How we estimate private-company revenue

Private companies don't publish their revenue, and the numbers scattered online are usually one unsourced figure copied from site to site. We wanted something better: a repeatable, testable estimate you can reason about — with the error bars shown.

Estimate it two independent ways, then reconcile

There's no single public number that gives away a company's revenue. But there are two independent views of it:

  1. The website funnel. How much traffic does the site get, and what share is real acquisition traffic vs. logged-in users? Published SaaS conversion rates turn that traffic into signups, then paying customers — multiplied by listed pricing, an estimate of self-serve revenue.
  2. The demand model. How much do people search for the brand, how mature is the company, how long has it existed? A model trained on companies whose revenue is public turns those demand-and-scale signals into a total-revenue estimate — including enterprise, sales-led revenue a funnel can't see.

We reconcile the two. For a self-serve product the funnel leads and the demand model confirms; for an enterprise, sales-led company the demand model leads and the funnel is a floor. The result is a single range.

We look things up — we don't invent them

Every rate in the funnel — visitor-to-signup, signup-to-paid, churn, customer lifetime, tier mix — is a published industry benchmark, not a knob we tuned to get a nice answer. Only the demand model is fitted, and it's fitted on real reported revenues.

Where a real number exists, we use it

For companies whose revenue has been credibly reported (press, filings, reputable data providers), the page shows the reported figure with a Verified badge, not our estimate. We never call an estimate "reported."

How accurate is it?

We can grade ourselves, because we know the real revenue for a set of these companies. So we test blind: hide the real number, run the full model as if we'd never seen it, and compare.

Ranges and confidence, not false precision

Every estimate is a range with a confidence label. High confidence means a reported anchor or two methods that agree closely; low confidence means thin signal and a deliberately wide range. Read the range, not the midpoint.

What this is — and isn't