Creating a good brand is one of the hardest tasks of an entrepreneur. Aaron Patzer, CEO of Mint, stated that you should “expect to pay $3 to 15k for a 6-8 letter, single word, English domain name.” In fact, he paid $180,000 in equity for “mint.com”, which took over three months of negotiation. A domain name should be spelled unambiguously to prevent losing word-of-mouth referrals. Sean Cheyene of Herbal Ecstasy believes that brand is everything. Without the brand name, it would have been nearly impossible to create a $350 million dollar company from selling a plastic baggy with herbal pills — with no reputation and money.
Today, there are 88,312,535 million top-level “.com” domain names registered, which makes it difficult to purchase a domain like “efax.com”. (To put this in perspective, Webster’s Unabridged Dictionary has 475,000 words.) Thus, any combination of existing words we come up with for a “.com” domain name will probably be taken. We can browse existing domain names for sale at sites like BrandBucket, where they have “brandable business names and unique domain names”. However, a domain name like “Zables.com” goes for $5,000 and I am still not happy with that name. Perhaps then, a good indicator is to look at expiring domains, which I can try to bid for — for only $69.
I recently stumbled upon Namejet, which receives over 16,000 soon-to-expire domain names per day. I can bid on these domains, although it’s difficult to navigate through the list and find a good name, thanks to it being filled with domains like sxwzjhxx.com. Also, we have biased and limited searching capabilities. For example, if I am looking for a real-estate domain, I would try all possible combinations of keywords related to real-estate (as would every other person building a real-estate website) — then find that the names that I’m interested in are overpriced and difficult to win.
Since a company’s success can greatly increase from a good domain name (think Mint vs. Geezeo), I decided to use my machine learning knowledge and build a predictor to determine the probability that a given domain name resembles an English word: P(w=domain_name). After downloading then sorting 340,000 domains with a “score”, I found reasonable domain names with ease: flipcast, drawmash, idealix, shopvolt, swerp, raideye, geekleaf, wirednut, moccah…just to name a few. (I computed the probability score using n-gram letters. If the first 3+ letters of the domain matched the dictionary (e.g. shop in shopvolt), then the probability for that section of the domain is set to 1.0 to handle transitions such as “pv”, which are not found in the English dictionary.)
While it’s still difficult to find a domain name for a specific product, I just found meaningful domain names for a streaming video website (Flipcast), a creative company (Drawmash), a gaming company (Raideye), a shopping CSE (Shopvolt), and some new blogs (Geekleaf, Wirednut). Win.