A platform for pricing private companies. It takes the documents a seller actually has, the bookkeeping file, the tax return and the owner's own model, and reconciles them into one defensible valuation with every figure traced back to the page it came from. It began as a question asked in class at Columbia Business School in September 2023.

What is the Columbia DSI capstone?

A term-long project in which a team of Columbia data science students works on a real problem brought by an outside organisation. In Fall 2026 one of those teams is working with Value Alpha on private company valuation data. The work is covered by confidentiality agreements, so the specifics stay with the team.

Can you build a company while doing an MBA?

In my experience the useful version is not splitting your time between the degree and the company. It is bringing one real problem to every class and letting each subject attack it from its own angle. Statistics, corporate finance, operations and strategy asked very different questions about the same problem, and the answers compounded because they were about the same thing.

Why are small private companies so hard to value?

Because the same business produces different earnings figures depending on which of its own documents you read. The bookkeeping file, the tax return and the owner's model are each prepared for a different purpose and rarely agree. Public companies have audited statements and a quoted price. A company worth a few million has neither, so price is often settled by anecdote.

Building It

The problem I took to every class is now somebody else’s capstone

Three years ago I was the student bringing one stubborn question to every subject. This term a Columbia team is working on the same question, and I am on the other side of the table.

In September 2023 I started at Columbia Business School with one question I could not put down. Why is a company worth a few million dollars still priced by anecdote?

I did not have a plan to build anything. I had a question that would not leave me alone, left over from years of pricing companies where the numbers were audited, the comparables were public and the arguments were at least about the right things. Then you look at the other end of the market, where most businesses actually live, and almost none of that exists.

So I did the only thing that seemed useful. I took the same question to every class.

One problem, four angles

This is the part I would tell anyone thinking about doing a degree while building something. The temptation is to split yourself, to treat the programme and the company as two claims on the same hours. That is the version that leaves you doing both badly.

What worked was the opposite. One problem, carried into every subject, and each subject allowed to attack it in its own language.

Statistics asked what the sample actually was, and whether the thing I called a market was a market or a convenience sample of whatever data I could reach. Corporate finance asked which method survives when the inputs are unreliable, which is a different question from which method is correct. Operations asked where the work actually goes, and the answer was not the arithmetic, it was reconciling documents that disagree. Strategy asked the only question that really matters for a company like this: if the data is so hard to assemble, who else is going to bother?

None of those is a business plan. Together they were the outline of one, and they compounded because they were all pointed at the same thing.

What the question turned into

By the autumn of 2024 it was concrete enough to be a finalist in the Columbia AI Startup Challenge. It has a name now, an office at the Columbia Startup Lab, a team, and a set of answers I did not have in that first term.

The shortest version of what I learned: the hard part is never the valuation arithmetic. The hard part is that a single company routinely produces three different earnings figures from its own paperwork. The bookkeeping file, the tax return and the owner’s own model are each prepared for a different purpose, by different people, with different incentives, and they rarely agree. Everything downstream, the multiple, the offer, the loan, is built on whichever one somebody happened to open first.

That is not a technology problem in the way people expect. It is a reconciliation problem, and it is unglamorous enough that very few people want to do it.

The part that surprised me this month

This term, a team of Columbia data science students is working on that same problem as their capstone, with us. We meet weekly. The work sits under confidentiality agreements, so the specifics stay with the team, and that is as it should be.

What I did not expect was how strange it would feel to be on the other side of it.

Three years ago I was the one carrying a question into a room full of people better qualified to answer it than I was. Now I am the one bringing the question, and watching a group of people pick it apart with tools I do not have. They ask things I stopped asking a year ago because I had quietly decided I knew the answer. Several times now I have been wrong about that.

If you have been living inside a problem for three years, the most valuable thing you can do is hand it to people who have not. They have no investment in your previous conclusions. That is the whole point of them.

The unglamorous parts, since this is supposed to be honest

A few weeks ago I worked out the cost of attending an independent sponsors summit here in New York. It is the densest room of exactly the people this product is for. The pass for a platform like mine was $3,499. I did not go.

That is the actual texture of this stage. The company is pre-revenue. The rooms where the buyers are have a price on the door, and you say no to most of them and find the ones that do not. There is a version of a founder essay that leaves that part out. It is not a useful version.

The other unglamorous part is cadence. A standing research meeting every week since the summer. A bi-weekly call with my co-founder. A capstone session every Thursday morning. None of it looks like progress on any given day. All of it is the only reason there is anything to show after three years.

Why I am writing this down

Partly because the arc amused me and partly because of what it says about the original question.

A question is worth carrying if it survives contact with people who are not you. Mine has now survived four academic subjects, a startup competition, a few thousand valuations, and a room of students who were in secondary school when I started pricing companies. It is still the same question, and it still does not have a good answer in the market.

Why is a company worth a few million still priced by anecdote?

I am less annoyed by it than I was in 2023. I am more convinced it matters.

Go deeper

If you are pricing a business, buying one, or selling one, and you want the earnings reconciled before somebody else does it for you, that is the work I do.

Book an advisory session Try Value Alpha The Valuation Index

Tomasz Felpel is an investor, founder, and advisor in private markets, based in New York. He publishes the Felpel Private Company Index and has priced more than a thousand private companies.