Why Capital Refinery Exists

Before Capital Refinery, I spent a year inside private-markets firms helping investment teams turn AI from a boardroom promise into actual workflow.

A mid-market private equity firm first. A private credit firm after that. My role was not to talk about AI in the abstract. It was to make the tools they already had — Perplexity, Copilot, internal document workflows, research processes, and deal-team habits — useful on real work.

That meant sitting with analysts, associates, operating teams, administrative staff, and senior executives. It meant watching where the tools helped, where they created leverage, and where they quietly made the work more dangerous.

I saw the promise. I also saw the failure mode.

The clarifying moment came in spring 2025. A private credit firm handed me a deal pack — CIM, three years of financials, a rent roll, marketing decks, and an Excel model with broken external references, artifacts from earlier deals, and assumptions copied forward from a prior transaction.

The ask was simple: could AI build the model and produce the research?

The honest answer was no. Not in a way that would survive an IC, a lender, an auditor, or an LP. AI could summarize the files. AI could write convincing prose. AI could pull together external market research. But it could not reliably tell which assumptions had been validated against source evidence and which had been carried forward on faith from another deal. It could not preserve the decision record. It could not prove where every number came from.

The polished output made incomplete diligence look complete.

That was the part I couldn't unsee.

It was not unique to that deal pack. Across private-markets workflows, I kept seeing the same pattern. Analysts became file archaeologists. Associates became spreadsheet plumbers. Partners inherited polished memos whose numbers no longer tied cleanly to the source evidence the team had operated on months earlier.

When the LP asks, “When did you first know?” — when the auditor asks what supported an assumption at IC — the room searches folders. The answer is somewhere in the model, the inbox, the chat history, or someone's local files. It is not structurally preserved in one place that survives the deal moving.

What this is really about

People do not go to business school, train in credit, or spend years inside deal teams because they want to maintain spreadsheets. They do it because they want to think clearly, analyze hard problems, test judgment, and make decisions that matter.

The right systems should give that work back to them.

AI alone does not do that. In many workflows, AI makes the problem worse — because polished output looks finished before the evidence has been adjudicated. The team trusts the prose. The auditor cannot verify it. The LP cannot defend it. The cost arrives months later, when nobody can reconstruct what the team actually knew at the moment of decision.

Capital Refinery was built around the opposite principle:

Every important number needs a trail. Every conclusion needs an anchor. Every assumption needs to remain connected to the evidence that made it defensible. The moment of IC approval should not disappear into a PDF. It should become a living decision record that can be tested as the deal changes.

AI belongs inside that system — but not as the foundation. AI is useful when the evidence is anchored, the definitions are controlled, and the human judgment remains explicit.

The deterministic substrate is the brain. AI is the translator.

Spring 2025 did not start the build. It clarified what I was actually building.

The result is what I needed the day I got that deal pack: a system where the polished output and the evidence underneath it stay tied to each other for as long as anyone needs to defend the decision.

Then the conversations started coming from the other side

The buy-side build came first. The sell-side surface came from the conversations that kept landing in my inbox alongside it.

I am an AI strategist and fractional Chief AI Officer. The work that pays the bills — and that I genuinely love doing — runs through Enterprise Refinery: helping operating companies, SMB owners, and the consultants and AI shops working with them figure out where AI and modernization actually move the business. Different rooms, same conversation, repeating week after week:

A founder spends six figures on an AI workflow build and cannot tell whether anything in the business is actually different. A CFO modernizes the close cycle and has nothing to hand a potential buyer that proves it. An operator hires a fractional team to clean up reporting and ends up with a prettier dashboard and the same diligence risk. A consultant builds something impressive and a year later cannot demonstrate the engagement changed the company’s institutional posture in any way a buyer would credit.

The pattern was the same every time. Real money spent on real work, with no framework for measuring whether it landed. People picking partners blind because there is no neutral artifact to point at. The consultants frustrated for the same reason — they knew the work was good and could not prove it in language a buyer or lender would credit.

That is the realization the sell-side surface came from. The same 10-axis engine that grades whether a PE deal would survive institutional review grades whether an operating company would survive that same review — same axes, same evidence discipline, same refusals. The only thing that changes is who’s reading the artifact and what they want to do with it: a buyer’s IC, a board, a lender’s credit committee, a partner’s engagement renewal.

Capital Refinery now runs three doors from one spine: investment teams underwriting with evidence, portfolio operators tracking institutional durability over time, and owners + advisors + consultants preparing for sale, financing, modernization, or board review. The Modernization Impact Review and the Readiness Gap Review came directly from those conversations — partners and operators who needed an independent, evidence- backed measurement layer to attach to their work.

The methodology is the same in all three rooms. The discipline is the same. The refusals are the same. What changes is the audience the artifact is written for.

The smaller belief

People did not train in finance to chase broken spreadsheet links.

If a system makes the work that drew them into the field harder to do — and the work that drove them out of the field easier to fake — that system is failing.

Capital Refinery is built on the opposite bet: that the right tools should remove the clerical burden so the humans in the room can do the work only humans should do.

Challenge the evidence. Weigh the trade-offs. Decide.

— Duaine McDonald, Founder