What this covers

Patrick interviews Jack Kokko, founder and CEO of AlphaSense, about how the market-intelligence search engine was built and where it is heading. The story starts with fragmentation. As an investment banking analyst, Kokko logged into dozens of systems, and the product's promise became one question in one place, with confidence that absence there means absence everywhere. Financial content sat in paywalled silos without the links and click data that let Google rank the web, which Kokko gives as the reason Google retreated from enterprise search.

The middle of the conversation turns to LLMs. They let a system grasp a query's intent and read documents more richly, but Kokko says robust RAG systems are far harder to build than the impressive demos suggest. He is surprised that some professional users over-trust public chatbots. His answer is to keep sentence-level traceability to source documents as the guard against hallucination. He also describes a model-agnostic design that lets the team swap models by task, and he names reasoning as the most valuable capability right now.

The business half covers small hedge funds as early customers, and investor relations teams as the beachhead into corporations, which he says raised the addressable market roughly tenfold. It also covers the Tegus acquisition, expert interviews as the only systematic window into private companies, and the near-billion-dollar raise that funded the deal. Kokko adds views on M&A culture and integration, and a contested claim that Europe lags in producing technology companies. He names his biggest mistake as not selling the larger vision early enough.

Sharpest takeaway

Kokko argues that AlphaSense's durable advantage is proprietary, hard-to-aggregate expert and filing data plus a fast-moving, model-agnostic orchestration layer that keeps users able to verify sources, so LLMs raise the value of the data rather than commoditizing it.

  • Fragmented paywalled data was the original hard problem, and the web-style authority signals Google relied on did not exist for it
  • LLMs make natural-language intent understanding and cross-document synthesis possible, but users still need citations to trust outputs
  • Expert interview data fills transparency gaps in public and private markets where no regulatory disclosure exists

The claims · ranked54 claims · weighted by value

This asset isn't compiled yet

You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.

0.80

Acquisitions that make strategic sense, where two products belong together so that one plus one equals much more, get investor support very quickly, but acquisitions done just to buy scale should be avoided.

normativehigh valueestablishednovelty 2/4durability 4/4· Jack

“If you have strategic logic and two products that really belong together and are one plus one is going to be four or five or ten”

0.75

Despite a background in investment banking M&A, Kokko found no rulebook or instruction manual for acquiring and integrating a company, and business school does not teach how to acquire, integrate, and leverage a business, so acquirers must learn by doing.

factualhigh valueestablishednovelty 2/4durability 3/4· Jack

“I couldn't find any rule book or instruction manual to it. It's shocking how little there is really out there”

0.75

Building a startup ecosystem is a chicken-and-egg problem because entrepreneurs, investors, and talent must reinforce each other, so it takes a long time to build, and early attempts can fail because they arrive before the flywheel exists.

causalhigh valueestablishednovelty 2/4durability 3/4· Jack

“But it's hard when it's a chicken and egg problem, having the entrepreneurs, the ecosystem, the investors and everything feeding itself”

0.71

Large language models let a system understand a natural-language query's true intent in a way keyword matching could not express, and also give a much richer understanding of the underlying documents, so search moves from keyword lookup to answering business questions.

causalhigh valueestablishednovelty 2/4durability 3/4· Jack

“which now tells the machine a lot more about what the true intent is that you're looking for in a way that couldn't be expressed in those keywords”

0.68

Europe lags the US in producing technology companies because European entrepreneurs find it harder to raise scale capital, which often requires going to the US or London, harder to find experienced founders and employees, and face a more risk-averse culture where failure is a terminal state rather than a step.

causalhigh valuecontestednovelty 2/4durability 3/4· Jack

“In Europe, you've got more risk aversion. It's sometimes just hard to find people that are willing to join an early-stage startup”

0.68

Private companies have almost no systematically accessible public information because there is no regulatory requirement to disclose, so expert interviews are the only systematic source, which makes this the largest new content vacuum to fill as private markets grow.

factualhigh valueestablishednovelty 2/4durability 3/4· Jack

“there is no regulatory need to put out information. So this now fills a much larger vacuum”

0.60

The fundamental problem AlphaSense solves is fragmentation: as an investment banking analyst, Kokko had to log into dozens of systems, and even aggregated tools are unsatisfying unless users can ask one question in one system and trust that absence there means absence everywhere.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Jack

“I think for us the fundamental problem has always been about fragmentation”

0.60

Successful acquisitions depend on cultural integration: acquirers should imagine being acquired, anticipate what employees need to hear on day one, and address fears quickly, because no system exists for evaluating talent and integration is still manual work.

normativehigh valuespeaker onlynovelty 3/4durability 4/4· Jack

“always when you try to imagine yourself on the other side being acquired, what do they need to hear day one”

0.58

A data set is valuable to the extent it adds net new information to what already exists, meaning it offers new perspectives rather than overlapping with what is already available, and such truly new perspectives are rare.

normativehigh valuespeaker onlynovelty 3/4durability 4/4· Jack

“To me it's the net addition to the information volume that's out there already”

0.57

Financial content sat in thousands of paywalled silos without the links, authority signals, and click data that let Google rank the public web, so building enterprise search over it was far harder than web search, and Google itself retreated from enterprise search.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Jack

“It wasn't like the Internet, where Google and others had billions of web pages with links between them”

0.57

Large conservative banks were hard to sell to, but small hedge funds became good early customers because they are under intense pressure to find edge and will pay for proven advantage, though they often kept the tool as a secret weapon rather than promoting it.

factualhigh valuespeaker onlynovelty 3/4durability 3/4· Jack

“But small hedge funds we figured out were actually really good customers”

0.57

Professional users, though most do not, sometimes default to trusting public LLMs too much, which surprises Kokko because AlphaSense's serious researchers want to read the underlying content, check the analyst, and understand context before making seven-figure bets.

factualhigh valuespeaker onlynovelty 3/4durability 3/4· Jack

“we occasionally see and hear about professionals using those same tools and defaulting to trusting them”

0.57

Investor relations departments became the beachhead for corporate expansion because they had to understand what investors research, and peer-to-peer word of mouth across firms spread the product quickly without changing it.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Jack

“We found that we could do expand quickly, we could get them to talk to their peers across firms and the word of mouth really spread quickly”

0.57

Consumers using ChatGPT, Perplexity, or Gemini in daily life trust LLM answers too much, because without citations a credible-looking answer feels true, validation is too time-consuming, and the model does not know where its memory-based claims came from.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Jack

“in my mind trust these LLMs too much. And I find myself doing that too”

0.56

To plan for unpredictable model progress, AlphaSense built a model-agnostic system that can use any LLM and quickly swap models for different tasks, such as a model better at multimodal inputs or larger context windows, reducing latency and adding intelligence.

normativehigh valuespeaker onlynovelty 2/4durability 4/4· Jack

“try to create a system that is pretty agnostic, that can use any LLM that's out there”

0.52

Although generative AI enables new capabilities, building a robust retrieval-augmented generation system involves many components, so it still creates new engineering difficulties that a company with a decade of search experience is better positioned to handle.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“as people started calling this RAG (retrieval augmented generation) and started to look at all the components of creating a robust software system, the reality has still been that”

0.52

Expert interviews with former executives, customers, partners, and competitors provide impartial perspectives without skin in the game, which is valuable even for public companies because company filings and sell-side research are filtered to show the best story.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“even that is filtered. The companies try to do their best to show their best sides of their business”

0.52

A key design task was keeping AlphaSense's search UX, where users see individual sentences and snippets and trace them to source documents, while adding generative capabilities, because guardrails against hallucination depend on that traceability.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“some of the beauty was really in the user experience of being able to really see the individual sentences and snippets of text, where the insights came from”

0.52

Investors and analysts should not hang their careers on the ability to find and synthesize information, because better tools make markets more efficient and allocate capital better, which Patrick concludes and Kokko affirms.

forecasthigh valuespeaker onlynovelty 2/4durability 3/4· Patrick

“if you're an investor or an analyst, you better not hang your hat on the ability to just go find information and synthesize it”

0.52

Building a patient investor base with deep pockets, where capital is not needed until a deal arises, allowed AlphaSense to raise close to a billion dollars for the Tegus deal very quickly when the opportunity appeared.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“it was worth patiently building an investor base that has deep pockets where you didn't need the capital”

0.52

Proving that the platform could serve corporate professionals, not just financial professionals, increased the addressable market roughly tenfold and turned AlphaSense from a niche financial tool into a horizontal solution with large scale potential.

factualhigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“proving that addressable market was 10 times bigger, covering really every type of business”

0.52

Kokko's largest mistake was not being confident enough to convince stakeholders of the bigger story earlier, and as a result he under-invested and under-scaled, and he learned to pursue large visions without waiting for milestones.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“maybe that's the one mistake of failure where I wasn't confident enough to tell them that this will be amazingly successful”

0.52

Reasoning is the most valuable model capability for AlphaSense right now, since a system must break a research question into sub-tasks such as market size, players, and growth rates, and native reasoning would simplify the pipeline.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“Certainly reasoning capabilities are valuable right now. We are building that in our own system”

0.52

Retrieval-augmented generation made it easy to build impressive demos, but building a robust software system requires solving many components, so the new technology brought new difficulties alongside its new leaps.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“as people started calling this RAG (retrieval augmented generation) and started to look at all the components of creating a robust software system”

0.52

In an acquisition, the acquirer should anticipate what people at the target fear, show up on day one with answers to their questions, and be honest about what is not yet known, while framing the combined company as a growth story.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“trying to anticipate what's in the minds of people on the other side then trying to address their concerns very quickly”

0.50

Selling to corporations is hard because the titles of people doing research or influencing million- and billion-dollar decisions vary widely, so vendors need insiders as guides and must find buyers across many functions and global departments.

factualhigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“The variety of titles is so broad that it was hard to find those people”

0.50

The model capabilities Kokko most wants are better reasoning so systems can plan research the way an analyst would, plus larger context windows, faster speed, lower cost, and multimodal analysis, all reliable enough to rely on natively without layered verification.

forecasthigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“But maybe the LLMs can get really smart at reasoning and do this type of work natively.”

0.49

Rapid frontier model improvements raise consumer expectations, because users who experience ChatGPT-level capability expect the same in professional solutions, which makes it harder for incumbents but gives fast-moving startups an advantage.

causalhigh valuespeaker onlynovelty 2/4durability 2/4· Jack

“those consumers expect that you have that also in your professional solution”

0.49

Five years out, AlphaSense aims to remove friction so the cost of acquiring information and insights becomes very low, letting machines do work like asking the same five questions across 500 documents in minutes instead of weeks of intern research.

forecasthigh valuespeaker onlynovelty 2/4durability 2/4· Jack

“We're trying to take all the friction and barriers out of the way and try to make the cost of acquiring information and insights really, really low”

0.47

AI can generate and simulate alternative paths an industry might take, but it cannot tell which path will actually happen, so its value lies in scenario generation for investors to explore.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Jack

“And it's not going to be able to tell you where it will go, but it can tell you different paths of where it might go”

0.46

Current LLM products require sequences of models for planning, doing work, checking the work, and cleaning output for the user, because individual models are not yet accurate enough to be relied on natively, and a smarter model could collapse these steps into one.

factualhigh valuespeaker onlynovelty 2/4durability 2/4· Jack

“one is interpreting what work needs to be done and planning the work and then doing the work. And then another one is checking the work”

0.39

Machine learning progressed from categorizing words and local context, to models like Google's BERT that understood longer passages and KPIs and sentiment such as whether a company is bullish or bearish, to large generative models that read whole books and form contextual conclusions.

factualestablishednovelty 1/4durability 3/4· Jack

“Later on, you got the simpler earlier language models, like Google's BERT, where level of understanding expanded”

0.35

Americans treat failure as a step toward success and admire entrepreneurs who build new things, while Europeans fear failure more, which discourages risk-taking and slows ecosystem formation.

causalcontestednovelty 1/4durability 3/4· Jack

“Failure is considered just a step in the way to success, whereas in Europe it might be a...”

0.34

Companies building on AI should structure teams and org charts for agility because breakthroughs arrive frequently and unpredictably, even frontier model developers cannot forecast them, so leadership must track research and pivot quickly.

normativespeaker onlynovelty 2/4durability 3/4· Jack

“you have to try to organize your team and the org structure to be agile because you just have to expect the unexpected”

0.32

In post-acquisition communication, leaders should be honest that they do not know everything, while expressing excitement and offering a growth narrative that makes the change a career opportunity rather than a loss.

normativespeaker onlynovelty 2/4durability 3/4· Jack

“Trying to be honest that we don't know everything”

0.31

Assessing cultural compatibility and identifying talent during an acquisition has no systematic tooling and remains manual work, which makes it a gap in the M&A playbook.

factualspeaker onlynovelty 2/4durability 2/4· Jack

“There isn't any system that I'm aware of for that. It's still kind of very manual work so far.”

0.31

Large corporations' IT teams, given budgets by CEOs to do something with GenAI, quickly tried to build internal search over their own fragmented memos and insights, and this produced demos that often won funding, which pushed the market toward building the same kind of product that Google had abandoned.

causalspeaker onlynovelty 2/4durability 2/4· Jack

“Now you see some large corporations and large financial firms tinkering around their IT teams have been given a budget by a CEO”

0.31

Investment professionals often keep useful tools private because their edge depends on secrecy, so the investor segment does not market products by word of mouth the way corporate buyers do.

causalspeaker onlynovelty 2/4durability 2/4· Jack

“If you're going to be an investment analyst, you might not be tweeting things as much”

0.31

AlphaSense expanded into corporate buyers by moving from investor relations to corporate strategy, financial planning, competitive intelligence, corporate development, strategic marketing, product, sales, and sales engineering, adding personas over many years.

factualspeaker onlynovelty 2/4durability 2/4· Jack

“we started from investor relations and investor relations took us to corporate strategy and financial planning and competitive intelligence”

0.31

Foundation model developers are scaling faster than any company has scaled, having made LLM access that cost 100 times more a few months earlier available for pennies, even as LLMs are being open-sourced and commoditized.

factualspeaker onlynovelty 2/4durability 2/4· Jack

“a couple of months ago was 100 times more expensive. Somebody really got it to be in the market for pennies”

0.31

Large corporations and financial firms, given GenAI budgets by CEOs, are discovering that their internal memos and insights are hidden in disparate places, which creates demand for the same retrieval problem AlphaSense has solved.

factualspeaker onlynovelty 2/4durability 2/4· Jack

“we've done all this work to produce all these insights and memos and so forth, but when we need it, we don't find it.”

0.31

Kokko envisions AI systems acting as a large team of superhuman analysts plus a group of interns that churn through work that previously went undone, enabling higher diligence and letting the system draw tentative conclusions investors can then test.

forecastspeaker onlynovelty 2/4durability 2/4· Jack

“a large team of superhuman analysts that get smarter and smarter”

0.30

Kokko's motivation comes from earlier frustration as an analyst with the manual, inefficient process of gathering information in short time windows, and he believes his nerdy interest in financial disclosures gives the company an edge in making capital allocation more efficient.

normativespeaker onlynovelty 1/4durability 3/4· Jack

“this is the one thing where my nerdy interest in financial disclosures gives me an edge”

0.27

Serious professional researchers want to read the underlying content, see the analyst who wrote it, and get full context before making seven-figure or larger bets, so AlphaSense's role is to route them to the right source, not to replace their judgment.

normativespeaker onlynovelty 1/4durability 3/4· Jack

“before making seven-figure or bigger bets based on this information”

0.26

Kokko now spends a much larger share of his time on product development and reading research himself, receiving a filtered feed from a team that reads everything, with roughly one in ten developments reaching him as impactful.

factualspeaker onlynovelty 1/4durability 2/4· Jack

“I find myself spending a much larger percentage of my time now in just product development, reading everything myself”

0.26

The kindest thing anyone ever did for Kokko was that his parents let him pursue electronics and order components from around the world as a nerdy teenager, which signaled that his passion was worth backing.

normativespeaker onlynovelty 1/4durability 2/4· Jack

“my parents allowed me to just go wild in diving into electronics and ordering components from around the world”

0.26

AlphaSense did not need to change its product to sell into corporations; it only needed to sell first to investor relations teams who wanted to understand how investors analyze their stock.

factualspeaker onlynovelty 2/4durability 2/4· Jack

“And then saying, okay, this works, we didn't have to change the product. Let's just go.”

0.24

Generative AI accelerated AlphaSense's roadmap and created more demand because it moved the product closer to the founders' original vision of a machine users could talk to for answers to any business question, a vision Kokko thought was decades away.

factualspeaker onlynovelty 1/4durability 2/4· Jack

“I could pick up my phone and pick up the AlphaSense app and ask a question, any business question, get the answer. And I thought that was decades away realistically”

0.24

Kokko found public-company disclosures and prospectuses fascinating, and this interest in financial disclosure as an information source gave him an edge that became the basis for AlphaSense.

normativespeaker onlynovelty 1/4durability 2/4· Jack

“my nerdy interest in financial disclosures gives me an edge”

0.24

An analyst-style question such as the market landscape for semaglutides requires the system to decompose the task into market size, players, and growth rates, which illustrates the planning reasoning AlphaSense builds.

factualspeaker onlynovelty 1/4durability 2/4· Jack

“what is the market landscape for semaglutides? And the system breaks it down”

0.21

AlphaSense began by trying to build a semantic search engine that reads financial and business content line by line like a human analyst, indexing billions of data points across millions of documents and mapping them to a user's query to surface the right insights.

definitionspeaker onlynovelty 1/4durability 2/4· Jack

“We wanted to build a semantic search engine that would understand financial and business content, read it line by line, almost like a human analyst would”

0.21

After ChatGPT-era models arrived, AlphaSense's first obvious builds were document summarization and cross-company and cross-industry insight synthesis, tasks that had been too tedious to do before.

factualspeaker onlynovelty 1/4durability 2/4· Jack

“we used to spend the first few months thinking about the first very obvious things to build, like summarization of documents and summarization of insights across an industry”

0.21

AlphaSense is doubling down on expert interview content, which Kokko believes is the most exciting new area of company and industry insight, because it fills gaps in both public and private markets.

normativespeaker onlynovelty 1/4durability 2/4· Jack

“I'm still thinking this is the most exciting area of new content and new insight about companies and industries out there”

0.21

Kokko's early work on EASDAQ, a European NASDAQ-style exchange for startups, was too early because Europe lacked enough exciting tech companies to sustain it, and it might succeed today.

factualspeaker onlynovelty 1/4durability 2/4· Jack

“Europe didn't have enough exciting tech companies to sustain that European NASDAQ”