AlphaSense for Finance Analysts: What It Does and What It Changes

AlphaSense is an AI search engine for business research. Launched in 2011 and based in New York, it searches what it counted in June 2026 as more than 500 million documents, from company filings and earnings-call transcripts to Wall Street research and interviews with industry experts, and answers questions in plain English with citations back to the passages it drew on.

For a junior analyst, that changes the first hours of any project. The old job was finding the right documents and reading them. The new one starts from a cited draft that arrives in minutes, and the work is checking it: opening the sources, judging whose words they are, and knowing what the library cannot see.

AlphaSense at a glance

2011

Launched

New York, by a former investment banker

500m+

Documents in the library

as AlphaSense stated in June 2026

200,000+

Expert-call transcripts

as AlphaSense stated in August 2025

$7.5bn

Valuation

June 2026 funding round

What is AlphaSense?

AlphaSense sells a subscription research platform to banks, funds and large companies. Its founder and chief executive, Jack Kokko, put the ambition in one line in 2018: "We're doing to business information what Google did for the internet."

The comparison holds in one important way: most of what AlphaSense searches, it did not write. It licenses and indexes what companies, banks and news publishers put out, arranges expert calls whose transcripts join the library, and since 2023 has built generative AI on top of all of it.

AlphaSense is not AlphaSights

The names cause a real mix-up, so it is worth settling early. AlphaSights is a separate company, an expert network founded in 2008 that connects clients with people who know an industry from the inside. AlphaSense is the search platform.

The two have drawn closer since AlphaSense began arranging expert calls of its own. In January 2025 AlphaSights sued AlphaSense for trademark infringement in the U.S. District Court for the Southern District of New York, arguing that the move had caused confusion among customers.

What AlphaSense changes for a junior analyst

From finding to checking

What AlphaSense changes shows up in a junior's first week. Picture the request: a managing director wants to know what a client's five closest competitors said about pricing on their last two earnings calls, before a pitch on Thursday. Done the old way, that is a search problem, and most of the day goes on reading.

With AlphaSense, it becomes a checking problem. The question goes in once, in plain English, and a synthesized answer comes back in minutes, each point tied to the transcript or broker note it came from. What is left for the analyst is the part no machine signs off on.

The finding job

• Pull ten transcripts and a stack of broker notes
• Read each one, guessing at the right keyword
• Draft the first version from a blank page

The checking job

• Open each citation and read the paragraph around it
• Decide how much weight each speaker deserves
• Start from the draft and fix what it got wrong

Why the skill moves to the sources

A cited answer is only as good as what sits behind the citation, and AlphaSense mixes very different speakers in a single result: a chief executive on an earnings call, a sell-side analyst with a rating to defend, a former employee who left the company months ago. Reading an answer well means knowing which of them is talking.

So the useful knowledge is less about buttons than about sources: what is in the library, who wrote it, and what is missing from it.

What do analysts search in AlphaSense?

The library, then, is where understanding starts. It runs from public filings to sell-side research and expert interviews, and each kind of document speaks with a different voice.

Broker research and the entitlement rule

Sell-side research is where the library differs most from one firm to the next. AlphaSense splits it in two. Real-time research, delivered as soon as a bank publishes it, depends on an entitlement, and a broker typically approves one when the reader's firm has a relationship with it: a trading relationship for a buy-side fund, an advisory one for private equity, a coverage one for a company's investor-relations team.

Without an entitlement, a reader can still get what AlphaSense calls After Market Research: many of the same reports, typically zero to seven days after real-time readers had them. It is metered, with each user or team holding an allowance of pages, and a page opened once can be reopened within 24 hours without using another. Corporate clients get a separate collection of bank research, sold as Wall Street Insights.

Expert-call transcripts

The expert library is the most distinctive part of the platform, and its core came with Tegus, the expert-research firm AlphaSense bought in 2024. AlphaSense said in August 2025 that the library held more than 200,000 transcripts across 25,000 public and private companies, and was adding thousands more every month.

Buy-side analysts lead the library's calls, which join it after an embargo period; clients can also run calls of their own, with a human interviewer or AlphaSense's AI one. Each transcript is labelled by the speaker's perspective, such as former employee, customer, competitor or consultant. AlphaSense requires experts to be at least six months out of the role they are describing, and a team of compliance reviewers works to keep material non-public information out of the calls.

What do AlphaSense's generative AI features do?

On top of that library, AlphaSense has added generative AI in stages, and the product names have changed faster than the jobs they do. Anchor on the capability and treat each name as dated.

Search, summaries and answers

Underneath everything is the search layer AlphaSense calls Smart Synonyms: a search for "layoffs" also finds "reduce headcount" and "job cuts", which a plain keyword search would miss.

Summaries came next. In June 2023 AlphaSense said every earnings call on the platform would get an AI summary within a few hours, each bullet linked to the exact place in the transcript it came from, and that early-access customers had cited savings of 2 to 14 hours a month.

Generative Search is the question box: type a question in plain English and the answer comes back with citations. AlphaSense rebuilt it in January 2026 as what it calls a multi-agent research system. It runs at three speeds: an automatic mode, a Think Longer mode that typically answers within one to three minutes, and a Deep Research mode that takes ten minutes or more.

Users can pick the underlying language model from a drop-down, though AlphaSense recommends leaving it on Smart Default, which chooses one per question.

Grids and research agents

Generative Grid puts questions in columns and documents in rows, up to 12 questions across up to 400 documents, with an answer in every cell and its source a click away. Deep Research writes the longer pieces: a primer on a company or industry covering its business model, markets, competitors and the current bull and bear case, or a screen of possible acquisition targets that can go straight into a slide deck.

SuperAnalyst goes further. AlphaSense describes it as working like a user on your behalf: monitoring filings, earnings and research 24 hours a day, identifying experts and running calls, and producing briefs, models and presentations as events occur.

Inside Excel and PowerPoint

The add-ins bring the same engine into the files juniors build. In Excel, an analyst can ask it to add a quarterly revenue build or layer in an LBO debt schedule, and AlphaSense says the model extends its logic instead of replacing it; the files never leave the analyst's machine. In PowerPoint, the add-in can re-spin a pitch book for a new target, build comps tables and pull precedent transactions.

How do analysts use AlphaSense day to day?

AlphaSense's banking page lists four jobs the platform is built for: pitch books and market research, finding buyers and targets, transaction comps and valuation, and deal preparation and diligence. One associate at a bulge-bracket bank, quoted on that page, said a project that "would typically take five hours over the span of a few days" took ten minutes, "while we were on the phone with the client."

Three jobs it now drafts

In a junior's week, those four turn into three recurring tasks.

1. Getting up to speed on a sector before a pitch. A Deep Research primer gives the first pass; the junior's job is to test it against the latest filings and the partner's view of the market.
2. Earnings season. Each call's summary arrives within hours, and a grid then asks every transcript in the peer group the same questions: guidance, pricing, margins. The junior reads the cells that matter in full.
3. Diligence on a private company. With no public filings to lean on, the expert library carries more weight: former employees on how the company sells, customers on why they buy, competitors on where it wins. AI-led channel checks add a read on demand, pricing and supply.

A worked example: the peer grid

Scale the Thursday request up to a full earnings season and the arithmetic looks like this. The reading and checking times are assumptions for the illustration, not measurements:

• Fifteen peers, each with two recent earnings calls: 15 × 2 = 30 transcripts.
• Three questions for each transcript (guidance, pricing, margins): 30 × 3 = 90 answers, each with a citation.
• Reading every transcript at an assumed 45 minutes each: 30 × 45 = 1,350 minutes, or 22.5 hours.
• Checking every cited answer at an assumed two minutes each: 90 × 2 = 180 minutes, or 3 hours.

On those assumptions, 22.5 hours of reading becomes three hours of checking. The grid could hold far more, since 400 documents and 12 questions is 4,800 cells. The limit is not the machine. It is the checking, which cannot be skipped, because a wrong pricing comment in a pitch book is the analyst's error, not the software's.

Can you trust AlphaSense's answers?

Confident answers, checkable sources

The software can make that error first, and AlphaSense is blunt about how. Hallucination, it says, is "when a genAI model provides inaccurate or misleading information, usually with total confidence and conviction."

The wider category has a documented record on exactly this material. In FinanceBench, a 2023 benchmark built from public-company filings, GPT-4-Turbo paired with a retrieval system answered incorrectly or refused on 81% of the questions. That test did not involve AlphaSense, but it shows what can go wrong when a general model reads financial documents.

AlphaSense's answer is to narrow what its models can say. Answers are grounded in its own content rather than the open web, citations point to the exact snippet an answer came from, and when Generative Grid finds nothing relevant in a document it says so rather than inventing an answer. The company says its Office add-ins are built to flag anything that looks inconsistent or unsupported rather than paper over it.

What sits outside the library

Generative Search runs inside what AlphaSense calls a "walled garden" of premium, curated content. Web results come in only when a user switches web search on; Google's Gemini model then finds the pages, and AlphaSense summarises and cites them.

Add the two gaps from inside the library, research a firm is not entitled to and experts at least six months out of the role, and an answer can be faithful to everything it saw and still miss the latest thing that happened.

Who uses AlphaSense, and how big is it?

Inside those limits, the platform has grown fast. AlphaSense is private, so its scale comes from its own announcements, and read in order they show recurring revenue rising from above $100 million to above $600 million in about four years.

In June 2026 AlphaSense said it served a majority of the Fortune 500 and "nearly all of the world's largest financial institutions," and it named customers ranging from Amazon, Nvidia and Pfizer to the D. E. Shaw Group. CNBC ranked it eighth on its 2025 Disruptor 50 list.

Banks as customers, investors and builders

Banks sit on both sides of the relationship. JPMorgan Chase is a named customer, and J.P. Morgan Asset Management was one of three leads on the June 2026 round. Goldman Sachs's asset-management arm co-led the 2022 round, and Goldman Sachs Alternatives was still among the company's investors in 2026.

Banks also build their own tools, which do a different job. JPMorgan's LLM Suite and Goldman's in-house AI assistant are general assistants for each bank's own staff, and Morgan Stanley's AskResearchGPT, launched in 2024, searches the bank's own research, which it put at more than 70,000 reports a year. What AlphaSense sells is the library none of them owns alone: research from many banks, plus expert calls, in one index.

The difference shows against Hebbia, one of the specialist AI tools Citi's head of banking, Vis Raghavan, said in May 2026 the bank was piloting: it runs a similar table of questions, but over documents a firm supplies itself.

Where AlphaSense came from, and how it is changing

An analyst's frustration

The library none of the banks owns alone began as one banker's frustration. Jack Kokko, a former Morgan Stanley analyst, was an investment banker in the late 1990s, and he spent days hitting Ctrl+F, one keyword at a time, across thousands of PDFs, still worried about what he had missed. He later described "fearing that you're missing something critical that's putting a billion dollar deal at risk."

The idea took shape about a decade later at Wharton, where Kokko and Raj Neervannan were in the MBA class of 2008. A class project, Kokko said, was "very similar to my analyst days," and it showed him that "this problem still was there, that it was still incredibly hard to find information. It was very manual." They launched AlphaSense in 2011. Kokko is its chief executive and Neervannan its chief technology officer.

Buying the content

For its first decade, AlphaSense mostly indexed what others published. Then it started buying content: Stream, a library of expert-interview transcripts, in October 2021, and Sentieo, a research platform built for investors, in May 2022.

The largest deal was Tegus, for $930 million, announced in June 2024 and closed that July. Tegus brought more than 100,000 expert-call transcripts and two research tools it had already bought, BamSEC and Canalyst. The integration had a cost: in November 2024 AlphaSense cut 150 jobs, about 8% of its workforce, to remove the overlaps the deal created.

From search engine to agent

Read in order, the launches in the features table point one way: from finding documents, to reading them, to doing the work around them.

Kokko's view of what that means for analysts is optimistic. Asked by CNBC in June 2025 about fears that AI would replace Wall Street analysts, he said, "It's a popular narrative," and then: "But I would not be so sure." His case was productivity: "That person will be operating with a higher ROI [return on investment] and companies don't cut high ROI people."

Of the expert interviews, he said: "There are a hundred on a single company, and no human can read it all, but Deep Research will read it all and ask questions." Whether the banks see it the same way is a separate question, covered in what banks have said about AI and analyst jobs.

What to know about AlphaSense for an interview

AlphaSense is most useful in an interview as evidence that you understand how research work is changing, not as a product to drop into an answer. Four points carry that:

1. Name the source, then show how you checked it. Say which kind of document answered the question, a filing, a transcript, a broker note or an expert call, and why that changes how far to trust it; then trace any number back to the filing it came from.
2. Show you can cover a blind spot. Pick one thing a cited answer could not see, such as the newest research or a change since an expert left, and say how you would fill the gap: the latest filing, a second expert, or a direct question to the company.
3. Keep the names straight. Confusing AlphaSense with AlphaSights, the expert network, tells an interviewer the research stopped at the logo.
4. Talk in capabilities, not product names. The names change quickly. What holds up is knowing what a summary, a table of answers or a research agent does, and where each one fails.

If your school licenses it

Some business schools buy AlphaSense for their students. Harvard Business School's Baker Library offers it to current students with remote access, no downloads of expert transcripts, and 25 company-model downloads over the life of an account; Stanford's Graduate School of Business gives accounts to current students too.

If yours does, use it on a real question, such as a company you follow, and be ready to describe one answer you checked and what you found behind it. The interview readiness tool walks through the wider questions, and the guide to the banks' own AI covers what individual banks have built for themselves.

The bottom line

AlphaSense makes finding cheap. A sector primer, a peer comparison or an earnings summary now takes minutes, and that speed is what banks, funds and companies pay it for.

What it does not make cheap is judgment: knowing whose words sit behind a citation, whether the newest research was even visible, and whether the number matches the filing. A junior who can do that checking, and explain it, is doing the part of the job the software hands back.