GoDaddy may not spring to mind as a developer of cutting-edge AI technology, but the internet company is currently employing new tech to help small businesses compete with tech giants.
“If you have the local bookstore that has built their website on GoDaddy, that local bookstore needs to compete with Amazon,” GoDaddy director of engineering Jason Ansel told VentureBeat in an interview. “And Amazon’s using a lot of machine learning. Amazon is a machine learning powerhouse. [So] basically, how can we use our machine learning expertise at GoDaddy to help that little bookstore compete in an increasingly machine learning-dominated world?”
One of the most significant issues facing those small businesses is a shortage of data compared to their huge competitors. But Ansel says GoDaddy is in a position to pool information across its massive customer base to create intelligent systems that help them all.
The first project in that vein is a system the company claims can value internet domain names better than a human can. It uses new advancements in artificial neural networks to achieve superhuman results, with the goal of creating a valuation metric for domain names that works much as Zillow’s Zestimate works for homes.
“You have this domain valuation industry, which is dominated by a small number of experts who know a lot about how to value domains and have these really large portfolios,” Ansel said. “And it’s also an industry where there’s a huge variation in prices. People don’t really know what domains are worth. And so there’s a lot of people who either pay too much or pay too little.”
GoDaddy tested its system against a group of human experts using a random sample of domain names and their sale prices. The algorithm was more than 20 percent better than the human expert, judged on root mean square error.
To get there, GoDaddy created a new system to better generate prediction ranges for the price of a domain, since many previous techniques assume data distributions that don’t match those the company has observed.
Fueling that system is a complex set of features designed to capture all the complexities of valuing a domain name. To begin, GoDaddy trained a model to separate and evaluate the words in a domain to help determine its value. That’s a complicated task in itself since domain names typically don’t have characters separating them. The system also considers over 100 other features beyond the words in the domain to understand additional factors that would affect its value.
“There’s also things like how long [the domain name is], which you’re better off encoding directly because the value of three-letter dot com often has nothing to do with the words, it’s just that it’s really short,” Ansel said.
In addition to length, GoDaddy’s AI analyzes factors like top-level domain use (.com, .net, .de, .pizza, etc.); which company hosts a particular domain; when it was sold; and where the sale took place. All of those factors help paint a better picture of the demand for a specific piece of digital real estate.
To create this system, GoDaddy needed a massive amount of data, which it has in spades. Ansel said the company has millions of historical data points on domain name sales, which can then be used to perform transfer learning on the resale data set, which has 250,000 data points.
“So really, GoDaddy is one of the very few companies in the world that could produce this, because we’re the only ones with the data,” Ansel said. “GoDaddy in the United States is around 60 percent of the primary domain market, so we’re the largest domain name registrar in the world. In the domain name aftermarket — which is basically reselling domains from person to person — we’re also the largest by the count of sales.”
Right now, the domain name valuation tool is available in open beta.
Correction 9:40 a.m. Pacific: This story initially said that the domain valuation tool was not yet available. GoDaddy has released it in open beta.