How many sold comps does a valuation actually need?

A comp-based valuation is a claim about a distribution: businesses like yours have recently changed hands somewhere in this band. Every such claim rests on a sample, and a sample of one or two carries no information about the band — only about those two sales. Quote a median off three deals and you have published a number that would move by half if one more deal landed next week.

So our calculator enforces a hard minimum of five usable comparables. Below it, the API returns no estimate and says why — Not enough comparable sold deals yet (n found, 5 needed) — rather than returning a figure with a caveat attached that nobody reads. The same floor governs what the paid report will price, and what the medians on this site are published against.

Five is not a statistical threshold anyone can defend as the right number; there isn't one. It is the point below which we judge a published figure to be actively misleading. Treat it as the line between "no answer" and "a weak answer", not between "weak" and "good".

The five are counted after filtering, not before, which is why a category holding hundreds of sold deals can still fail to produce an estimate. Three things happen to a comp set on the way to a number:

1. Rows with no usable multiple drop out. A sold price alone cannot produce a multiple; the source page has to state earnings or revenue too. Most published sales don't.
2. Impossible multiples drop out as data errors. Anything at or below zero, or at or above 100×, is a parsing failure or a mislabelled figure rather than a comp, and is discarded before any median is computed.
3. The tails are trimmed once the sample is large enough. At twenty or more surviving comps we cut 5% from each end, so one spectacular sale cannot drag a category's median on its own. Under twenty, nothing is trimmed — there is not enough there to spare.

The five-comp test then runs against what is left. That ordering matters: a category can show four figures in its "deals tracked" column and still, correctly, publish nothing.

Three properties of a sample matter as much as its size, and none of them improves just because the count went up.

Spread

Twenty comps clustered inside a narrow band and twenty scattered across an order of magnitude support very different claims, and the median looks identical either way. That is why every estimate we publish is an interquartile range — the middle half of comparable sales — rather than a point. If the range comes back wide, the honest reading is that comparable businesses genuinely sell for wildly different prices, not that the midpoint is your number.

Venue mix

Twenty comps that all came from one marketplace describe that marketplace. Sold prices are published very unevenly — auction sites list closes, brokers mostly don't, private deals never do — so a comp set is a sample of what gets published, not of what gets sold. Our estimates therefore report the venues behind the sample and what share came from the single largest one. When that share is close to 1.0, you are pricing against one venue's economics.

Size match

A comp set whose typical business is four times larger — or four times smaller — than yours is not describing your deal, however many rows it has. We flag that mismatch explicitly rather than letting the count imply a comparability the sample doesn't have. What makes a sold comp comparable takes that apart properly.

Loading live coverage from /api/stats…

Coverage is measured over the paid report's window — the most recent 200 priced sales of that type — so these counts describe what a report would actually price against rather than everything in the database. The free calculator reads a wider window of 500. Both apply the same five-comp floor.