• by Laurie Sullivan , Staff Writer @lauriesullivan, April 16, 2014

The Seattle-based startup Blab wants to become the crystal ball of media advertising. It created technology allowing brands to predict, with 70% accuracy, trending topics up to 72 hours in advance by analyzing social conversations.

Blab launched late last year with a handful of some of the world's largest companies. On Thursday it will make available to all companies the second iteration of its platform and announce partnerships with several agency partners such as Horizon Media, which invested in the company. The platform makes about 1 million predictions per minute on chaoticdata based on time, volume, and time and volume by media channel. The accuracy will increase as the platform learns.

Predicting trends, an ambitious goal requiring the ability to index live conversations, means understanding a variety of vernaculars, languages, and formats such as images, text and videos. Blab CEO Randy Browning, former CEO of Publicis in the West, said the company isn't a social monitoring tool. "Think of us as the Google for conversations," he said. "Google's job is to index the world's information, and we're indexing the world's conversations."

The real-time platform processes about 100 million conversations daily from 50,000 sources worldwide, per Browning. The information comes from every major news source, blogs, and social platforms. It can predict the media platform that will drive the conversation and for how long. The data gets classified into topic segments. Brands and agencies can discover the topics to support campaigns, rather than doing searches. "We don't get stuck in natural language processing or the linguistics processing that everyone else does," Browning said. "If you do it that way you can't succeed. You can't look at German and Farsi as well as videos and images. It doesn't work."

The technology doesn't filter social conversations. It pulls words and analyzes the psychological meaning to predict the actions consumers will take three days in advance.  "We have a seeding term that's typed into the engine and then the engine finds all the associated term and terms and brings in conversations," Browning said. "Think of us as a mathematical vs. a linguistics approach. We're not looking at the content except for the association of the term around the original term."

Lots of companies are getting into the data biz. Twitter, which has played down its data licensing business known as the "firehose," acquired Gnip, a company offering access to historical Twitter data, the full Twitter data firehose, and APIs that provide data from other social networks like Bitly, Instagram, Reddit, and Tumblr.