The form was never the product

I shipped a business valuation calculator. It lives at jmemaxwell.com/business-valuation and it's free.

The short version: it values owner-operated private businesses, it shows all of its work, and the same engine that powers the web page is also a tool your AI can use directly. That last part is the reason it exists, and I'll get to it. First, the calculator.

What it actually does

Ask "what's my business worth" and the answers on offer today are a proper valuation engagement (thousands of dollars, weeks of back and forth) or a multiple somebody heard at a barbecue. There's not much in between, and the in-between is exactly where most owners are when they start thinking about succession, a buyout, or just whether the thing they've built is worth what they hope it is.

So this sits in the in-between. It runs a capitalized free cash flow valuation, which is the income approach a real advisor would use on a business like this, not a multiple pulled out of the air. You feed it real financials. The P&L side: revenue, EBITDA, depreciation, what you actually pay yourself versus what it would cost to hire your replacement, plus any normalization adjustments for the one-off stuff that shouldn't count against a buyer. The balance sheet side: cash, debt, receivables, inventory, payables, the tangible assets the business runs on. Then the risk side, which is where the value really gets decided: industry, size, how concentrated your revenue is, and how much of the business walks out the door when you do.

The discount rate is built up from sourced components (Damodaran, Kroll) with presets you can override. And here's the design decision I'd defend in a bar fight: anything you leave blank falls back to a disclosed default with the source attached. Not a hidden assumption. A visible one, with a citation and a retrieval date. You can start with a handful of numbers, get a defensible ballpark, and tighten the answer as you replace defaults with your real figures. If you want to go deep there's a detailed mode with working capital timing, growth patterns, and a full WACC build-up.

The report

What comes out the other side is the deliverable I'm proudest of: a proper report, free, and written for humans.

Headline value with a range. A cautious-buyer reality check, because the headline number is not what someone writes a cheque for. Market multiple cross-checks to corroborate the income approach. A sensitivity table so you can see how much the answer moves when the assumptions do. A bridge from enterprise value down to what actually lands in the owner's pocket. And then the full audit: every formula shown, every assumption listed with its source and date, a glossary at the back for the terminology.

It's written so an owner who has never read a valuation can follow the logic from revenue to conclusion, and detailed enough that their accountant can pick a fight with it. (Please pick a fight with it. That's what it's for.) It's a planning estimate, not a professional valuation opinion, and it says so on every page.

The real reason I built it

I didn't build this because the world was short one valuation calculator.

I built it because I wanted to ship one piece of software that serves humans and AIs as equals, from day one, off a single set of business logic. The web page is the human side: forms, results, a PDF. The AI side is an MCP tool. MCP (Model Context Protocol) is the standard AI assistants use to call external tools, and it means you can hand this calculator to Claude the way you'd hand a colleague a spreadsheet. Paste one URL into your AI's connector settings and it can run the exact same valuations the website does.

Same math. Same sourced defaults. Same report. There is exactly one copy of the valuation logic, and both the web page and the AI call it. The web page formats numbers. The MCP endpoint speaks protocol. Neither of them calculates anything.

That discipline is the whole experiment. Because the alternative is what most companies are about to do by accident: build the product for humans, then bolt on an "AI integration" eighteen months later that reimplements half the logic in a second place and quietly drifts out of sync with the first. I've watched versions of that movie before with mobile, with APIs, with every channel that arrived after the original product shipped. The bolt-on always costs more than building the second door up front would have. I wanted to feel that cost first-hand instead of theorizing about it.

The honest answer: it's cheaper than I expected, and the expensive part wasn't where I thought. The math engine took discipline but no heroics. The thing that fought back was plumbing. claude.ai's connector dialog takes a URL and OAuth, nothing else, so partway through the build my little calculator website had to grow up and become its own OAuth provider. Nobody puts that on the brochure for "just add AI to your product." It's on mine now.

What it's like to use

The part that genuinely surprised me is how different the AI door feels. You don't fill out a form. You say "value a landscaping business doing $2M revenue with $400K EBITDA, the owner pays herself $150K but a manager would cost $90K" and the AI maps that to the tool, runs it, reads back the conclusion, and then argues with you about the discount rate. You ask "what if revenue concentration is worse than I said," and it reruns the valuation while you finish your coffee.

That's when it clicked for me. The form was never the product. The math was. The form was just the only interface we had.

Try it

• The calculator: jmemaxwell.com/business-valuation
• Connect your AI: jmemaxwell.com/business-valuation/connect (in Claude it's Settings, Connectors, paste the URL)

My prediction, worth the usual $0.02: within a couple of years, shipping software without an AI-facing door will feel like shipping software without a mobile view did in 2012. Not wrong exactly. Just dated, and obviously so.

Cheers.

JME

P.S. If you run it on your own business and the number offends you, the inputs are right there. Argue with the assumptions, not the calculator.