ZoomInfo, the source of business information on people and companies, last week launched ZoomInfo Lists. The company calls it a powerful direct marketing tool that provides email, phone, and print direct marketers with the ability to create targeted marketing campaigns from a CAN-SPAM compliant database, updated daily, of millions of heavily indexed people and companies.
Among the features of the new service are the ability for marketers to email targets as many times as they want over the course of a year; the option to focus campaigns to specific audiences based on its 24 categories of information; and in-depth profiles of prospects including their career history, education, and memberships on boards or trade organizations.
What’s behind the new service is ZoomInfo’s semantic search engine and artificial intelligence and natural language algorithms, which CTO William Wechtenhiser says are the force multipliers in helping marketers target their campaigns to those who are likely to be most interested in their offerings, most likely to respond to it — and indeed most likely to want to be found by the marketer.
“We’ve got 50 million people who are very heavily indexed, associated with companies, and we have a lot of information on those companies themselves. So to slice and dice this on semantic principles that are interesting to your business is pretty cool,” he says. Even when this results in smaller lists of prospects, the value is they are the prospects they actually want to contact.
At a very high level, ZoomInfo takes unstructured or semi-structured content off the web and alters it into structured data that can be semantically searched, Wechtenhiser explains. There’s a lot of sloppy stuff on the web, so the challenge is keeping its data complete and accurate. Semantic and natural language technologies such as sentence-based extraction and information unification enables ZoomInfo to make sense out of two different profiles of a person named Tom Smith, for example, so that it can conclude whether they might be the same person.
“There is lots of data on the web that contradicts each other, either because something is old or false or it was a typo,” or for other reasons, Smith says. “At the end of the day we get our best guess of who this person is, or who the company is,” so that marketers — or other searchers — are able to get results that correctly correlate that information based on the criteria they set. ZoomInfo looks at hundreds of millions of web pages and gets tens of billions of facts from them. For example, it can see a sentence in a press release that says something like John Smith left Company A and joined Company B, and has a new title, and use that information to update its records so that the existing John Smith’s data is updated rather than duplicated.
Its proprietary code and rule sets handle the complexity of the content that populates its taxonomy trees for people or companies, mining keywords in extracted unstructured data for indications about, for example, what industry a particular company is in, and what products and services it offers, or its geography or acquisitions. “So you can do a high level search to find all the executive-level or higher marketing people in the health care industry within 100 miles of Atlanta Georgia,” he says.
ZoomInfo has been around eight years, before many of the new Semantic Web standards were developed. But even if they had been, Wechtenhiser says they wouldn’t on their own be able to service ZoomInfo’s requirements. RDF triples, for example, have some limitations in terms of being able to express only binary predicates, or requiring an OWL inference engine for transitory reasoning. “In our case we have code written that does these complex things and we have our own rule sets,” he says. He could see that Semantic Web standards may some day have a role to play in tree structure for industry or people taxonomies, but that’s about as far as it’s capable of going for now. In terms of the data on the Web that ZoomInfo collects, “there’s no real substantial place to get nice structured data via RDF or other semantic web standards. It’s not adopted widely enough,” or where it is, it is generally exposed in a very limited set via APIs.
Down the road, Wechtenhiser sees potential to enable notification services to ZoomInfo subscribers that a piece of data has changed-for example, a company was acquired by another entity or someone left their post. This would be useful especially in cases where companies have integrated ZoomInfo with software such as SalesForce.com, which lets companies import ZoomInfo data into their CRM system or update existing records.
“We don’t just want to submit data in a static snapshot but keep track of what is changing over time and which of those changes you care about,” Wechtenhiser says of notification services. That could go beyond the flat list of people you already sell to to include, for example, any new VPs of marketing in the industry you serve, who might be a potential new customer for you. “Getting that notification on the day we find the press release saying there’s this new person — that has value,” he says.
That opportunity presents its own challenges on the semantic front. “Semantic search is generally expensive, uses more servers than you want it to, and as it gets more complicated it gets bad fast,” Wechtenhiser acknowledges. ZoomInfo has taken steps to mitigate that in its current offerings, and will need to do so as it contemplates new options like notification services.
“You can’t afford to run 1 million searches against your data every day or every minute — that process doesn’t scale. So those are concerns for us. In every case we build special purpose systems where we try to use the underlying platform, like search, as far as it goes and flatten things into a format that is friendly, because raw search will take you far if you can translate problems into a domain that’s friendly for it,” he says.
Let the raw scalable systems do what they can so ZoomInfo technologists can focus on the hard parts of the problem — for example, special software might generate a single alert for ten people on the lookout for new VPs at health care companies within 100 miles of Atlanta, Georgia, rather than an alert for each person watching for that information. Says Wechtenhiser, “There’s all sorts of performance tuning and tweaking to be done up and down the stack.”
ZoomInfo says ZoomInfo Lists was in use by some 30 beta sites. “ZoomInfo Lists allows the use of multiple selects to build a list from the ground up; using criteria of their choosing to specifically target people that have the characteristics of buyers for their products,” says senior product manager John Kelley.
” This is very different from other lists sources as a user typically starts with a set list type, and then uses 1 or 2 selects to further segment the set list down.” As an example of this, he says, one customer was looking to target CFOs, but it was not enough just get a broad list of CFOs across America. “Using ZoomInfo Lists, she was able to first narrow the list down to CFOs in the tech sector, then narrow the list down to one region of the country — Mississippi. Because ZoomInfo uses natural language processing she could take it even further by focusing the list on CFOs who graduated from OleMiss (University of Mississippi). In the end she had a targeted list of 50 or so CFOs who she had a direct connection to in one area of the country.”