S5 Ep120: How Decisions Enabling Market Intelligence Works With Raj Neervannan of AlphaSense

Great. Great to have you be here and thanks for having me here, Nicolar and Shalat. It's a wonderful morning here and a wonderful evening there, and it's always good to chat with people who are in the know with the product that we work on and we're so passionate about. You must have seen the pain that you go through looking for alpha, and I'm sure your audience knows alpha is above average return for same risk between two assets. That's the holy grail, the thing that everyone looks for. We are happy to hear that you're from this industry as well.

Yep, absolutely. It's a very big question; I think I loaded four different questions in the same question. But I'll start with your own background. You have been feeling what it is like working under the gun with limited time, trying to look for information that will move the market or derisk something. You have a variety of hypotheses in your head, and you want to get to it. Your choices are you go to Google, or typical processes before we came on board were terminals like Bloomberg or other information sources with structured content, balance sheets, instant news, and a library of content. Tens of thousands of sources, then you Google finance, Yahoo, or other tools. The typical search process: my co-founder was Jack Koko, an analyst at Morgan Stanley, and we met when we were doing an MBA at Booth. I've been a serial entrepreneur, and he has been one too. We have been through this multiple times. When we met, we were keen on working together and brought complementary ideas. He mentioned that he used to open a document, search on Google, get 10 out of a million, download one, then control-F to find what he was looking for, highlight it, go back to the next one. You must have done this in your universe. I thought this must have been solved before because Google has been around for 30 years and search engines for a long time, and so have other large vendors. When we looked into it, we realized they solved it as keyword search. You look for a keyword, it finds it. It's more of a keyword finder served towards the internet user looking for a toothbrush or a ticket. It's not meant for a professional like you or your audience. Even for people looking for any hypothesis, it's not easy to get. It's easy to find documents, not easy to find insights. Those are two different things. The main difference is that information sources like CB Insights get you 30-40% of the way and then leave you. You have to go through the remaining 60-70% and you give up along the way. You think you've done enough and move on, but you end up making only 20-30% more progress with a lot of effort. So there is a big problem with finding enough and finding the right information fast enough. We thought this process was highly friction heavy and needed to be addressed. Some people had given up. That's the pain point. Today, AlphaSense is called a market intelligence and search platform, a loaded term because everyone calls themselves market intelligence. But truly, a real intelligent platform helps you find insights quicker, make decisions faster and smarter, and hopefully not miss the most relevant information. That's what we strive to address. It's a long road, but we have made good progress. The technology can come into the picture, and that's where my background came in. Having done this many times and with a bit of finance, I understand the pain point with my co-founder. We can solve this by providing the context of what they're trying to search. We help you find key ideas and topics inside documents and spotlight them for you to quickly get to and move on to the next one. Instead of giving the top 10 documents, we find the spotlighted sentence context on the 13th page of a document and then go to the next one, and we give you a summary. So you are quickly reading a series of snippets and contextual ideas instead of going through all the friction. As you read, you can zoom in and out and find exactly what you are looking for and get to that end state of having enough insights to decide which is relevant. That's what we do.

The differentiation is that we are heavily technology driven. As opposed to an add-on for one more keyword or search thing when you have a lot of content, that's the main difference. In some ways, it's akin to Amazon. They sold goods, but their presence is not because of selling goods; people had Walmart and other retailers. Amazon used technology to deliver quicker, find things you want, and the online experience was better. In many ways, I think we are like that, but in the business world, it's completely different. We have wide appeal with this kind of scale to a lot of different content sets. We see the vendors you mentioned as partners who provide us content, and we leave the technology heavy duty of finding search to us.

Yeah, 100%. First, Google indexes content for consumers, whereas we index it for business and financial research for knowledge workers. It could be anyone from a hedge fund to an investment bank to life sciences. Life sciences loves us because they do so much in-depth research. Anyone that wants to do serious in-depth research loves us. For that, not only is the search important; content search is only as good as the content you search on. So, the information they look for, if you're in life sciences or any specific field, they must have tons of content, specific blogs, specific sources they listen to. Absolutely, that is one of the main additional differentiators. We collect lots of quality sources everywhere. It's not just public content sources; public content sources are overly searched because consumers and internet engines give you popular results. A 10-K that came out is not very popular unless you search for it specifically. It's not about popularity with us; it's about our customers' need for insight. Insights can be anywhere whether popular or not. Quality sources are quality sources, so private sources are quality content. We go find them. Clients tell us, and we look for all the research content from brokers, all the insights, verticalized industry sources, blogs, news, transcripts of earnings calls, anything spoken that is useful. Another friction point is not just search; finding the content sources also becomes a friction point. People have to pay for this but not that, and go to other sources. There is entitlement with paying content. We want to make that problem go away. We buy everything and anything useful, handle all the payment, and give you a flat fee. So you don't have to worry about entitlements; you just get great content, great search, and you get to it. Our strategy has been to enter into multi-year long-term contracts with our partners and make sure they are involved in our ecosystem and able to reach their customers through us. We also have our own expert insight network; we acquired a company called Stream last year. Stream conducts in-depth expert insight interviews on all topics of interest, and we collect that content. This used to be one-on-one and cost thousands of dollars; now it falls into the flat fee model. That's our model for getting all useful content, whatever the format, and giving you a flat fee.

Well, to put it lightly, as humans we have been seeking answers for a long time, so I don't think we are going to stop even if there is an expert. But I think what we should strive for is to make the process as frictionless as possible and get to where we want, good enough or great enough, so that given the time and energy you have, you have done what you could. This tool gets you a lot further. To say you will be doing 100%, no one is going to be able to do that. To understand why, we try to make the tool as close to human as possible; that's the utopian goal. In order to do that, we try to understand the question. That's what we do in search first: understand you ask a question. You could say earnings or sales; I interpret that as revenues or backlog. We map that question to a lot of different concepts in our mind. That's the first step. Second, there are possible answers. In this case, it's an engine that has not just two or three facts but billions of documents, already read and contextualized and tagged in enriched ways. So when a query comes, it can match it close enough. It's not just keyword level; it's super enriched. If a human listens to a question and reads a document, they would enrich it in their mind and answer back. That's the process except it's a machine. So we do that beforehand and store the enriched information. Third, a matching process between the enriched query and enriched information. Fourth, not just matching but ranking and sorting to give the relevant ones in a format that makes sense to you. Those four steps are what we do. There's always an element of human after that. For example, sentiment: we have a sentiment engine and understand tonality, but it's constantly changing. The contextualization requires proximity searches. You ask one thing, and the answer could be a few sentences below. So we show you result with context around it. All this variance of thought processes has to be algorithmically implemented so that AlphaSense understands greater context, even trending themes. The engine tries to understand the context and does it religiously back and forth. That's where the value comes. It gives you the actual stuff very close, and then you may differ and iterate a little further.

Yeah, I think they see this as complimentary. We are not trying to displace them. There is tons of structured information available that they have been using for a long time. We want to add value because so much value is in the portion that is 80% of alpha, the unstructured information that is not even tapped. That's the job we want the machines to do. Our users have developed trust with other tools and will continue to use some of those. We have similar content here but use it differently in surfacing insights. We pull a table quickly and compare across multiple tables in an AI-driven way as opposed to an Excel spreadsheet way. Sometimes they feel that what they were trying to do with another tool, they can do with this, so they use this instead. But that doesn't mean that other tool doesn't offer other benefits. So the biggest banks use both.

Great question. That's the beauty of this model. It used to be that a company wanted to speak to an expert and a mediator would set it up for thousands of dollars, and the discussion would get buried. In this model, we pay the interviewee some money, but also the interviewer gets paid enough and gets access to the content they contribute to. They get useful information in return for giving their own.

Right. So we seek experts based on topics, research who can be the best one. In some cases, clients mention few names. There is an army of people in our organization whose job is to sort out what topics to speak on and where clients contribute. We have our own industry knowledge looking at what's trending. That's one input. Second input is who is well equipped to answer those questions, contributed by clients, experts, and our own research. Then we have a team of people who are knowledgeable enough to interview the expert on the topic. For example, you and I discussing: you come from this field, I come from this field, so we can go deeper. Similar to that, we bring someone to interview, or we interview ourselves depending on the topic. It is not a marketplace where random people come in and discuss. We want to control the quality of content, so we curate the experts. We have interviewers also curated, and they discuss intelligent topics that matter to people. The difference is it's democratized enough that the price is low enough for the flat fee model.

Great question. I'll quickly go to the office side because that's where the proximity, sentiment, and all the algorithmic things are done. That's done more from expert teaching the algorithms in a broad sense, but also experts in each industry saying here are the topics that mean specific things. Apple in tech is different from apple in agriculture. That's the whole AI thing, general knowledge and common sense. That is done in a more tech-centric way to scale. There is a process for figuring out new concepts and changing contexts. There is an army of people and engines for quality control. When it comes to Stream, it is done in a more in-depth fashion. Our product represents across industries, so it certainly covers many. Financial services has a lot of depth because AlphaSense started off there. It tends to be way better because it started earlier. As it grows, it gets deeper and broader. We have the highest number of interviews added monthly, more than any other platform. As more topics are requested, it gets broader due to sheer volume. We are the largest provider of this type of content, and by doing more and more various topics, we become broader. Our focus is to be the broad-based expert insights platform, but the interviews go a lot deeper.

Great question. This is where humans come into play. Humans have artificial general intelligence; we computers are nowhere near that spectrum. Humans can read something hand scribbled in a filing and understand it. But if you look at point solutions, we are dealing with AI in a narrow sense. Making a computer understand text was a big stretch from keyword to doing all the things we do now. That wasn't doable 10 years back. We have made a lot of progress from numbers and structured tables to making computers make sense of text. That's a huge step. But the text may not be in an organized form, not in XML or computer-friendly form. AI now understands pictures and words even if not as characters. If you can write the same letter in five different ways, it needs to understand that it's handwritten. Computers have to be smart enough to figure that out. Understanding tables from PDF or image is something we do with table extraction. Computing technologies are making inroads into point solutions that try to understand aspects of human perception we take for granted. We apply that not just for text but also for human conversation and other abstract discussions. Computers have made a lot of progress with image recognition, self-driving cars. We apply similar techniques in table recognition, OCR, etc. Second is voice: today you have Siri and Alexa understanding what you're trying to say. Many podcasts and discussions can be transcribed reasonably well, so humans don't have to sit and type as much. We still do a lot, but we can get help along the way getting consumption out there. That can be searched, auto-corrected, maybe even fact-checked. There are a lot of things you can do from a call: what was being said, was it referred to something else, verify numbers. This area of transcribing and digitizing raw analog into digital form is where the biggest gap was. AI can help bridge that, even if harder.

Sure. We built this company based on our experiences. We wanted to make sure our investors are our users. Often, our investors are our users; they come to know we exist because they use the product and say, this is amazing, how can I help further? They are also content providers; they create a lot of content and contribute in a monetized form. It's a healthy relationship. We see them as partners and users. Almost all of them use our product heavily. That's the context. The biggest synergy is that they understand our long-term vision. We are constantly making the product available for lots of different content. For example, if you are a large company creating a lot of content for internal consumption, you need a tool like this to search that content, and you can't upload to the cloud. AlphaSense in a box or coming to you would be a great way. There is a lot of room for technology to streamline that information and help in more ways than one. Our customers, we never know who will become investors, but almost all are users.

Thank you.