A company that does not build ChatGPT and does not chase general AI raised $130 million, reached a $700 million valuation, and became profitable. Why?
In July 2024, a financial AI company called Hebbia closed a $130 million Series B at a $700 million valuation.
The financing itself was not shocking. Larger AI rounds were everywhere. The surprising number was in the fundraising coverage: $13 million in revenue, and already profitable.
In 2024, an AI startup becoming profitable was more interesting than an AI startup raising $130 million.
Even more interesting: in the same month, Hebbia CEO George Sivulka published a deliberately provocative essay titled “Productive Individuals Don’t Make Productive Firms.”
The core argument was this: AI has made every individual faster. People write emails faster, write code faster, and make presentations faster. But no company has become meaningfully more valuable just because each employee can move a little faster. Where did the productivity go?
This is not a rhetorical question. Hebbia’s product philosophy is built on the recognition that organizational efficiency is not the sum of individual efficiency.
01 The Right Way to Build Financial AI: Help Institutions, Not Individuals
If you open Hebbia’s product, Matrix, you do not see a ChatGPT-style chat box.
You see a workspace that can ingest tens of thousands of documents, a natural-language query layer across all of them, an output engine that can automatically generate presentations, memos, and emails, and a system called Skills that turns the working methods of the best people on a team into something repeatable, so everyone can produce to the same standard.
In other words, Hebbia is not designing a “better brain.” It is designing a better organizational collaboration system.
Most AI products answer the question, “How can I finish my work faster?” Hebbia answers a different question: “How can our team produce higher-quality work?”
That difference determines every product decision:
- It needs a permission system, because analysts at different levels can access different layers of analysis.
- It needs knowledge accumulation, because today’s analysis should become tomorrow’s starting point.
- It needs auditable citations, because financial output cannot be “AI guessed this.” Every conclusion has to trace back to source documents.
- It needs repeatable methodology, because best practices should not walk out the door when one person leaves.
So why have most AI companies missed this?
Because individual productivity is a natural product metric. It is easy to A/B test, easy to show users, and easy to put in a fundraising deck. Organizational productivity requires a much deeper understanding of how a team actually works, which is far harder than building a chatbot.
But Hebbia made the right bet: the customers willing to pay high prices, especially financial institutions, are not buying individual speed. They are buying team output quality.
02 The AI Version of the Palantir Model: Hiring the Right People Matters More Than Building the Right Feature
Hebbia’s hiring page reveals its go-to-market strategy:
- Forward Deployed Banker (AI Strategist): former investment banker, now on-site AI strategist.
- Forward Deployed Investor (AI Strategist): former private-equity investor, now on-site AI strategist.
“Forward deployed” borrows from the strategy Palantir famously developed: send domain experts directly into the customer’s office, build around real projects, and drive product adoption from inside the workflow.
Sivulka’s own interview comment was even more direct: a top global bank chose Hebbia over a leading model lab because the model lab’s people needed the customer to explain what a CIM, or confidential information memorandum, was. Hebbia’s people did not.
That is Hebbia’s core competitive barrier. It is not more model parameters or a larger training set. It is that the team includes people who have actually done the work.
These former Evercore analysts, former private-equity investors, and former law-firm partners have the title “AI Strategist” at Hebbia. They do not write model code. They do three things:
- Understand the customer’s specific workflow, such as how the team runs due diligence.
- Build Skills, encoding best practices into the product through Markdown files.
- Train the customer’s team and make sure adoption actually happens.
This model means Hebbia is not simply “selling software.” It is selling capability. The customer is not buying a tool. It is buying a domain-fluent deployed team.
The lesson for B2B AI founders: if your industry is vertical enough and complex enough, hiring an industry veteran to deploy the product can be more effective than hiring three engineers to build more features.
03 The Technical Moat: Why Hebbia Gave Up on RAG
Hebbia was already working with semantic search and RAG in 2023. Then it announced that it was moving on.
The title of its technical blog post was simple: “Goodbye, RAG.”
The reason was also simple: standard RAG is not enough for financial workflows.
Financial professionals are not asking only, “What did this document mention?” They ask questions like:
- “Which statements in Google’s latest earnings report sound like lies?”
- “What loophole exists in Clause X of this NDA? What risk emerges when it is combined with Clause Y?”
- “If Fund Z’s position table shows Stock A has a higher weight than Stock B, and Stock A has lower liquidity than Stock C, what happens during market volatility?”
These are compound questions involving multi-hop reasoning, conditional judgment, and synthesis. Standard embedding-based similarity search cannot solve them.
Hebbia’s answer was to rebuild the information retrieval architecture. It uses a system called the ISD architecture to separate document retrieval from reasoning. The result: every output can be traced to precise locations in source documents, the system can handle batch tasks across 30 billion tokens, and the outputs are reliable enough for financial institutions to use.
The lesson for AI founders: if your customers need trusted, traceable, cross-document reasoning, do not use RAG as a shortcut. There is no shortcut here.
04 The Most Underrated Innovation: Skills
Hebbia has a system called Skills. I think it is the company’s most underrated competitive moat.
Skills are essentially digital canning for industry methodology.
There are three levels:
-
Hebbia Skills: Standard working methods encoded by former bankers and lawyers on Hebbia’s team. For example, a “comparable transaction analysis” Skill can encode the workflow of people who have done deal valuation countless times.
-
Institution Skills: The judgment logic of the best analyst inside the customer’s own firm gets encoded into the system. The analyst whose work always wins unanimous support in investment committee can now help every new hire produce analysis at a similar standard.
-
Personal Skills: Individual preferences and shortcuts.
Imagine a new investment-banking analyst on their first day. They no longer have to ask a senior banker, “How do we build this model?” The team’s accumulated best practices are already inside Skills.
How strong is the lock-in?
Leaving Hebbia means losing the encoded methodology of your team’s best people.
That is not technical switching cost. It is knowledge switching cost. Technology can be migrated. Methodology is much harder to move.
05 Verifiable ROI: Giving Procurement Nothing to Argue With
Hebbia’s customer case-study pages provide concrete, verifiable numbers:
- Provident Healthcare Partners: time to produce core M&A deliverables was reduced by 40%. Document coverage increased by 30% in the same amount of time. The firm expanded business capacity without adding headcount.
- Troutman Pepper Locke: a migration project involving more than 1,000 contracts “would not have been possible without Hebbia.” The team estimated the work was 3 to 4 times faster than doing it manually.
- Orrick: due diligence time fell from days to minutes.
These are not vague claims about “improved efficiency.” They are specific metrics a procurement team can put directly into an annual contract ROI review.
The lesson for B2B AI founders: do not sell “efficiency.” Sell the specific compression ratio of a specific workflow.
A managing director who goes to the CFO with “40% faster core reports” has a stronger case than one who says “10x productivity.” The former means the same team can do more deals. That is a revenue-growth logic, not merely a cost-saving logic.
06 The Honest Caveat: What Is Hard to Copy
The first half of this article covered many things that can be copied. Now for the parts that cannot.
Four Forms of Luck That Are Hard to Copy
-
The founder’s strategic foresight: Sivulka bet on “organizational AI” in 2020, roughly 3 to 4 years before the market caught up. That judgment cannot be copied as a method.
-
Capital-intensive R&D: Rebuilding the information retrieval architecture took years and a lot of money. Not every company has that patience or balance sheet.
-
Brand trust with top financial institutions: Getting a top investment bank to let your people sit in its office and build workflows requires years of trust. Backing from a16z, Index, and GV helped significantly.
-
A proprietary data flywheel: Tens of thousands of daily annotations across private financial documents create a data barrier that public datasets cannot replicate.
Data Disclosure
- $130M Series B, $700M valuation, and $13M profitable revenue: from TechCrunch’s July 2024 reporting, a credible third-party source.
- 40% faster deliverables and 30% more document coverage: from Hebbia’s customer case-study pages, not independently audited by a third party.
- Product feature descriptions: from Hebbia’s official blog, a primary source.
- Current ARR: Hebbia has not publicly disclosed updated post-financing ARR. The $13 million figure refers to data reported around the mid-2024 financing.
07 Five Actionable Lessons for B2B AI Founders
-
First answer one question: are you helping individuals move faster, or helping organizations become better? If it is the latter, your product needs permissions, knowledge accumulation, and output workflows, not just a clever chat box.
-
Hiring industry veterans for product deployment can matter more than hiring engineers for new features. One person who understands the customer’s workflow and sits with them for a week may be more effective than three engineers building remotely for three months.
-
Sell with specific, verifiable ROI numbers. “Efficiency improvement” is abstract. “40% faster core reports” is something procurement can sign.
-
Make best practices repeatable. Can your product help the least experienced person on a team produce work close to the level of the best person? If not, you have not solved organizational productivity.
-
Do not use RAG as a shortcut. If your customers need trusted, cross-document reasoning, you need your own information retrieval architecture. There is no shortcut, but once you get it right, it becomes an irreplaceable barrier.
Hebbia’s story tells us that in the AI era, the most valuable companies are not the ones that look the smartest. They are the ones that turn intelligence into something repeatable.
Data note: The core data and facts in this article come from TechCrunch’s July 2024 reporting, Hebbia’s website, and Hebbia’s official blog. All customer ROI data comes from Hebbia’s customer case-study pages and has not been independently audited by a third party. Product and technical information comes from Hebbia’s public technical blog.