Why I’m bringing an agentic AI engine room to private equity funds and family offices.
Early in my career I was an analyst at Ernst & Young, working with private equity firms on a problem that sounds simple and never is: finding the cash trapped inside their companies. Working capital, payment terms, balances scattered across accounts, money leaking out of vendor relationships nobody had examined in years. The tools were spreadsheets. The unit of progress was weeks. You would spend a month assembling a picture of one company’s cash position, and by the time you finished, the picture had changed.
A few weeks ago I watched a fleet of AI agents do that same analysis in minutes. Not a demo. Not a prototype. Working systems, running in production, that I built and supervise myself.
That gap, twenty-plus years compressed into minutes, is what this post is about, and it’s why Fern Capital now offers agentic AI transformation to private equity funds, family offices, and their portfolio companies.
What I did in between
I spent the intervening two decades inside money movement. Healthcare payments at Mastercard and Change Healthcare, then eleven years founding and leading Finexio, an embedded B2B payments company. We grew it past $7 billion in annual payment volume across a network of more than 500,000 suppliers, growing by a thousand businesses a month, serving large public universities and integrating into some of the most complex enterprise systems in the country. Our largest shareholder and strategic partner was J.P. Morgan. When the most sophisticated bank in America is your biggest investor and your rails move billions, compliance, security, and trust stop being a page on your website. They become the DNA of everything you build.
The part of the Finexio story I have talked about less is what happened inside the company over the last few years, because I led an AI transformation there before it was fashionable to claim one.
We built an in-house predictive engine to decide which payment method each of 500,000 suppliers should be paid with, replacing manually written Python. Prediction effectiveness improved by more than 80 percent, and overall prediction quality across network routing and payment acceptance rose above 90 percent. That wasn’t cost cutting. That was intelligence driving revenue and network economics.
Then we went after the cost side. We completely eliminated a full-time QA team, replaced by an agentic system. We cut our product development cycle by 60 days by automating requirements gathering and delivery to engineering. We reduced full-time engineering headcount by more than 80 percent while shipping more code at higher quality. We eliminated dedicated data analyst and data scientist roles entirely. In total, a seven-figure EBITDA improvement came out of that work in two years, while output and quality went up. I believe that is far more than most CEOs can claim to have actually achieved, rather than announced.
When I left Finexio this summer, I rebuilt my own firm the same way: parallel coding agents, autonomous content and CRM workflows, a self-hosted stack I supervise from my phone. I built it as a novice in this specific tooling, and it now sits in the top few percent of what I see among my peer set. That taught me something important. This is more achievable, more affordable, and far faster than almost anyone running a fund believes.
It also clarified why this combination is rare. Plenty of people can raise money. Plenty can operate. A smaller number know the financial side of a business inside and out. Very few have done all of that while staying deep and close to the code, building and supervising these systems with their own hands. That intersection is the reason to work with me.
What sponsors are getting wrong
Most sponsors I talk to think AI means cost takeout and cheaper software development. Maybe some basic tooling. That is where the thinking stops, and it misses the actual prize.
The prize is an intelligence layer across the portfolio. Every company conversation, email, call, meeting, and financial feed, standardized into a live picture of what is happening at each company, why, and where. From that layer, the opportunities surface themselves: which workflows to make agentic, which products to build, which cash is trapped and where. Insight flows up to the management team and the board, agentic systems get built against the highest-value targets, and performance gets measured against all of it. Most sponsors do not know this exists. It does, and I run a version of it today.
Here is the uncomfortable part. SaaS is dead, software is effectively free, and you can now pay AI for outcomes instead of paying vendors for seats. But none of that is available to you until the data and intelligence framework exists across your companies. And expecting the management team of a thirty-year-old manufacturing or business services company to build that framework themselves is not a plan. Owners already sense this. The push has to come from ownership, because it will never come as a pull from management.
For a PE fund, the math is unforgiving. You need to return a multiple on capital in a leveraged structure where every dollar of EBITDA is worth five or six at exit, and the hold clock is running. For a family office holding permanent capital in durable, old-economy businesses, the question is different but the answer is the same: run lean, free the trapped capital, and recycle it into the next generation of the family’s investments.
Why the office of the CFO is where I start
Every business touches money. Money in, money out, balances, fraud prevention, bank account security, vendor analytics, working capital management. All of it can be agentic. All of it can be transformed by data. And almost none of it is visible to the owners today, because the cash trapped on a balance sheet, or across an entire portfolio, doesn’t announce itself.
This is the exact problem I worked on as a young analyst with spreadsheets, and the domain I spent twenty years operating in at institutional scale. It is the natural entry point in any portfolio company, in any industry, because it pays for itself first and builds trust for everything after.
How it works
Fern Capital’s Engine Room delivers through dedicated engineering pods under my direction, with senior AI, data, and full-stack engineers. Every engagement runs the same sequence: diagnose the company or the portfolio, design the target operating model, build production systems within weeks, and sustain them with clean handover and maintenance.
There are multiple paths in. Some clients want us to come in, solve specific problems, get paid on the value delivered, and leave their team holding the keys with a light maintenance retainer behind it. Others want Fern as their embedded AI partner: the ongoing transformation and value-creation function for the fund, at a fraction of the cost of hiring full-time operating partners and lead engineers to run it. Both work. The greatest value comes when standardized systems run across the entire portfolio, because every build makes the next one cheaper.
Twenty years ago the tool was a spreadsheet and the answer took a month. Today the tool is an army of agents and the answer takes minutes. If you are responsible for a portfolio and you suspect there is more inside it than you can currently see, you’re right, and I’d like to show you.
Ernest Rolfson is the founder of Fern Capital and the founder and former CEO of Finexio. Begin a conversation.