Retail vs auction: the same engine, two very different answers
Measured on 2,000 recorded sales since 2020, the engine’s figure is 1.2x the sale price at retail marketplaces and 11.8x at wholesale auctions. Those are medians: only 33% of retail names and 10% of wholesale names land within 2x of their sale. The figure is a retail estimate. It is not a prediction of what a name will bring at an expired-domain auction.
What sold, and what the engine said
| Where it sold | Names | Median sale | Median engine | Within 2x | Over 2x high |
|---|---|---|---|---|---|
| Retail marketplaces | 1,000 | $2,088 | $3,575 | 33% | 38% |
| Wholesale auctions | 1,000 | $235 | $3,670 | 10% | 88% |
Retail here means Sedo, Afternic, BuyDomains, Uniregistry. Wholesale means GoDaddy, NameJet, DropCatch, Dynadot, Park.io. 88% of wholesale sales sold for less than half the engine’s figure. At retail the errors run both ways: 38% high by more than 2x, 28% low by more than 2x.
Where the gap is widest
Engine figure divided by the sale price, median per name, on a log scale. The line marks 1x, a figure equal to the sale price.
| Cohort | Retail names | Retail ratio | Wholesale names | Wholesale ratio |
|---|---|---|---|---|
| All names | 1000 | 1.2x | 1000 | 11.8x |
| Keyword | 351 | 1.3x | 200 | 14.6x |
| Brandable | 332 | 2.4x | 370 | 23.4x |
| .com | 742 | 1.5x | 878 | 13.0x |
| Other extensions | 258 | 0.4x | 122 | 3.9x |
The gap depends on the price, not only the venue
Sorting the same sales by what they sold for shows the pattern. The engine’s figure sits in a narrower range than sale prices do, so it runs high on the cheapest names and low on the dearest.
| Sold for | Retail names | Retail ratio | Wholesale names | Wholesale ratio |
|---|---|---|---|---|
| under $250 | 77 | 62.7x | 523 | 19.3x |
| $250 to $1,000 | 80 | 8.1x | 380 | 9.9x |
| $1,000 to $5,000 | 661 | 1.4x | 78 | 3.2x |
| $5,000 and over | 182 | 0.2x | n/a | n/a |
- Rank agreement between the engine’s figure and the sale price is 0.04 at retail and 0.16 at wholesale (Spearman, 1 is the same order, 0 is none). Inside these sales, which mostly fall in a narrow price range, the figure does not reliably tell a cheaper sale from a dearer one. Do not use it to rank a list of names that sold in the same range.
- This is also why the accuracy page can show the engine below the sale price. It is scored on a different set, recent reported sales, which run dearer, and on the dear end this sample reads low as well.
What to take from it
- Retail and wholesale are different markets. The median wholesale sale is $235 and the median retail sale is $2,088, in the same ledger, and the engine is built on the retail one.
- Brandable names show it most. At retail the engine runs 2.4x the sale price for them, and at auction 23.4x. A name that does not split into known words is priced on thin evidence, and the sale records lean toward names that sold well.
- Other extensions run the other way at retail: the engine’s figure is 0.4x the sale price, below 1x. The error is not one direction.
- A single divisor would not turn a retail figure into a wholesale one. Only 10% of wholesale sales land within 2x of the engine’s figure, and a flat cut moves the median without narrowing the spread. A wholesale estimate needs its own model, built on the ledger’s venue labels. Appraisals now show one where the evidence allows: a wholesale band under the retail figure, from what names in the same segment (extension, structure, length) cleared at investor auctions. Thin segments get no band. On later auction sales, 66.8% landed within 2x of the band; the retail figure as-is managed 9.3% on the same names.
How this was measured
- Source: the 461,352-row sales ledger as of 2026-09-23. One row per domain, its latest sale, sold since 2020-01-01, at $10 or more.
- Sample: 1,000 names per side, taken at even steps through the ledger ordered by a fixed hash of the name. The same ledger gives the same sample and the same numbers.
- Each name is appraised with its own sale excluded from its comparables, so the engine cannot see the answer. Names in the frozen test set behind the published accuracy figure are left out.
- The ratio is the engine figure divided by the sale price, taken per name; the tables show the median. Cohorts under 40 names are not shown.
- Venue labels are as recorded in the ledger. Sales reported without a venue are not in either group, and a recorded sale is only what was reported, so the retail group leans toward names that sold well.
- Nothing on this page lists a name or a per-name figure. The numbers are medians and shares over the sample.