Emerging Tech Roundup with Kris Hammond, Narrative Science

May 1, 2015



We’re excited to bring you The Emerging Tech Roundup, a new podcast series launched in partnership with Boston-based Blue Hill Research.

Each month, my co-host James Haight and I will be joined by industry experts and thought leaders shaping the future of business through emerging technology. We’ll discuss where technology is headed and how it impacts businesses today, covering topics from 3D printing and artificial intelligence to biometrics, next generation security, predictive analytics, and more.

Subscribe via iTunes

This Episode’s Guest

kris hammond_circle“I think that the notion that everyone in the world has to be data literate is ridiculous…. If everyone has a data scientist that hangs out with them and explains things to them, that’s what Quill will be.”

— Kris Hammond, Chief Scientist and Co-Founder, Narrative Science


Listen Here

In This Episode



James: Hey everyone, welcome back to the podcast. Today we have a great guest for you  Kris Hammond. He is the chief scientist and cofounder of Narrative Science. Kris, we would love to get to know you just a little bit better. Can you introduce yourself? We’ll go ahead and take it from there.

Kris: Okay, James. I am in fact indeed Kris Hammond. I’m the Chief Scientist and one of the co-founders of Narrative Science. I’m currently on leave from the computer science department at Northwestern University, where I’m usually a professor, but right now I’m really focused on taking Narrative Science technology and moving it both into the marketplace in terms of building a business and in terms of turning it into something that can really have genuine impact on the world.

James: Fantastic. For us, we’re really interested in Narrative Science in general. Kyle and I were talking about how interesting you guys are as a company. We’re excited to have you on the podcast. Welcome aboard.

Kris: Thank you.

Kyle: Can you give us a little background on Narrative Science?

Kris: Sure, well, the core technology that is Narrative Science is actually a piece of technology we call Quill. It does one thing in the entire world. It takes data, numbers, and symbols, unambiguous structured data. It looks at that data, figures out what’s going on in the world on the basis of that data, then turns it into a story that it expresses in natural language.

It can take any data, any kind of structured data. If there’s a story in there, it can extract that story and tell that story in a way that is absolutely readable by a human being and absolutely indistinguishable from something a human being would write.

Kyle: You know what got me really excited about it was the stuff that you’re doing with Forbes. Can you give us a little bit of background on that?

Kris: Oh, sure. Forbes is actually one of our earliest clients. What we do for them is we take current data associated with historical information for a company in terms of its earnings. We take analyst’s predictions, we take other companies that have already announced their earnings, and we do earnings previews for Forbes. And that is to say we write the story about what we think is going to be happening based upon what the analysts have said and what the past performance of the company is.

It really was the case that with Forbes in particular, they do a magnificent job with the earnings previews they could produce but given any earning season they might do a couple hundred. We can do thousands because we were able to take the analysis that they would do, the tone they would use, the information they would look at, and the style of their writing, map it onto Quill and then have Quill take what was a process, which is not a huge process for them, a couple hours of a staff writer writing a story. We can take that process and bring it down to a second.

That’s actually kind of exciting that we can take that kind of analysis and prose and produce it so quickly that we can scale it to not do 200 stories or even 1,000 stories, but 10,000 stories. We can do thousands upon thousands of stories because we have the ability to scale.

It’s the beauty of the system that we can take a human skill, bring it to the machine, and allow it to actually run at machine speeds, with machine precision and machine consistency, to scale it.

For us the thing that was really exciting was it meant that the people who were writing these commodities stories at Forbes could actually end up writing stories that they really enjoyed more. That is more strategically focused, more humanly focused, more based upon picking up the phone and making a call to a CEO rather than looking at the numbers and producing a story on the basis of that.

James: That’s fantastic. One of the questions I had — you said indistinguishable from say, if a human wrote it. We’re talking grammar and we’re talking prose. You’re not talking formulaic sentences is what I’m assuming.

Kris: Yes, the human element of Quill really comes from the fact that we have people who know how to write who are configuring things, and even at the end of the day, picking the collections of words that are used to express something. Also, it will pick exactly what it is that it’s going to say.

I’m trying to think of a really good example of this. If I’m writing a story for you about your portfolio, I’m going to actually tell you about the places where you can actually take action. The words I’m going to pick are going to be words that are perhaps nontechnical words if you’re that audience, and really words around performance, words around the selections you have made.

I’m not really going to get into the technical nitty gritty of exactly what kinds of analysis I’m doing to tell you whether or not things are doing well or badly. If I’m doing the same sort of story, but I’m handing it off to someone who is managing a fund, for example managing, it’s the same kind of analysis, the same kind of reasoning underneath it. But in fact, the language is going to be much more technical, much more controlled, much more in line with regulatory concerns. Both of them will read like they were written by a human being who actually understands the audience.

That’s a huge component of what we do. We understand who we’re writing for. We can translate that into the decision making that Quill goes through in terms of taking things down to the individual word level.

Kyle: All the data that you’re bringing in, are there limitations to consuming and reporting the information you have now as it currently stands?

Kris: I think for us the data limitation really is about less the nature of the data and more about the nature of the story. Sometimes there’s a story an organization will want to tell that is like some sort of performance review or some sort of profiling, and the data just isn’t there. They might have a lot of text but they don’t have something that’s structured, well behaved, and normalized and all those things. That’s where the limitation is.

The limitations are usually on the nature of the data. The size of the data pool usually is not an issue because the reality is that although you might for a given story, you might actually refer to something in the aggregate. You’re not going to look at a petabyte of data for one given story. You’re instead going to look at the results of doing analysis against that data handed to Quill for the actual storytelling.

James: That’s really interesting. When we talk about limitations what if we look at it from the reverse side? I’m curious about what the limitations are for companies out there today. We know that scale is a limitation or the amount of writers I can hire is a limitation. I know you guys do a lot of work with the financial reporting. What are the top one, two, three pain points that you’re seeing out there of why Narrative Science matters?

Kris: I think the massive paying point that we see in the world really has to do with the history of big data and the history of data analytics in general. We’ve collected outrageous amounts of data. We have analytics, we can run across that data to discover and report on absolutely almost everything that data touches. We have visualizations where we can just display that data, but when push comes to shove the people who need the information that’s associated with that data usually do not have the depth of data analysis skills to pull it out on their own. Literally, a company will have tens of millions of dollars of investments in data analytics, in data scientists.

Then the bottleneck becomes they need someone to look at that dashboard, figure out what’s going on, and turn in not one report but usually hundreds of reports. If they really wanted, thousands of reports so every single stakeholder associated with the data gets the right information. That pattern we see over and over again.

We’re seeing a shift away from, “Let’s grab all this data and do something with it.” That was really driven by engineering. We’ve seen a shift to the business side of business, saying, “Look, there’s things we need to know. I need this information. Not the data, but the information. I need a mechanism for turning that data into information I can actually read, ingest, and deal with so I can make decisions.”

That is a constant pain point. Everything else pales in comparison to that. If I’m looking to report, if I need to know how all my sales force is doing, I can use sales.com, I can look at the spreadsheet,  at the visualization, but it’s not going to tell me. If I map that data into stories for every single sales person, every single manager, all the way up the hierarchy of the organization, then everybody knows what’s going on. They can understand it and respond to it at exactly the level and exactly the way in which it is the most powerful for them to understand things and really take those actions.

That’s the kind of story we keep hearing over and over again. Even if you look in the world of visualization, clearly they’re giving ways to look. We’re going to help you tell the story. For us, it’s less “we’re going to help you tell the story” and more “Quill will tell the story.”

We’re not a tool to help you tell the story. We’re a tool for telling the story of what’s happening on your factory floor. What’s happening with your logistics? What’s happening with your sales team? What’s happening in terms of the effects of your marketing? All of those areas have tremendous masses of data, which is absolutely going fallow because we do not have the people to look at it, figure out what’s going on, and report on it to the people who make decisions. That for us is huge.

Kyle: Yes, it leads me to a quote I heard a long time ago, which is basically, “We must move from numbers keeping score to numbers that drive better actions.” I think that’s what you’re talking about. We can collect all this data constantly and try to figure out what it’s saying, but we want to actually drive actions in business and do instead of just report constantly.

With that said, where do you think the future lies? What is the grand vision for what you guys are doing? What’s the future of everything that we’re talking about, other than Terminator coming back?

Kris: Well, there’s that. In all the conversations about AI, and AI’s going to kill us, we look at Quill. Yeah, Quill is going to explain you to death.

James: You’re going to know so much that it’s going to blow you over.

Kyle: A slow death.

Kris: For the core technology, it’s a wonderful horizontal technology. We have our focus in financial services and marketing services, performance, looking at how things are performing. But the reality of the technology is that anywhere there is a collected data set, and that data set was collected for a reason and the specific kinds of information you want from that data set including things that are advisory and predictive, Quill can actually take on the role of explaining what’s going on. Not in the data but in the world through the lens of that data.

The really long term of this is that I fully expect the Quill technology to be the face of data, to be the voice of data. Wherever right now there is a spreadsheet, there will instead be an explanation of what’s going on in the world based upon Quill reading that spreadsheet and explaining it to you.

People will look at spreadsheets the way my generation looks at computer punch cards.

Even when I was a kid it was something that was already archaic. That made no sense anymore. But we will look at that level of data, a machine communicating with us on its level as being archaic. Some people will have nostalgia about it, but it will be outmoded because everyone will be thinking, “No, no, no. When I want information from the machine about the data it’s collected about my world, my business, my government, myself, my health, all the equipment I touch — when I want that information it’s simply going to tell me. It’s going to tell me because it knows how to look at that data, figure out what’s going on, and explain it to me.”

That is the future for us. To get there, we’re going to make money in financial services. We’re going to make money other places. Build the company, grow the company and think about where we can integrate into places where there’s data, but there might not be information in sight yet. Our growth for the next few years is thinking about where are there places where people have invested in data, invested in trying to get that data out, but still are tremendously unsatisfied? We’re going to start living there as well.

James: A couple of things that popped to mind — maybe this is too out there futuristic or whatever, but something like the Chief of Staff, the President or you CEO, you get all your briefings. You’re not showing charts. You don’t have to have someone be the intermediary that interprets it. You just get it right there.

Kris: Absolutely, yes. Absolutely. This gets interpreted strangely sometimes. I think that the notion that everyone in the world has to be data literate is ridiculous. If you know how to run a country, if you know how to run a business, you shouldn’t have to know how to deal with the level of noise when you’re doing time series analysis. That’s just a ridiculous thing for you to have to know, too. Having a system that knows about that, but also knows you and knows what you need, that’s the beautiful thing. If everyone has a data scientist that hangs out with them and explains things to them, that’s what Quill will be. It will the data scientist that constantly explains stuff to you.

Kyle: I would love a personal data scientist that just follows me around.

James: Kyle and I are just looking at each other and nodding how awesome it would be.

Kris: For me, it’s always funny because I’ll talk to people who are living in the world of data. I’ll give a public talk and I’ll say, “Who within the last month has looked at a spreadsheet?” Almost always everyone has got their hands up. Then it’s a question of, “Okay, who found that a pleasurable experience?” Nobody. Who wants to wrestle with correlations between columns, flipping between pages, looking at this graph against this graph? When in fact that process is usually for anyone who does the same kind of spreadsheets in the same job in the same way, that kind of process is really codifiable. You can really be brought into Quill.

Would you rather have it be that instead of a spreadsheet? You’ve got a one page executive summary of what the hell’s going on. If you’re the CFO, you’ve got your financials. If you’re running logistics, you get this telling you where all the bottlenecks are. Your factory, all of those sensors that are associated with the machinery are now telling you what’s being over utilized, what’s being underutilized, and where you might see a problem popping up in the next month. All of that data can be turned into that information. You just need the technology to do it.

James: What about wearable devices? I know there’s a huge trend towards personal health body biometrics. It tells you what you’re doing. I’m sure it can integrate with what diagnoses might be or other trends.

Kris: Yes, I think the quantified self-communicated in a qualitative way through a narrative about how you’re doing where it’s not just a single data set from Fitbit, but from your diet, from you weight, from your medical information, all brought together to tell the story of you on an ongoing basis would be tremendously valuable.

Actually, there’s another place, people often don’t see it. That is there’s a lot of talk around the democratization of data. Chicago has pushed out tremendous data sets associated with what’s going on here. But the reality of the democratization of data is that most people can’t look at that data and figure out what’s going on. The only people who can do so are people with analytics skills.

It’s really not democratization. It’s the meritocratization of data. If you have the skills you can get the information. We like doing this. We’ve grabbed publicly available data associated with city of Chicago, that sensor data about our waterfront. You tell the system your zip code. It tells you the story of your beach. We’re entering the season for people going outside in Chicago. It’s very short. We all go to the beach, but the system can say, “Here, given what I’ve seen of all these sensor readings, Oak Street Beach is not looking very good. You should go a half-mile up. You should go to Belmont.” It’s completely democratized information based upon analysis that is a data scientist’s piece of work that’s mapped onto a communication that everyone can read. [Editor’s note: read more about that project here.]

It’s that moment for us that’s super exciting, because the realization that it’s really hard to genuinely democratize data if you have to interpret it, if you have to calculate against it, but democratizing information by turning that data into short descriptions of the things about the things that are important to you. That turns out to be something that is a powerful moment for us. We just want to keep marching through that kind of space, as well.

Kyle: You’re also talking about the ability for a brand to create extreme personalization for somebody, as well, right? Especially when it becomes the smart devices and IoT and all this stuff we are talking about. That has to do with quantified self. Is there any other interesting opportunities that you see for innovation in general for what you guys do, looking towards the future?

Kris: I think that the world of innovative things as it’s opening up is going to be actually huge for us. We’ve already got one client in the aerospace space where we’re looking at data associated with what’s happening inside of a plane. We’ve got another client that’s looking at management of large-scale power stations and providing ongoing reporting there. Where again, what you need is to be able to in one instance, get to what’s happening, a situation assessment without having to look directly at the data. You might not have time to look at the data. You need to communicate with people who don’t understand the data but still need to know what’s going on with their power plants.

I think there is that to begin with. Then, in terms of the quantified self, I think that we really are looking at the ability to integrate across a wide range of devices to literally have the story of yourself where it’s delivered every day. It’s there for you. It will be delivered on whatever the next version of Glass is, but your story about how you’re doing, what you could do, how you can do better will be available to you immediately. That can only happen in terms of all of this because the data is now there, the machine has something to tell us. There’s our technology to bring it into a story. In terms of varying your terms though, I can give you a quick example of something that’s personalized and tremendously impactful.

This is a thing that we’ve done as a prototype and we do not currently have a client around it. That is looking at test data for kids who are living in a No Child Left Behind testing world. Right now a kid takes a standardized test and what they get back is the number. They get back what their score is, what their percentile is. Maybe they get historical, you’re getting better or worse. The data under there about all the questions and why the questions were asked and what the questions were looking for is underutilized. We actually built a prototype where we would look at the tests themselves and actually report back to the individual students.

“Here’s not just how you did, but here are the things that you did where you have problems where other people weren’t having problems, and things you should look to. Find in the materials a way to learn more about magnetics, electrostatics, or biology etc.” Going through and actually teasing out from the data layer advice.

It’s not, “Here’s how you did.” It’s, “Here’s how you did. Here’s exactly the places where you have problems. Here’s how you can do better.” For us, we look at that, there’s no way that humans will ever do that. It’s not feasible. We don’t have the people. We can provide a way to communicate performance reviews and improvement advice to students in a way that they can ingest. They can read it. It’s just for them and it can potentially change the way in which we treat testing and education because we’re integrating personalization that’s completely driven by a machine. That for us is an exciting thing.

James: Yes, absolutely. My day job is data analytics analyst for Blue Hill. One of the biggest trends everyone is all abuzz about is cognitive computing, machine learning, and everything. There is this huge ability that’s changing, now I can ask questions of my data just by typing it in into Loss Analytics or Microsoft Power BI, whatever it is. If you’re able to not just automatically throw out data to get someone that you think is personalized to them, they can ask a question and that adapts. It’s going to be very interesting.

Kris:  In the long run there’s a data discovery component of where we’re certainly going to be going. The nice thing about having a system that understands what it just told you at a very precise level is that it knows then what else you might be ready to ask about. It can ready itself so that when you ask questions it can easily disambiguate what you’re asking about. They get you to the next piece of information. We think that interaction is certainly in our future.

James: This is really interesting. Wrapping up, is there anything that we haven’t thought of, we haven’t talked about yet you think is pretty important? What’s out there that we haven’t touched on?

Kris: I think as we’re looking at the growth of artificial intelligence in the world, in particular in the businesses, I think that there is going to be an absolutely fundamental component of that. Moving AI into offices, into the business world. That is, we are going to need to be able to trust it. I really love the Watson technology but it only provides an answer. It has a hard time providing an explanation. An answer alone can’t be trusted.

I think that what we’re going to find is for these different kinds of technologies that are genuinely brilliant pieces of AI, they are always going to need to be able to fall back on, “Oh no, let me explain to you what I’ve done.” I think that explanatory capability is going to be crucial for partnering with AI systems and using AI systems as opposed to simply listening to and being ruled or run by AI systems.

I think that’s actually a huge component of what we care about, what I personally care about. I have a little thing about people’s nervousness about AI. A lot of people worry about somehow AI making us less special. For me the reality of artificial intelligence, the smarter the machine gets, the less we have to be like it.

Right now if I’m going to do data work, I actually need to know how to do stuff at the machine level, the things that the machine can understand.

As we make the machine smarter and able to communicate with us on our level, we can actually be more human. The smarter the machine is the more human we can be. It doesn’t take anything away from us. In fact, it amplifies us as creative strategic thinkers as opposed to making us rote slaves to the mechanics of what the machine demands of us right now in order to get information out of it.

Kyle: Perfect, thank you so much for joining us today. If people want to educate themselves on you guys and what you’re doing, where should they go? How do we connect with you?

Kris: Go directly to narrativescience.com. I’m occasionally a Twitter user @KJ_Hammond. Narrative Science has a pretty active Twitter account @narrativesci. Those are the three main places, and read Forbes. Because they have some brilliant writers working for them including a machine. That’s the easiest way to get a hold of us.

Kyle: Fantastic, thank you so much. Ladies and gentlemen, our guest today has been Kris Hammond from Narrative Science. Thank you.

Kris: Oh, thank you very much for having me. This was super fun.

Stay Tuned for the Next Episode

Subscribe via iTunes