How to Secure Investment from Top-tier Investors in AI

00:01 Desi Some would say that there’s an AI startup for almost anything and that’s probably true. But one industry that really needs it is media and fake news. Dhruv Ghulati is the founder and CEO of Factmata an AI startups aiming to restore trust in news media. He started Factmata in 2016 as a university project and the company has already been backed by world-class investors like Mark Cuban and Craig Newmark, the founder of Craiglist. Can you imagine that?! Here’s where Dhruv is taking Factmata next.

00:52 Desi Hi Dhruv, great to have you on The Product Show!

00:56 Dhruv Ghulati Yeah, you too! Thanks for inviting me!

00:59 Desi First, before we kick off, I just want to say how excited I am about our conversation. I come from a journalistic background and 10 years ago factchecking was the most dreadful and time-consuming part of my job. So if someone had told me back then that a product like Facmata would exist, I would have been very excited! In 2020, when trust in media and social media is declining so rapidly, it’s amazing to have an AI product in the content space. Tell us how you started. How did you come up with the idea? What’s the story of Factmata?

01:38 Dhruv Ghulati Back in 2015-2016 I went to UCL, University College London and I was working in the Natural Language Processing lab and I was thinking of interesting applications of natural language processing and one of them was automated fact checking. The reason is it’s quite interesting is that it requires human reasoning. To be able to do fact checking, you need to be able to do not just look at the text you’re looking at but actually find evidence from outside of that text. That might form evidence for your reasoning and your fact checking, your argumentation. So it’s a really hard problem, it fascinated me as a problem and being able to apply it to social good content in terms of checking what politicians say and making sure that there’s some protection for the ordinary citizen. It seemed like a really exciting thing to do. And really Factmata grew out of that thesis and has evolved to be the business conception of the way you could take that research and bring it to a new media platform of the future.

02:58 Desi So it started as a university project, that’s very interesting! How far you have come. You were in Techstars in 2015 and Entrepreneur First in 2016. Were they good bootcamps for becoming a successful entrepreneur? How much did you learn there?

03:13 Dhruv Ghulati Yeah, I think they’re definitely very, very good. They’re very good at the early stage. They’re very good at telling you the tips and tricks how to get something up and running. Whether that’s getting early traction understanding how to best speak to your customers and find those quick wins, thinking how to prioritise features, thinking how to delight customers and product-market fit. Those are the things that accelerators are very good at doing, especially when people don’t have that natural skillset. I basically used those experiences to… It definitely accelerated me in being able to launch Factmata without having any support myself, cause I knew those techniques to be able to rely on

04:00 Desi When you were starting, what was your go-to-market strategy? How did you acquire your first customers?

04:07 Dhruv Ghulati We had this platform-focused approach which was essentially let’s first build the system that is able to take customers data and score it and find hate speech and hate news and give it back to them. So I guess our go-to-market strategy was relying on expert consultants who knew our customers and had relationships with them. Because we were going into a space where I didn’t have any experience, I’m talking about advertising technology, AdTech. So we would find some AdTech consultants and they would get us in those first trials. What we then did, essentially the trials were exactly the same. The customer would give us some content, some urls. You’d then got and score them and say which percentage did you think contain propaganda. But then we came up with a really templated version which essentially even if we did one trial you could argue that the customer should be happy, they’ve seen a case study from another trial, they shouldn’t be bothered. Every single customer want to see if it works on their own data and so we had to go through lots of this different trials but what we did was. We just had a very template report that we would give out to people that had all the key questions we knew that they wanted to get answered. Accuracy rates, sample things that we picked up feature developments, some ads that had been placed on content that had propaganda. That was our kind of go-to-market strategy, creating resales reports.

05:53 Desi At the moment, what’s your typical target audience, in terms of customers. Is it mainly publishers? PR companies? Who are they?

06:02 Dhruv Ghulati Yeah, that’s been a journey to kind of figure that out our key customers are PR agencies and brands that have communication needs. And that has really evolved from where it was in the beginning which was ad exchanges and ad networks. Really big brands that have constant opinions spreading about them constant rumors starting and they would actually lose marker share and revenue if they’re not able to manage the narrative.

06:41 Desi So fake news, let’s talk about fake news. Tell us about the main signals your system is monitoring I’m pretty sure it’s much more complex than just answering in one question. But I’m just very curious – how do you start? Where do you start from? How do you identify this is fake new?

07:01 Dhruv Ghulati I think that the key thing we started this process is saying. As a startup with limited resources in the beginning what’s the quickest win we can have with the product that adds value I remember this anecdote from one of our investors who built Brightmail, which is an email spam company. When he built the first email spam algorithm, it was operating in about 40% accuracy, but it was the best algorithm on the market. So it was good enough to get clients and deals and then improvements and iterations later on our strategy is being let’s build what’s better than what’s out there on the market and not get too hecked up about perfection and accuracy. Let’s get into the client, work with them and use the training data and the feedback they’re giving us to improve the algorithm. In short, the way we though about detecting fake new is a signalling mechanism to a human analyst. Can we give them lots of different signals and the more the better, so they can actually analyse and see for themselves – this is actually fake news. They’ve got signals for them to put a filter to maybe ten potential items. That might be is the language propaganda, is the language racist. Is it a clickbait title, is the language non-objective. Does it contain hate speech? Is the language controversial? We have now about 10 different signals and it’s purely based on the language. That’s actually a unique thing you can look at spread patterns, who’s spread the content and so on but as a company strategy we thought actually. Because we started with natural language processing – let’s start the best in the world with that and let other people build other things that we can tack on and partner with and that’s how we’ve basically taken it on for the last 3-4 years. 

09:20 Desi Saying that an article is right-wing or clickbait is a very subjective thing, isn’t it? Sometimes it could be who’s written it, sometimes it could be who is reading it or sometimes it could be just bad writing. How do you manage to stay impartial, especially when politics is involved? Have you ever been accused of being politically biased?

09:44 Dhruv Ghulati No, we actually haven’t been accused of being politically biased, we have competitors who definitely have. And that’s because those competitors are actually using human being to rate articles human being to rate websites. We actually don’t have any human beings in a way, but the algorithm certainly you can argue has a bias and that’s because people train that algorithm. But the way we combat our bias is say. First of all, we’re not rating a website based on its history the age of the website, the author, the person who wrote it, when the wrote it, what topic is about. We don’t really care about that at all and that’s why I told you we only focus on the language. All we’re saying is that everyone should be judged fairly purely on what they write. Not who they are, not what publication they work for not what country they’re at and that’s why. Even though you can build a really accurate algorithm, you don’t want to actually do that. You can sacrifice a little bit of accuracy for being fairer. And saying – look the language of this is written in a very impartial way and the way we trained the algorithm is we had journalists from about 12 different countries, about 50 of them go and rate different articles. We actually don’t do left-leaning or right-leaning, we purely do. Is this propaganda? Is this hyperpartisan? Is this extremely one-sided? Or gender-driven or hyperpartisan as an article. It’s a binary classify rather than left-wing or right-wing. Because even any person politically, left-wing and right wing differs in different countries. That’s a purely political decision, it’s purely putting yourself into a different camp. Whereas something like “You have said something that’s very politically biased” It’s really biased one side or the other, is much easier. 

01:48 Desi How do you currently identify opportunities for building new features or adding new functionalities to your system? Do you rely on your gut instinct? Do you have a system? Do you work closely with your partners and they told you “That’s what we need”

12:06 Dhruv Ghulati I was a product person before building the company and one of the things that I’d say is sometimes I do have a bit of a contrarian view on product management. Because when we were building the business and we’d say “Hey we want to tackle fake news” Implicitly you’re actually taking an estimate, an educated bet that people want to tackle fake news. This is the strangest company, because you’d think that everyone wants to detect misinformation. And wouldn’t it be great to have something that just checks your articles, as you said in the beginning of the interview, and tells if something is correct or not. The challenge is that: A) that’s gonna take you many, many years to build and then you gotta say – would you pay for it? The problem in this sector is that the technology is so hard you also can’t get upfront payments from people, because they all say of course I will buy this product, when it’s ready. The challenge in product management here is you actually need to have some innovation and have some ideas of your own on where you think the market is gonna head. What product features are gonna be needed by the customers in the future. I’ll give you an example. Four years ago when I was starting up the business I said we’re gonna have to just look at urls and content and tell you If the content is biased, hyperpartisan, it’s really politically biased. We want to remove that content. A lot of people were saying at that time, our competitors were saying “Actually let’s just build lists of websites” and we will rate them. Those competitors – great! They got some deals and they got some traction. But what I knew it was gonna happen is that the Adtech world and the internet operates at scale so just giving people a list of websites will go out of date naturally and we had to build something that’s a little bit harder which is analysing an article for its complexity around bias. Because I knew that at some point those customers we were selling to would say: We’ve got the short-term solution but we need something that scales now I think you have to be willing to give up a bit of early revenue knowing where the market is heading in the product.

14:43 Desi From your experience so far, what do you think is the Aha! Moment for Factmata?

14:49 Dhruv Ghulati When we first started the business about 2017 it was more of a social issues, governments were talking about it. There was some interest but there was no real budget assigned. Our sales strategy was probably wrong, because we were asking these customers. Hey – give us your content, we’ll score it and will find something and then hopefully you’ll like it. We had to get NDAs for the data. We had to prepare a data sample, they didn’t know where to get it from. There was a lot of friction in the process. Our main Aha! Moment was saying – “Look, why do we need to get for those customers let’s just scrape and scroll the internet every day and find things and create reports, even when customers don’t want them. We just show them. And suddenly they get amazed what they could find some fake news about them or the could find a narrative that’s evolving and they never new about. And then they were ready to get their credits cards our and sign up for the full licence.

15:53 Desi You have an impressive lineup of investors. Just tell us how you did it! So we’ve got Mark Cuban, we’ve got Biz Stone, the co-founder of Twitter Craig Newmark the founder of Craiglist. So how did you do it? What’s your secret? 

16:09 Dhruv Ghulati All of my investors came on from cold emails I basically emailed a lot of people, I messaged them, I annoy them and occasionally… People generally… And that’s been so humbling for me as a new first-time company founder with a big goal for my business. People are very, very helpful on the internet. If you have something to say if you work hard and you’re building expertise in a certain topic instead of having an imposter syndrome, there aren’t that many people who may have the expertise in a specific niche subject. I feel that through Factmata and what we’re building, we do have that we’re one of the top companies in our space. It was not too difficult for me to get a reply back from these investors. As the business goes on, obviously there’s a bit of structure to this but my personal philosophy is that the world should operate in this way. You shouldn’t need to have to take 4-5 people out for drinks and schmooze them to be able to get someone to care about what you do. You should be able to reach out to them and if people don’t like it. That’s fine! We should be operating in an objective, factual way about anything. And that includes raising investment. Ease of access, transparency, being able to reach out to people and to connect with them is the philosophy I’ve had in investment. 

17:48 Desi What’s your vision for the media industry and the big publishers? Do you think that they’ll be relying more and more on AI?

17:56 Dhruv Ghulati The media industry is a real tough nut to crack like I said there’s a lot of things against it VCs don’t want to invest in this space, quite frankly it’s not like investing in SaaS software for financial… Quickbooks management or something or investing in Zoom. The media industry has that issues where there’s not that much funding dollars coming in. Post Facebook and Twitter no investor thinks there’s gonna be any other new media company. They think it’s just a dominating market, monopoly. And a lot of investors made a lot of money out of Facebook so there’s an incentive to keep that going, right? Number 3, journalism itself, there’s a pride in doing thing in human ways in that industry. Just the people who are the execs in those media companies they want to do the things in human ways. My vision for the media industry is that regulators will gonna have to get involved to push some changes that’s gonna mean that platforms are no longer monopolistic. And they’ll be able to get broken apart in some ways, or the news part of their business are forced to have to compete. I think the media industry is gonna be a lot of individuals writing content, rather than The New York Times or The Guarding writing content, it’s gonna be us having blogs, podcasts, shows and project our own content. But in this world where everyone is a creator, we gonna have to have way more quality control than what we have. Facebook’s got billions of dollars spent on quality control but what about the wider internet? Someone will have to build the layer of infrastructure that does quality control and is able to understand objectivity, quality and moderate better. That’s where I see Factmata being that layer for the internet everyone is using it – when they create a new social network or they’re creating a new news aggregator, Factmata is there empowering it

20:12 Desi Brilliant! Thanks a lot! That’s been a very interesting conversation Have a great day!

20:18 Dhruv Ghulati Thanks a lot Desi! Bye!