a deal.
June 5, 2026
The spreadsheet tax: what rebuilding a model on every deal actually costs
Most mid-market M&A teams rebuild the financial model from scratch on each mandate. We added up what that habit costs in hours, capacity, and lost mandates — and why the number is larger than it looks.
May 31, 2026
Equity story vs. financial model: two halves of the same argument
A great valuation narrative and a defensible model are not separate deliverables. They are one argument told two ways, and buyers are expert at spotting the moment they stop agreeing.
May 24, 2026
The model is the deal: why a financial model should live where the deal lives
On most mandates the financial model sits in a folder, detached from the documents it was built from. When the data room moves, the model goes stale. Here is what changes when the model is attached to the deal itself.
May 17, 2026
How to write a better IM: lessons from running 50+ documents through AI
Five structural mistakes that show up in weak IMs, three patterns that distinguish high-converting ones, and the five-day workflow that replaces the four-week cycle.
May 16, 2026
From 2 weeks to 1 hour: the real economics of AI in M&A boutiques
A breakdown of where the 80+ hours that go into a typical Information Memorandum actually disappear, and what changes when an AI deal analyst does the structural work.
May 14, 2026
Equity story patterns: what makes some firms close more mandates than others
After reading several hundred IMs, the firms that win mandates are not the ones with the best financial models. They are the ones with the most disciplined equity stories.
May 11, 2026
Where AI extraction breaks in financial documents: failure modes by document type
Excel, scanned PDFs, management accounts, cross-document semantics: each fails differently under AI extraction. A practitioner-level map of what actually breaks, and how to prevent it.
May 8, 2026
The four things AI still can't do in M&A (and the three things it does better than humans)
An honest split: equity story emphasis, founder dynamics, fee negotiation, and reading a buyer's bluff stay human. Cross-document reconciliation, fatigue-free output, and full data room recall go to AI.
May 5, 2026
How AI hallucinations actually break M&A deliverables (and how to prevent them)
The failure modes that matter in 2026 are not the fabricated-quote kind. They are extraction errors, source confusion, and silent unit mismatches, all preventable with the right architecture.
May 1, 2026
What I learned watching analysts grind through 16 documents at 2am
The bottleneck in M&A document production is not intelligence. It is coordination, the kind of work human brains are uniquely bad at after hour ten.
April 27, 2026
Why I'm building NaS_OS for boutiques first, not banks
The conventional answer is to go where the budgets are. Here is why category-defining vertical software companies almost always start at the bottom of the market instead.
April 22, 2026
The hidden math of M&A staffing: why junior analysts cost more than you think
Managing partners quote analyst base salary at $110K. The real loaded cost is closer to $366K, and what that $366K is actually producing changes the AI tooling question entirely.
April 17, 2026
Rogo, Hebbia, and the boutique tier they don't serve
Rogo and Hebbia are building excellent AI for finance, but for the institutions that least need help. The thousands of boutiques running mid-market deals are still waiting for a tool built for them.
April 12, 2026
Why the 4-week IM is dead (and what replaces it)
The 6-week Information Memorandum production cycle is no longer a sign of thoroughness. It is becoming a competitive liability. What the new 10-day workflow looks like.