Finding Solutions for Local News with Razmig Hovaghimian (Hoodline) | Disrupt SF (Day 3)
8:29PM Sep 10, 2018
We're going to have a great session. Next up now, you know, there's a lot of discussion about the news media these days and people talk about,
we're constantly told that there's such thing as fake news by a certain individual who shall remain nameless. But the real point about news often affects people's lives locally. I certainly once cut my teeth as a local newspaper journalist in London. And, and it was very important to people because that's really where you live, right? And so we can actually get into that subject. Now. We're the startup called hood line, which is created by with the CEO Ramsey on I'm totally going to get this wrong. I'm so sorry. Just I knew I had it. But now it's gone from my head. Raz, Raz Raz. Meg. Huh?
Hi. I'm not gonna do it.
Her give her veganism. Not like okay, I'm not gonna do this. I'm not even gonna attend. So, but anyway, Rasmus is going to be in conversation with our very own Ingrid London from London with TechCrunch. Give him a round of applause, please.
hi Rosalyn. Attempt your surname. You're gonna try. I know. Okay. I hope I
hope I gave me I think it was
10 when I learned it. Okay. So thanks
for thanks for joining us today. Thanks for having
hood line, why don't you tell us what's going on over at hood line. Tell us a little bit. So
a little bit, the genesis of the headline story. And 2016 was an EIR at Rakuten, you had covered us generously advocate. Yeah, Singapore, Japan move here. wanted to go back to an old passion, which was using data and we started a company called pixel labs. Then we merged with the hood line, which is the San Francisco local news site, we rebranded the entire company hood line. So what we do effectively, we take data signals and turn them into local content. What a combination of editors and engineers, so I think data science machine learning, and we're in about 20 cities. Today, we take about 250 terabytes of data on a monthly basis, identify the signals and right automated stories that know covered and us as artists that are out there today. Okay, so
you're doing automated news,
with a combination of editorial to set the template and the tone and variations of it. And data sets a lot of triangulation and validation. So there's no fake us here. We take that to market
Do you think you're killing off local news? Or was it already dead? Yeah, I know that question
is coming. No local news. It's been tough. Like when people talk about us deserts, they're talking mainly about not having sort of the basis of information, credible, good local information markets there about 1000 communities still that don't have them. This was about 2000 communities that had lost her local news about maybe in the last 10 years or so some are starting to get filled back end. So it was already dead, it was starting to shrink. But what the issue that we're trying to fail as try and fill those gaps, trying to amplify the signals of the stories that are out there. And when we talk with our publishers and partners, we realized that nine out of 10 things they want their own cover, they might be covering crime reports, but they want to get the beat and the pulse of the community. So data sets allow us to actually provide that data and insights to them. So they can write stories that are deeper, more journalistic. We're not a newsroom, but we don't for the signals at at the very bottom medical neighborhood, polygon level, if you will. And we write their stories very quickly. So we understand what's happening in a neighborhood. But oftentimes, the signals could be amplified. When the newsrooms are the ones that are taking it to market they can go deeper and add more soul to it. Okay. And
do you I know that right now, a lot of the stuff that you guys are doing is sort of around I was looking at sort of restaurant restaurant reviews, kind of local listings turned into better narratives, right, right. Okay. Do you think that you might try to get more ambitious over time as your algorithms become more, you know, sophisticated, and so on. So what what's really missing in local news is, you know, really hard hitting crime reporting and, you know, something uncovering, you know, corruption in, let's say, a housing Housing Association. Yeah, you know, the stories that are, you know, shedding light on things, you know, maybe maybe, you know, something that local councillors are not doing right, that they should be or shouldn't be, right now, are you able to touch that kind of stuff with what you're doing? Or is that the ambition or not, or is it more about relieving,
you know, some of the work that's being pressured onto local newspapers so they they can focus on, it's actually both,
so we are maybe covering about 10 or 12 different categories. So restaurant openings could be one, but it could be high school sports, it could be crime could be anything that's happening around you, for us, anything really would alert and long at a hyper local level we want to cover but then we can go at a city level would love to build this data out with all this data that's coming from the clicks we get when our stories are powering local discovery, let's say inside of an MSN or an ABC or CBS or Hearst. But at the same time, with the clicks and more data that's coming into our platform, I'd love to be able to find this patterns and create infographics and create sort of data sets that newsrooms can can use to go find those investigative sort of reports and be able to break stuff is that we cannot cover so we're trying to find the information in the signals and tournament two quick stories that inform like, fill the gap. Because local news is big, like 40% of searches of local intent. If you look at what's happening, people searching for what's in their damn near me, searches and quotes has gone up 400% over the last couple of years, we want to fill that gap. But we want the journalist to be able to take those data sets and stories and open it up. And my dream would be if if we are at scale, and we'll figure out the supply of content and the right unit economics for it to open up to open up our platform for newsrooms that are trying to write the stories that are interested in the data sets for any small newsroom to be able to play with that data and write their own. Okay, so you've
kind of demonstrated that there's an appetite for local news evidence by local news searches, so and so near me, etc. And yet, we've seen, you know, a big collapse in local news, right? What has failed in the business model? And what is it that you're doing that you're able to start prevent that and make it into a profitable business model? And not assuming Yeah, well,
now, they don't know we're not when we are, I'd love to come back to you and say, we figured out a model that maybe we can work with others to replicate it is it is a tough space, because people are discovering their news in many different places from your Twitter's to Facebook's, whatever sites they might be on. So aggregating that demand is extremely hard. And to aggregate that demand typically, people are chasing clicks, any story that has sort of the right headline that might be a clicked on. But there are a lot of stories at a very local level that might not have those network effects. A story in a certain neighborhood might not travel at a national level, I might not ripple if you will, to a national level. But still, it's so important. And if you add all those neighborhoods together, it scales. So we figured that there are two things we want to solve. One is a supply of content, there's so much demand, but the content isn't there. So with the data sets, turning that into content, we feel we're able to sort of fill that demand because it's their second is the unit economics, it's extremely hard to just do it purely from the editorial side. But this day and age with a lot of data that's available from the government data sets, you're talking about fictitious business names, liquor licenses, crime reports, etc. Once you start analyzing that you're giving reporters tools to be able to write much faster. So that brings the unit economics of writing one story much cheaper than he was, let's say, five years ago. So fixing supply and getting the unit economics to a level where we can make sort of a profit per story and be able to scale that as what we're going after. We're not there yet. But that's what we're trying to crack. Yeah,
I mean, I think when I think about, like, the local newspaper, when I used to live in Monterey, California, I moved to England, it was awful, you know, and that was before the internet really hit, I'm old, you know, like, um, and, but, you know, like it, you know, it was awful, the, the stories were really low quality. And I think that, you know, was dying a death already. So I can understand, if you think that you can get more interesting stuff in there. Yeah, you'll get more demand for it,
I think. So, the more we're right, we're starting to see that they're more data sets starting to come into our platform, almost 90% embark now. And the more distribution we're getting on, let's say, the right side of our platform, the more data partners are coming, there's this weird net platform network effects that's happening for us where our bottleneck really is being able to work with all this data, make sense of it identified that that the right signal to noise ratio, if you will, maybe two or 5% of the stories we have in the platform, we publish. So we want to get better at it. We want to understand what people are clicking. And the beauty of being able to say it's powered by headline, rather than headline as a destination is, it's platform agnostic, right? We are not in a walled garden, we can be on CNN or Fox, right? Like, the data can be anywhere. So we understand what people want in their communities.
Okay, now, from what I understand, you guys have two parts to your business, you've got the this these automated stories that you're generating. And then you've got the recommendation engine. Now the automated stories they're running with a bunch of big media properties, right, you're working with like, tell us so the
ABC is a technology, CBS Hurston and we're talking with others that they've been great partners. And we learn a lot from from that relationship as well. So that product, we essentially have to pay you to
syndicate your content
or no. So it's totally free at the stage that we are in, I want to learn from the data, we want to be able to write better stories. And for us, that's distribution to learn what's working, how we're addressing the demand and and get better,
you're basically offering them free feeds
of your automated, we monetize our own stories, we could have an affiliate link, we can have an ad. And I would love to get to a point where we can share that back with with the publishers because this can create a new revenue stream for them. Okay, so that's another way that can help the local publishers. So that product where it's our fully syndicated content, we call it the data wire think a little bit like the Associated Press. It's almost like an API wire, the second product recommendation product that you were mentioning, except you have to pay for an AP feed, right? Yeah, so and this one is pay for you. In this case, we're giving great local content, and we are able to actually monetize it. So free is better because we're getting distribution. And that's how we're able to get to almost 100 million story recommend recommendations a month now, 100 million, and would love to scale that we're in 20 cities would love to be across all the DMS and the US the designated market areas, roughly, I think, the top 50 or 80% of the country, the second product is in module or recommendation, think about it, maybe like an outbreak for local except it's not ads, it's actually content. And we get tremendous amount of clicks on these because you are, let's say on a crime story, you see another crime story in the same neighborhood, or you are a different type of story, we recommend something that's very local, in your neighborhood or in your city. So that's how we're taking a second product to market either take the full feed, or just put a module
and then the module has like ads in it,
right. So in the stories, because since that our stories, there was ads, but we're also experimenting what an ad let's say every five story is every 10 stories or so.
Okay. So if you're working with all of these media organization, you know, obviously, one of the big areas where people are going for local news is also Google. So Google Maps, and just through search engines, and so on. So how much are you talking to? or looking at the likes of Google or Facebook, which has put in a massive local news push. Now, Twitter Hello, likes to think of it itself as the town hall. So where where are you on? Those are those conversations,
I think there's two ways they have so many signals will love to turn into content at the same time, would love to put the content we're generating does automated templates from other data sets, let's say the government ones on the platform, because it's clear that the demand is there. People want to understand local see local, so we want to be platform agnostic, maybe we're inside of moments, or we're working on events with Twitter, or maybe we're in the local feed Facebook is doing quite a bit, actually, a lot of things are experimenting with. So we're having conversations actually, with all of them. And I'd love to be able to just personalize those algorithms to show you those five pieces of magical content for you and your city. And let's say within half a mile off of where you are,
is that what Facebook is talking to you about?
We're talking about multiple things. I don't want to I don't want to speak for them. But local is definitely important for them.
Right? I mean, would they would they want to the thing is, is so far, a lot of what they've done has been about partnering with third party news organizations and trying to enable them, they seem to be pushing back on that now, you know, if you're, if you're a publisher working with Facebook, a lot of a lot of publishers now used to get so much traffic from there, now, they're not so. So I would, would you be somebody that they would potentially partner with, to replace some of that content
Do you think it's replacing, but if we really can, as I was saying, understand the pulse of the community, and what's happening, sort of that beat is important to them, they want us to is, and we almost in a way, if they're the social graph, we're trying to map a very hyper local graph of what's happening around you. And those stories will be valuable, but local for them. I think the definition might be a little broader, is not just the stories to publishers, but there's anything really would would location on it, it could be thanks to see things to do things to buy with merchants that are on Facebook. Okay.
So another kind of interesting development and local news that I've been noticing OUR SHOW YOU HAVE TO is this whole idea of citizen journalism. So it's not not about taking all those data sets, but about people on the ground and, you know, doing their thing, and you're seeing a lot of that coming through on, you know, Snapchat and Twitter. Big deal there, right. And Google started something earlier this year called bulletin, right,
yeah, bottom up news. They're getting another
Yeah. And it looks like I can't quite figure it out. But it looks like it's because it's just invite only now. But it looks like it's partnering with local newspapers and bringing in data right from from local people to crowdsource to fill those. Yeah, it's great to see
Google put so much effort into it, too, when they had their initial launch event, we were there and they earnestly
Are you were there at the launch of bullets.
Yeah, the line was there were talking with partners, they wanted to understand sort of what's on the what's out there. Part of what we do also is we have a tip line, we literally have phone numbers, if you're allowed to in Oakland, or San Francisco, where we get citizen sending tips, and we can turn all the tips, let's say, on a daily basis, into a template, it can be an automated template. So there is on the ground reporting that happens that way. I would love to be able to do it and in other cities to I think that's, that's a bit of a parallel with what they're doing. They're trying to understand on the ground. What's happening. Okay, how do you
how would you work with something like that, that was headline because you don't currently use that in your new or something like that. What
I would love to do is be able to take those signals and tournament two stories again, right, the left side of our platform, you all and then across Google, there are so many other places, including possibly YouTube because we do automated videos now. Yeah, so our, our templates, arm videos, as well as stories, articles, have distribution, just reach go to where the users are.
Okay. Congratulations. You've just made a perfect segue into my next.
Okay, so previous company. I know you of course, from your previous startup, Vicki, and which was a Why don't you tell us about that he was
a sort of sure Vicki was a few other Hulu or Netflix for rest of the world who started in 2010. And the whole premise was
content arbitrage effectively, we would license us shows, let's say for everywhere except the US because it would be $78,000 per episode here per year. But it's only $400. In Asia. We would do the same for Korean dramas, Japanese anime, we played with the data just like we're doing here from us. But there it was for TV shows or movies. And we built tools for a community of translators to create subtitles for free. So you can get a Korean drama and Arabic open up the Middle East, or you can take Let's end it movie, put touch on it and open up a new market. We grow it and Rakuten acquired a company I think you covered that you know what us from the early days.
And it's, it's, it's been great, like a Vicki's thriving and I'm not there. So maybe that's a good thing.
But it's great because for us was being able to build cultural bridges. And we feel like it's really important for extra medium of content, people understand it. And for 40 out of 200 languages on video or endangered languages. So the translations were actually one way of saving that language. So
is for me is quite interesting because you were you went you went into online video at a time when it was really just starting to take off I think, you know, it became really the buzz the buzz word, you know, at the owners of TechCrunch, you know, only wanted us to do video for a while, didn't seem to care at all about writing, you know, so, you know, when you think about that you guys were slightly ahead of the curve with with that you were there when YouTube was already big, but it was only set to get bigger before Netflix.
Really boom, before Netflix opened up there. Yeah, in fact, we're taking our content and our revenue model was to syndicated back with subtitles to Netflix and Hulu, right. That's how our revenues grew. Until we started aggregating that demand and having our own video is a medium that's just so universal now, okay, we started realizing that if we're going to be in every market, we have to ride that wave of like, open up Latin America, open up Europe. So video as the way of doing that. And also it allowed us to grow because we created the embeddable player that travels with the subtitles, yeah, in 80% of our traffic was from outside, like, fans would take the videos and embed them on their own fan sites. And they had their own followers. So there was this network effect there as well. Yeah, and you still
got the traffic from that. That's right. It was coming
directly on the player. And we would monetize that later. But it was just it was a hard thing like local is harder to be frank. And maybe that's why we're doing it. But being able to convince studios that fans can create quality subtitles and put them on was extremely tough. And they're worried about piracy of licensing and nationally as well. Yeah, there's a problem. Well, what
were the lessons from that, that you've kind of applied to building a next startup? Would you say, Are there any? Or is it so different?
Now it's time to data, I think, like we wanted to understand initially, when we launched fixer labs, we're trying to go consumer ourselves. And we're working with Twitter's data to see if we can understand local what, what are the local stories, and we realized only 4% of tweets, even if that had like location on them. So starting with understanding where the demand is, and having a point of view of what you're trying to solve, especially when your market making like you're not sure exactly how you're going to fill that gap. It takes a lot of experimentation, and it takes a phenomenal team. I think that's the fastest way if you're careful here building the team from the early days, and a shout out to what 27 of them were an orange Harrison and the mission come say, Hi,
do you think that headline might ever tried to tap into video itself? I mean, right now you're mainly are written were written, but we're doing automated
news, we're doing automated videos as well. In fact, for articles, I think we're going to get maybe in a month or so half of our article stories might have an accompanying video with it with those he like,
automatically generated as well.
They're automatically generated. In fact, they're easier done than the text versions, you can imagine. Let's say if it's a restaurant, opening a picture of the front of it, the picture and picture of the chef and let's say the dish with some ratings and an exact map and imagine
like slideshow type videos, rather
than there are some footage as well, you can do like we were working with some third parties where there's footage in a neighborhood level that we might be able to interlace. But yeah, it's it's not full video footage. But they're actually doing well with tests. And it's in high demand from the publishers, right, everyone wants more video inventory, because that's one thing they can monetize. And our videos are 15 to 60 seconds. It's early, but that's an area we're focused on. And the funding will help with that. Yeah, although I
guess if you're gonna, if you start putting in more crowdsource data, you could create a situation where you're asking people for more video that can become the basis Yeah, that's, that's very
possible. Yeah, automate that to be able to find the quality segments. And so that's what are the machine learning part of our team can identify the right foot edge, right images? And you've been proving
that and have you had a request for that sort of stuff from your publishing partners? video? Absolutely.
Right. It's pretty universal, I guess, demand for that, so
and so now, what is going to be next, then? Is it video? Or is it more sophisticated news? Or what? What do you think
learning from data style, but those two products, I feel they're different. And market the the wire product and the recommendation one in terms of content, both video and articles are important because people consume content in different ways. So those are areas that we see interest. But the more sophisticated we get with our data sets, the more validation you can do, and the better decisions, a better algorithm. So I think going down the path where we're able to personalize it for you and recommend better stories in your neighborhood. It's going to be important what I was saying, we're trying to solve unit economics and the supply I feel we're going to figure out the supply. In fact, that might be too much supply, like, how do we find just the right pieces of content
for the users? Yeah, I mean, I think it's kind of an interesting area, because you've got
the, the media is your endpoint here. But you what you're doing is a model that is being used in all kinds of things. There's this organization live stories, which is doing this with government data. Yeah, the Civic data on the East Coast? Yeah. Is that something that you might also look at, at some point to, you know, not just doing this for news, but for other kinds of organizations, not just media
interests from others? So what I think
they're aggregating data really well, at a city level e commerce, commerce could be possible to because if you understand sort of where the story is, and there are something that just opened around you, we work let's say would would coupons data. Yeah, we can talk about a deal that's right
around one example, there's very little reason to visit Groupon unless you're looking for Groupon. But if you create more narratives
in there, it naturally bubbles Exactly. It's in a way, almost like a local search bot, and reverse or things are showing up based on Western where you are on the Civic data point. It's extremely important, right, because when we're talking about news, deserts, even those publications that are still writing local, apparently, 17% of the stories are truly local, the rest is still this national stuff that they're getting in and the more data and information is available, the more they can start writing about their communities and their neighborhoods. So we absolutely love their data and that group, so that has done a really good job of mapping and I'm visualizing that data and that data, raw data and even the infographic so that could be very helpful for for newsrooms and from groups like us to make our stories richer. Okay. Something we
haven't really talked about. But I, you know, you come from outside of the US and Vicki had a co headquarters in Singapore, you yourself. You're Armenian, Egyptian? That's right. very exotic. The hard name Yeah.
What would you can you envision a kind of headline being applicable in other markets to I mean, it's just such a different or is it just such a different climate
in different places? No, no, I think I think the need is there. I mean, about 10 years ago, between my first and second years of grad school, I was working with the United Nations, literally, in Sudan, and South Sudan, before it split into a country and the iPhone was new around that time. And I was thinking, can you create a world map where you have stories that are not on other news outlets, not 70% the same and that was important for me to be able to do something at an international level. And the need is there be at a different market at a different country, if we can have the data in those markets. The way that we built our our machine learning models and a data science team is set up if we can get local editors to work with I think we can cover other markets, English speaking markets would be easier. But you saw Vicki, I started very international, and then start picking the markets this time around. We're starting in the US going in the other direction. I learned I learned a lot in the process. But
yeah, your algorithm is being taught,
I have an actual coasts uses everywhere.
Yeah, I'm in the other countries outside of the US. Have you had any interest yet in this sort of thing? Yeah, Europe and
Europe, Australia. And there's a lot of the inbounds we've gotten including actually Singapore, we've had interest from mobile apps where that's a messenger apps were thinking of having like a nearby button. Literally, you click on it and see what's happening around you. You travel to another country nearby, you open it again and see what else is happening around you and being able to power that but it's there's just we have to focus like we're still small and now it's an exercise of being able to write quality local content, being able to do it at the right unit economics then scale. Yeah. Okay. Well, thank you very much. Thank you so much. Appreciate it.