You once wrote: "[Agency campaign measurement] is akin to my fourth-grader grading his own homework, on a test where he wrote the questions, in a class he made up in his playroom. In today’s information-overloaded environment, this is clearly a risk for brands.” Do you stand by that statement?
I think it's gotten worse.
Since I wrote that piece, the amount of information continues to grow exponentially. If you're an executive, how do you get down to the synthesis of any issue that's important to you in the amount of time that it takes to make a decision that will actually have an impact?
The amount of time you have to make a decision, it's gotten shorter because of the nature of the media and how fast things change, except that pie of information has grown exponentially at the same time. So it will continue to get worse in terms of how we grade our homework on any issue until we have well-trained machines that understand the difference between what matters and what doesn't.
Because of the volume of data that's available, the ability to manipulate metrics has become so easy that it's causing problems for the space.
If you say to me, ‘I need our campaign results to be 10% better,’ it's not a novel task to be able to go and change query structures and things like that to make a campaign look better than it actually is. Corporate communications’ competitive advantage in the future is going to be getting to the truth. But to deliver the truth, you have to own the truth, and that's why we think it starts with getting better information.
Turbine Labs seems to emphasize ease-of-use in its messaging. Is that your main value proposition?
We really approached the entire problem completely backwards. What we wanted to do was not deliver a glitzy user interface that no one uses.
Director-level and above executives at medium to large companies and lobbyists and policymakers are our target audience. We looked for a long time at what their user behavior was, and I have never seen a CEO write in [the programming language] Boolean to obtain information that they need.
We designed what we call a natural language interface, which means that our subscriber asks a question freehand, much like you would ask or Alexa or Siri a question. User interface became way too complex through the last decade. Dashboards are not being used for substantial decisions; they are used for monitoring. Everyone is tired of new platforms, new apps, new ticker tapes, all that stuff.
But instead of preprogrammed answers, we take [their] question and map that to query sets that we built. [Those query sets] are continually monitoring for junk and various content and bots so that function doesn't have to be done by the requester anymore.
You hired Pawan Lakshmanan almost a year ago. What’s he been working on?
He was on the original development team at a company called Alchemy, which was a natural language processing company that was sold to IBM and basically became the base functionality of NLP for Watson.
He spends a lot of time watching how the information flow happens between outside on the internet into the eyes of our subscriber. We're very concerned about being able to take information no matter what the public source, ensure that we can classify it and bring it into context and then synthesize it down to something that takes an executive less than six minutes to consume.
We don't break copyright law in spirit or in letter, and that limits us in some ways to the data we can collect. We have instead spent time cutting deals with publishers directly so that they get paid fairly and then we can get a view into their full text, because our machine needs to read full text to understand if it sits within an output or not.
In addition to that, he has been spending a tremendous amount of time on our proprietary sentiment engine.
What are some changes you've made?
We spent five years [using] 18 million news and social media posts to train our machine to read like a human. This is our NLP. The only difference is that our machine reads 54,000 times faster than the average human. When you have that kind of computing power, you can consume and contextualize all of Tolstoy's "War and Peace" in about seven seconds.
With our ability to read all this text from start to finish, and ignoring every other keyword hit that might happen on a page, we can give our customer an entirely different view. [It’s not] the counting of [media mentions] but the impact of the corpus of media that's actually meaningful. It removes all of that noise and junk, all the instances where content is on sidebars and tickers and comments.
It reads this story like a human: What am I thinking about after I've read it? We've even melded in how consumers are now reading content on the web, which has changed over the last decade.
What can you say about your product roadmap?
Our roadmap is based on how we consistently take the friction out of that process of getting information. For example, it was really important for us to [start] creating a platform that has one single login where you can read thousands of paywall publications. That took us going to lots of publishers and [partnering with them].
Now instead of clicking on a link within our output, instead of going to hit a paywall, you come into the Turbine Labs hub, with one login, and you see a story that looks just like it would be presented in a Medium-esque model. The [contextual] metrics around that article are presented there as well. [We’ll] continually find ways to make the process of getting information to important people much more efficient and faster without any bias or whitewashing. It's just: Here are the facts.
This story was updated on February 20 to correct Fatzinger's quote about changes the company has made.