Decision MediaWorks

DCF: A Sign of Things To Come

We saw this article from the Miami-Herald about the Department of Children and Families new fraud prevention technology. The article says the new LexisNexis program “uses complex algorithms and ‘billions and billions of records across multiple decades’ to perform what it calls identity analytics.”

The Department would not identify the “secret sauce” of the program so we decided to demystify the technology for you.

The new program uses the same technique that Facebook uses to verify your identity. The program randomly asks applicants about a piece of information from their history, just like when Facebook asks you to identify pictures of your friends if you are locked out of your account.

The DCF program figures out what information to ask about by searching for public records associated with an applicant. The “algorithms” (a set of instructions to a computer) for searching these documents and matching them to an applicant are not really that “complex” in a technical sense. It works something like this:

1.The computer is told to “go find all records that match the set of information (name, phone number, social security number, address, etc.) provided by the applicant.”

2.The computer finds these documents and then ranks them by what documents look like the best match. This is kind of like how Google ranks your searches.

3.Another algorithm searches the best matching documents for predefined data fields (year, color, location) and generates a question to the applicant based on that information. For example, drawing from a birth certificate the question may be “What city were you born in?” Or from a car registration you could get “What color was your 2007 Honda Accord?”

The real cost drivers for these Big Data programs are the collection and storage of the data. Those costs however, continue to rapidly fall which will allow new programs like this come online over the next decade across the public sector. For instance, a similar program could not be used to verify voter identities.

As always, if you found this interesting please share and subscribe to our mailing list on the sidebar for our quarterly Data Intelligence Briefings.