How our domain appraisal works

Most appraisal tools return a number and no reasoning. This page documents the entire model behind ours — what it measures, what it multiplies, where the floors sit, how the confidence score is produced, and, just as importantly, what it cannot know.

What the model is trying to answer

There are two very different questions people mean by "what is this domain worth".

Retail

What an end-user buyer — a funded company that wants this exact name — would pay through a broker or marketplace. This is the primary range we publish.

Wholesale

What another investor would pay at auction, where the buyer is reselling rather than building. We show this as a separate, much lower band — typically 12–30% of the retail mid-point.

Wholesale-weighted tools are not wrong, they are answering the auction question. If you are a founder buying a name to build on, the retail range is the one that describes your actual situation.

The calculation, step by step

Step 1

Normalise the input

Protocol, www., path and casing are stripped. The name is split into a second-level label and an extension. Anything that is not a plausible registrable domain is rejected rather than scored.

Step 2

Segment the name into words

The label is matched against a dictionary of category keywords and common brandable business words to see whether it decomposes into real words. This matters because raw character count is a poor proxy for brandability: a seventeen-character string that reads as two clean words behaves like a much shorter name. When a two-word split is found, the model scores an effective length rather than the literal one.

Step 3

Set a base value from effective length

Base values step down as the effective length grows, because supply grows with length and memorability falls. Two- and three-character labels sit in the rarest classes; anything beyond roughly fourteen characters is treated as structurally weak regardless of what it says.

Two structural bonuses apply here: a single real dictionary word (the highest brandability class) and a clean two-word construction that reads like a real brand.

Step 4

Apply the extension multiplier

.com is the benchmark at 1.0. .ai sits marginally below it, .bot lower again, and the long tail of generic extensions is discounted heavily — not as a judgement of taste but because resale liquidity outside .com and .ai is genuinely thin.

Step 5

Score keyword and category intent

Terms with demonstrable commercial pull in the current market — AI, robotics, agentic, reasoning, inference, sovereign, fintech, biotech and regional anchors among them — apply a multiplier. Stacking is damped, so a name crammed with four buzzwords does not compound into a fantasy number.

An additional premium applies when both halves of a two-word name are category terms — an exact-match category phrase such as robotics + reasoning — because those attract strategic buyers rather than resellers.

Step 6

Apply structural penalties

Hyphens carry a steep discount; they are the single clearest signal of a second-choice name. Numerals carry a smaller one, because digits weaken verbal recall — you cannot say them over a phone without disambiguating.

Step 7

Score pronounceability

A vowel-to-consonant ratio in a balanced band earns a modest bonus. This is a heuristic, not linguistics: it catches unsayable consonant clusters, and nothing more sophisticated than that is claimed for it.

Step 8

Apply a retail floor, then a range

Each extension has a floor below which a clean brandable does not realistically broker. The retail mid-point is the floored result; the published range runs from 0.8× to 2.6× that mid-point, which reflects how wide genuine end-user outcomes are.

Step 9

Compute a confidence score

Confidence measures how much structural evidence the model actually had — a recognised extension, matched keywords, a clean word split, a pronounceable pattern. It is capped well below 100 and never rises above 92. A high confidence score means the model understood the name, not that the price is right.

Step 10

Check the live registry, and check our own inventory

Every report queries RDAP for the real registration status, creation and expiry dates and registrar. If the name is one we are selling, you see our live asking price and a link to the listing instead of a modelled estimate.

What the model cannot see

Being explicit about this matters more than the arithmetic above.

  • Comparable sales. The model does not query a sales database. Check NameBio and DNJournal for disclosed comparables before you agree any price.
  • Trademark risk. A name that collides with an existing mark in your class can be worth nothing to you and a lot to someone else. Clearance is always the buyer's to run.
  • History. Prior use, spam history and existing backlinks are not evaluated. A name that previously ran a real product often trades lower, not higher.
  • Who is buying. The largest single factor in any real transaction — a specific buyer's urgency and alternatives — is invisible to any model.
  • Non-English semantics. A string that is neutral in English may be unusable in another market.

An appraisal is an estimate, not an offer, an endorsement, or a guarantee of sale price. We do not publish an accuracy claim because the sales record is too incomplete for any appraisal tool to honestly demonstrate one.

Questions people ask

Is a domain appraisal the same as a market value?

No. An appraisal is a modelled estimate of the range a name would plausibly broker at. A market value only exists at the moment a specific buyer agrees a specific number. Two buyers with different alternatives will rationally pay very different amounts for the same name, which is why we publish a range and a confidence score rather than a single figure.

Does the tool use past sales data?

No. It is a structural model: it scores the name itself — length, word structure, extension, keyword intent, and phonetics — rather than matching it against a comparable-sales database. That is a deliberate limitation and the reason we tell every user to check NameBio and DNJournal for real disclosed comparables before agreeing a price.

Why are the estimates higher than GoDaddy or Estibot?

Those tools weight past auction outcomes heavily, and most auction sales close at wholesale liquidation prices. Our model estimates the retail brokered range instead — what a broker would list at for an end-user buyer. Both are legitimate answers to different questions. The report shows a wholesale range alongside the retail range so you can see the gap.

How accurate is it?

We do not publish an accuracy figure, because we cannot demonstrate one. Domain sales are mostly private, so there is no representative sample against which any appraisal tool — ours included — can be honestly back-tested. Treat the output as a structured second opinion, not evidence.

What happens if I appraise a name you are selling?

You see the live asking price for that name instead of an algorithmic guess, and a link to the listing. We do not use the model to justify our own prices.