Jack Kokko

The through-line

Everything Kokko says traces back to a single frustration from his first job out of college. Working on tech M&A deals at Morgan Stanley in San Francisco during the dot-com boom, he found that the pace of the work outran the tools available to do it: "I remember some client boardrooms where I'm sweating and barely awake, but also afraid of what did I miss and what am I going to be called on by the CFO or CEO of that company where I just missed something in my analysis" [4]. What made it worse was the contrast with consumer technology. "We all use Google all day long in our daily lives but nobody had built a Google for finance or business and that's what we set out to build with AlphaSense" [1]. That gap between what a professional needs and what a professional gets remains his organising idea, and he still measures the world by it: "I know a lot of people still today are doing their jobs with fairly manual tools" [4].

What has shifted is the ambition of the answer. The first version was a semantic search engine that understood financial vocabulary; the current version is closer to what he once described internally as an oracle. "I remember telling our team five, six years ago about how our system is going to be this oracle that you can ask any question in human language and it'll understand your question and go do the research, come back with a great answer. Frankly, I had no idea how and when we're going to get there" [4]. Large language models arriving felt to him like "sort of a Cambrian explosion of opportunity" [4], and the mission has widened accordingly, from serving investors to serving corporates and, in his framing, the market as a whole: "you can imagine you know the whole market benefiting from better intelligence better insights information and that's really our mission" [1].

On the tools not being up for the job

Kokko is precise about what was actually broken. It was not the absence of data. "Certainly we had the data terminals that a very financial professional has still today. That was a big part of the frustration that you could go and manually look for data but it was very hard to consume it at the scale and speed that was needed" [4]. The analyst's problem was breadth under time pressure: catching up on a new industry, new companies, and "the sort of cross-sectional information across different industries that you really should have known to be smarter with your analysis" before sitting across from executives about to bet billions on a deal [4]. He describes the prevailing method with something close to disbelief: "People were still control F searching for individual terms one at a time in PDF reports. Shocking but that was how work was done" [4]. The residue of that period has never left him. The boardroom fear of having missed something, he says, "has stuck to me, frankly, still every day when I walk into a boardroom" [4].

On building AI before the models could do it for you

He is careful to correct the assumption that any of this was easy once ChatGPT existed. "We did have an AI first vision from day one. The AI of that day was just a lot simpler" [4]. The early system was a semantic engine that "understood millions of terms and link them to core concepts. Understood that revenue is the same as topline across the whole vocabulary of finance," so that it could reliably surface every data point on a theme whether it appeared in a Japanese earnings call or a US SEC filing [4]. On top of that sat narrow classifiers trained at considerable cost: sentiment on a single statement in a paragraph, whether a broker report was an initiation or a price target change, company recognition in text. "We had a team of dozens of people in India tagging very large volumes of those statements to be able to train those models and train it to do that one classification task. But it does it so well that even today's LLMs still struggle to be able to do that kind of thing at that deep industry level understanding" [4]. That is his standing argument for specialisation: general models absorb much of the old work, "not all of it," and "there's still a lot of value in really specialization and refinement in that industry level deep dive understanding" [4].

On content, and why expert interviews changed the product

The starting corpus was, in his phrase, "the information that was hiding in plain sight, hiding because there was so much of it that was hard to get to the insights even though they were available to every professional in the market": global filings, earnings call transcripts, conference presentations, press releases, news, and broker research, which let a user compare "what is the company saying, what is an analyst saying about any topic, any company" [4][1][2]. His view is that this was necessary but not differentiating, since it was still obtainable elsewhere. The turn came with expert transcripts. Stream was "a vast library of expert interview transcripts... really a transformative new type of market intelligence content" that "shines the light on a kind of blind spot for the market" [1]. The appeal is that the source is not the issuer: "so they're not filtered so you get really get to kind of closer to the truth" [1]. Before that, he says, "you had to rely on what is the company saying, what are they putting out there in their press releases or saying in public forums and filings, but you really had to go talk to management to question that or get alternative points of view" [4].

Tegus extended the same logic at scale. Kokko credits it with pioneering "this concept and business model of investor-led expert interviews," with investors interviewing thousands of experts a month "that are in the trenches of almost every industry out there" [2], working with roughly a thousand buy-side firms across public equities, private equity and venture capital [4]. The decisive gap it filled was private companies, "an area where for investors it was really hard to get insights on before. In fact, that was, you know, the top request over the years from AlphaSense's customers" [2]. He now claims that as the strongest position in the portfolio: "There wasn't much qualitative research on private companies out there, but these expert interviews started to really pull that in. And today we feel like we've got the richest source of insights on private companies" [4]. Content additions continue, but he is explicit about the priority: the focus goes where "we see so much unique proprietary incremental value that we can add" [4].

On what generative AI actually changes in research

The interface change is the one he keeps returning to. Rather than keywords, a user asks a question, and the system "goes across all the half a billion documents in our system and is able to find the most relevant ones and then dig deep into them and ask those same questions from every single document" hundreds or thousands of times, returning a narrative answer [4]. His worked example is a research question a person would previously have handed to a junior: "let's say you want to understand what is AMD's AI chip strategy and how do they plan to compete with Nvidia? In the past, you would have asked that from a human analyst and they'd have to do a bit of work to get to the right answer" [2]. He also points to summarisation as the humbler but real win, letting users "see what is a long-form document talking about before they have to read the whole thing" and so "cover more ground" [2]. The same capability is being extended to customers' internal content, combining thousands of external sources with information "that's often been very hard to navigate and search" in one experience [2].

He frames the payoff in terms of three things at once. Clients tell him "a single deep research report that our AI produces now gives them same 10 pages that they spent three weeks producing with a team of people... Allows you to do much more diligence and ultimately be more confident in your decision-making and still be a lot faster. So it addresses the quality and the speed and the confidence all at once" [4]. This is why he regards an information platform as the natural home for the technology: "if you think about an information platform like ours, that's probably the best fit for this technology that you can imagine" [2].

On trust, citations and sending users back to the source

Kokko draws a sharp line between AlphaSense and consumer chatbots. "A crucial difference to what people are used to with these chat bots we all use as consumers is that we focus on taking users to those underlying documents. Our users are serious professionals that care about reading and getting deep into the context that is stated in a SEC filing or research report or expert interview" [4]. Answers are "granularly cited," and the interface is built so citations and source documents sit on the same screen, letting the user "dig deep into the underlying document and really understand the context and go deeper and lodge additional queries from there" [4]. The point is not deference to the machine. "You can get a very strong confidence in what you're reading because you know where it's coming from... and then judge for yourself. Not just trust that, but get the whole 360 degree view" and "gather more of the mosaic of information" [4]. The machine will hand you conclusions, but "you can question what the machine is giving you and ask different questions from different angles" [4]. That design choice was validated back to him in conversation with Lloyd Blankfein, who argued that a bare answer is untrustworthy and a bare list of sources is insufficient, and that what a professional wants is the answer with a bibliography [3].

On prompting as a management skill

Asked how to get more out of the system, Kokko reaches for the analogy of delegation. "I'd go back to how do you ask an analyst that's working for you? How do you make sure that you convey all the information that the analyst needs to know so that you can be sure that your request has been understood? It's the same thing with the machine. If you keep it too vague, it might misunderstand the question or it may make assumptions that you don't like" [4]. He notes the symmetry runs both ways: "you ask your colleague a question and you have to think did I give a good enough prompt that I can trust the answer and the same applies to these machine models" [4]. Effort should scale with cycle time. A fast generative search returns in about six seconds, so "it's cheap to ask lots of questions"; a thinking mode runs one or two minutes and dozens of searches; deep research takes ten to fifteen minutes and produces a ten-page report, and "when you're doing that longer cycle work, you're going to be more careful with your prompts. You don't want to wait and then realize that you weren't precise enough" [4]. He also treats prompt quality as the vendor's problem rather than the user's burden: "We feel it's our job to make sure we understand the user and understand what they're looking for," building toward a system with "the quote unquote intuition to understand what you didn't say," calibrated to a user's role, industry and company [4].

On where edge comes from when everyone has the same machine

Kokko accepts that his product commoditises work that used to be a moat. "We are raising the bar for sure as any technology that is introduced into the investment process. Now everybody's able to do things much more quickly, efficiently and move more in an agile way because the research can be applied to so much more that in the past you just had to ignore" [4]. Differentiation relocates rather than disappears: "It becomes now a question of who's asking the right questions and how are you asking them and what angles and how do you look at cross industry impacts and read-throughs from this company to that company," plus the plainer advantage of adoption speed, "about technology adopters early and late and how are you able to adapt to these new solutions and how well can you deploy them" [4]. Customers are already asking him to level the field internally, "how do we raise everybody to the same level so they're very good at asking the machine" [4]. On the fear of hollowed-out jobs he is unsentimental and optimistic: "this machine automation really just does the work that nobody wanted to do. The work becomes more interesting and it's easier ultimately to do the value adding work when the machine does the heavy lifting" [4]. He has also taken this argument into public debate about AI on Wall Street and about competing with Bloomberg [6][8].

On leverage, autonomy and risk

The question of whether AI lifts everyone or only the best is one Kokko says recurs inside his own company: "there's a debate that we end up having a lot. You can think about AI as a technology elevating everybody, but then your point is more about maybe it's the superstars that it elevates the most" [3]. Blankfein's answer, that everything now sits on a longer lever so decision-makers are worth more and the rank and file worth less, drew Kokko's agreement on the underlying mechanism, and he seized on the trading example of colocated servers as showing "how edge really matters even if it seems very minute can mean everything depending on what you're trying to do" [3]. From there he pressed the harder question: markets have already ceded autonomy to machines at millisecond scale, so at what point does the industry become uncomfortable handing agents more [3]. His own instinct is that banking culture holds the answer, "banks are kind of all about risk control. It's sort of letting ambitious people go and take risk but also building the right controls to prevent that," and with "a lot of ambition from technology players to go and build agentic automatic things. I think that risk control lens is really important too" [3].

On decisions as the unit of value, and the company's arc

Kokko frames the entire business around decision quality rather than data volume. "For every company, their market value is a sum of the decisions that they make, and we help them optimize that" [2]. That framing explains the customer expansion beyond investors: the same unfiltered expert content that helps a fund choose a stock also helps an operating company work out "how do I navigate my competitive environment" [1], and the acquisition of Tegus supported serving all types of companies rather than investment firms alone [10]. On capital, he treats each round as external confirmation of demand rather than an end in itself. After the $225 million raise he pointed to having "doubled our user base since the last year," which "showed to them that we have this kind of vast demand in the marketplace" [1]; the later $650 million round valued the company at $4 billion [2]. He has been consistent about listing: there is no near-term pressure because "we have enough money in the bank," but "we do see it as something that we'll eventually do. We have a big vision for the company and we'll need significant resources to execute on that, so IPO is certainly part of that road" [1].

Takeaways

• The founding insight was a consumer-professional gap: Google existed for daily life while analysts were "control F searching for individual terms one at a time in PDF reports" [4][1].
• General-purpose LLMs did not eliminate the need for domain specialisation; narrow models trained on hand-tagged financial statements still outperform them on deep industry classification tasks [4].
• Expert interview transcripts are the differentiating content because they come from customers, suppliers, former employees and competitors rather than the company, getting "closer to the truth" [1][4].
• Private company coverage was the top customer request for years, and Tegus was acquired largely to close that gap [2][4].
• The product is deliberately built to send professionals into the underlying documents with granular citations, so users "judge for yourself" instead of trusting a generated answer [4].
• Prompting should be treated like briefing an analyst, with precision scaled to the cycle time, from six-second searches to ten to fifteen minute deep research reports [4].
• AI raises the bar rather than conferring edge by itself; advantage shifts to who asks the right questions, spots cross-industry read-throughs, and adopts early [4].
• Kokko argues the banking discipline of risk control should govern the push toward agentic automation, given markets already cede millisecond decisions to machines [3].