The people pricing your software company have never opened it

I price private companies for a living. More than a thousand of them, and before that I spent my career in Fortune 500 corporate development working on transactions up to roughly $26 billion, on the side of the table that decides what a business is actually worth to a buyer. I also write software. Those two facts sit oddly together on a CV, and for most of my career I treated them as unrelated. They are not. On an online or software business they are the same job.

Here is the problem in one sentence. When you value a manufacturer, the assets sit in a building and a surveyor can walk around them. When you value a software company, the asset is a running system that almost nobody in the valuation chain will ever open.

The two questions, asked separately by people who cannot answer both

A seller going to market gets two kinds of professional attention. Financial people build the model: normalise the earnings, strip the owner add-backs, apply a multiple drawn from comparable transactions, produce a range. Technical people, if anyone hires them at all, read the code and report on architecture, dependencies and security.

These two reports almost never meet. The financial report treats the product as a black box that emits revenue. The technical report treats the money as somebody else’s problem. And the buyer, who is trying to decide what to pay, has to fuse them alone, usually in the two weeks between a letter of intent and a financing deadline.

That gap is not academic. It is where the price moves. A recurring revenue line looks identical in a spreadsheet whether it is produced by software a customer cannot leave or by a thin layer over a commodity API that a competent developer could rebuild in a fortnight. The model cannot tell the difference. The buyer can, eventually, and the discount arrives late in diligence when the seller has the least leverage.

What a multiple actually encodes

In my own transaction data, covering 7,601 recorded small-business sales, verified-sold deals clear at a median of about 2.58 times seller’s discretionary earnings, while businesses currently listed for sale are asking about 3.26 times. Two things are worth saying carefully about that pair, because it is the kind of number that gets quoted badly.

First, those are two different populations. The asking figure comes from live listings, the clearing figure from completed sales. They are not the same deals measured twice, so the comparison describes a market, not a transaction. Second, the size of the gap depends entirely on which number you put underneath. Expressed as the premium sellers ask over what buyers pay, it is about 26 percent. Expressed the way brokers usually express it, as a discount from the asking price, the same gap is about 21 percent. Both are arithmetically correct and they are not interchangeable. Anyone quoting one against the other without saying which is being sloppy, and I have been sloppy about it myself.

What matters is the shape underneath. That spread is not evenly distributed. It is not a market-wide haircut applied to everyone. It concentrates in the businesses where a buyer discovers something in diligence that the listing did not disclose, and on software businesses the thing they discover is usually in the product.

The four things a technical reader checks

When I open a product I am not doing a code review in the engineering sense. I am not counting style violations. I am answering a commercial question in four parts, and each part has a price attached.

Is the thing defensible, or is it a wrapper? A great deal of what is currently sold as an AI company is a prompt, an API key and a payment form. That is not a criticism of the business, which may be profitable and well run. It is a statement about what a buyer is acquiring. If the capability can be reproduced by a competent team in a month, the buyer is paying for the customer list and the brand, not for the software, and the multiple should reflect that. Whether the models are self-hosted or called, whether the data is proprietary or public, and whether there is anything the acquirer could not simply rebuild are now standard diligence questions.

What does it actually cost to run? Owner-operated software businesses frequently understate infrastructure, because the owner has been absorbing costs personally, running on credits, or sitting on a legacy pricing tier that will not survive a change of control. I have seen operating margins move several points on a careful read of a hosting bill. That flows straight into earnings, and earnings times a multiple is the price.

Where is the revenue concentrated, and is it contracted? This one is financial on the surface and technical underneath. The question is not only what share of revenue comes from the largest customer. It is whether the integration that customer depends on is deep enough that leaving is painful, or shallow enough that leaving is a Tuesday afternoon. You cannot answer that from an invoice.

Who actually built it, and can it be maintained? If the product was built by an agency that no longer exists, in a framework that has stopped receiving updates, by one contractor whose name appears in every commit, then the buyer is not acquiring a going concern, they are acquiring a rebuild with a customer list attached. That is a real business and it has a real price. It is not the price on the listing.

What a finding is worth in dollars

The reason to do this before you go to market rather than during diligence is that the same fact costs radically different amounts depending on who finds it and when.

A weakness you find yourself, and fix, costs the price of fixing it. A weakness you find yourself, and disclose without fixing, costs a small negotiated adjustment and buys you credibility on everything else you said. A weakness the buyer finds costs delay, and delay in a deal has a way of becoming money. And a weakness the buyer finds that you knew about and did not mention stops being about the weakness at all. It becomes a question about whether the other forty things you said can be relied upon.

That ladder is the entire argument for doing the work early. It is the same argument I would have made from the buy side, and it is the argument I made for years, just pointed the other way.

How to read your own before somebody else does

You can do a meaningful version of this yourself, and you should, before you pay anyone. Open your own product with a hostile eye and write down the answer to four questions. What here could a competent team rebuild in a month? What does it truly cost to run, at a price a new owner would actually pay? Which customers could leave without much pain, and what share of revenue do they represent? And if the person who built this stopped answering the phone tomorrow, what would happen?

Write the answers down honestly, then price them. Not in adjectives, in dollars off the offer. That document is worth more than a valuation certificate, because it is the document that determines what you actually receive.

The uncomfortable part is that most sellers cannot write it, not because they lack the ability but because they built the thing and cannot see it as a stranger would. That is the whole job. Somebody has to look at your software the way a buyer will, and price what they find, and those two skills almost never sit in the same person.

Common questions

Do you need to open the code to value a software company?

For a small online or software business, yes, in practical terms. The financial model cannot distinguish between revenue produced by defensible software and revenue produced by a thin layer over a commodity service, and a buyer will make that distinction eventually. Doing it before you go to market means you set the price rather than defending it.

Is this the same as a technical due diligence report?

No. A technical due diligence report tells you about architecture, security and code quality. It does not tell you what the business is worth. This work starts from the price and asks which technical facts move it, which is a smaller set of questions with dollar amounts attached to each one.

What is the difference between asking price and clearing price?

Across 7,601 recorded small-business transactions, live listings ask a median of about 3.26 times seller’s discretionary earnings while completed sales clear at about 2.58 times. Those are different populations rather than the same deals measured twice, and the size of the gap depends on which figure sits in the denominator. It is roughly a 26 percent premium over clearing, or roughly a 21 percent discount from ask.

Does this apply to AI companies specifically?

It applies most sharply there. Buyers have become direct about asking whether models are self-hosted or called through an API, and whether any training data is genuinely proprietary. A company whose capability can be reproduced quickly is priced on its customers and its brand rather than on its technology, and the multiple follows.

Is a pricing read the same as a certified valuation?

No, and the distinction matters. A certified appraisal is issued by a credentialed appraiser and is intended to be relied upon by a third party such as a tax authority, a court, an auditor or a lender. A pricing read is decision-grade work for the person making the decision. If you need the former, you need someone with the letters, and I will tell you so.

Related reading

• What acquirers really pay for, from a former corporate development operator.
• The transaction index: what small businesses actually sell for.
• How I price a company in 48 hours.
• Due diligence: what actually gets checked.

Go deeper

If you are buying or selling an online or software business and you want the number and the product read together, that is the work I do. A read takes five business days and tells you what a buyer will pay, what they will challenge, and what to change before you go to market.

Tomasz Felpel is an investor, founder, and advisor in private markets and healthcare, based in New York. He is a three-time founder of Value Alpha, an AI-powered private-markets valuation platform, Sonnerie VC, an early-stage healthcare venture firm, and Pond. Previously he led corporate development and M&A at Fortune 500 scale, pricing more than 1,000 private companies. Columbia Business School EMBA. Read the full story.