I demoed a successful AI startup and got a glimpse of what the jobs of investment bankers could begin to look like.

• Hebbia, a five-year-old startup, offers a glimpse of where the AI-powered future of banking is headed.

• I got an exclusive demo of the product and spoke with company executives about how it really works.

• Will it supercharge Wall Street or fundamentally change it? Here is my opinion.

Artificial intelligence is coming to investment banking, and when I heard that there was a startup that was generating buzz throughout Silicon Valley for streamlining tedious daily work in fields like finance, consulting, and law, I had to see it with my own eyes.

The company, Hebbia, offers financial services companies a variety of AI-enabled tools designed to deepen and accelerate their work. Founded in 2020, it has acquired clients including KKR, T. Rowe Price and Permira, the New York-based operation says on its website.

As a journalist who has covered how AI is reshaping finance, I’ve mostly focused on what banks are building internally, like Goldman Sachs’ internal AI assistant, for example. I hadn’t seen a third-party tool up close.

When I was given access to a one-hour live demo from Hebbia, I hoped it would give me an idea of ​​how investment banking might change in the coming years.

Hebbia’s big bet is that many institutions decide to buy access to these tools instead of incubating them exclusively internally. Tom Reeson Price, vice president of sales, told me that “hundreds” of positions at Hebbia are in the hands of sell-side bankers, although the buy side remains its largest market. Hebbia declined to share which banks are using its tools.

“It doesn’t make sense for every company to spend $10 million to $20 million, or even $5 million, on an internal build when you have venture-backed startups serving 150 customers like us,” founder Geroge Sivulka told me in an interview.

Sivulka left a Ph.D. program at Stanford to launch the startup five years ago. In 2024, it raised $130 million in a Series B round from Andreesen Horowitz, Google Ventures, and billionaire investor Peter Thiel.

Wall Street is already feeling the effects. Once upon a time, showing up with Excel knowledge meant you were up to date with the latest and greatest technology. Now, banking neophytes are beginning to dominate Hebbia. “They are like Hebbia’s analysts,” laments Sivulka.

A “Hebbia analyst”? As opposed to an exhausted analyst with bags under his eyes for the third night in a row? I had to know: Could this tool really usher in an era where the hard work of investment banking becomes a thing of the past, or will it simply help dealmakers get more out of their teams?

Here’s what I learned through my demo as I explored whether this product could really revolutionize how Wall Street works.

Free the bankers “to do the last mile”

Divya Mehta, a Hebbia executive who heads product development, began the demo by showing me a spreadsheet-style screen filled with rows of past, open or even potential deals and columns for questions. You can query the same data set in several different ways, adjusting your wording or criteria to run multiple simulations, generating a spectrum of insights from the same raw information.

I imagine how that could help negotiators think more creatively about a company’s acquisition targets or potential buyers, for example.

The system was designed to enhance the work bankers and investors already do, paving the way for faster trading and faster task completion. “They want to take analysts from zero to 90% and free them up to do the last mile,” Tom Reeson Price, head of sales at Hebbia, told me, describing what Wall Street clients are looking for.

Entering the ‘Matrix’

Next, Mehta showed me the heart of the Hebbia platform, what he calls its “Matrix.” It is the primary workspace where users can submit complex questions and watch the system solve them using massive collections of information ranging from spreadsheets and corporate files to PDFs and presentations.

Instead of typing a question and getting an answer at a time, Matrix works as a live research assistant, sifting through documents, extracting essential points, and putting them together in a way that’s meaningful to users.

Each column represents a query, such as “flash earnings” or “debt commentary,” categories through which users can analyze the activities of multiple companies. Each row represents a deal or document. Matrix extracts the relevant information from all of these files and compiles them into a simple table filled with custom results. Now, analysts can scan hundreds of submissions at a time and understand why the system reached its conclusions.

Speed ​​up question analysis and slide presentations

Below is one of Hebbia’s suggestions – you can see how it is structured and how it could apply to different companies.

One of the features sure to delight younger professionals is the ability to supercharge the creation of notes and presentations, potentially alleviating all-night drudgery on documents for clients that often end up in the trash before being read. The work can be heartbreaking.

“In fact, you can build the entire slide deck in Hebbia based on the output that the agent or the grid has created for you,” Mehta said. Work hours are compressed into minutes.

In another example, he described how the system can identify potential buyers for a company by analyzing historical transaction data and previous buyer behavior, a task that would be nearly impossible to perform manually. Law firms and consultants, he added, use similar processes to review large volumes of documents.

Write better questions on the fly

Mehta said Hebbia’s prompt development team refines previously generated prompts on thousands of examples. It’s an iterative process and after discovering the weak points of a message, the team refines the query until it is ready to be sent to customers.

Hebbia’s quick build wizard caught my attention, perhaps because it seemed like a meta of sorts. It is an AI tool that helps users create their own instructions… to feed back to the AI ​​system. Even if someone doesn’t know where to start or how to phrase their question, the system has a feature that can refine what users are trying to ask.

In other words, technology that is capable of teaching people how to communicate with technology. It made me wonder if this will soon influence how professionals in other industries learn.

Negotiation language, side by side

Mehta uploaded another example drawn from corporate earnings reports. These reports are often dense and unpleasant to read.

This screen synthesizes a variety of earnings results, debt load, and other relevant company information. Viewing these metrics allows you to detect subtle patterns that have gone unnoticed, such as recurring themes in management comments or changes in trading trends.

Hebbia maintains a library of reusable prompts for common financial tasks while offering customers the option to create their own. Some customers view their personalized directions as proprietary intellectual property from which they gain an advantage. Sivulka told me that it shows that Hebbia is working as intended.

“It’s really disturbing,” he said. “They use the software, but they don’t want us to know their use cases because they are alpha revenue generators.”

Pattern detection at scale

This screen compares regulatory filings and board recommendations on corporate events and shareholder positions. “We build rapid libraries by use case,” Mehta explained. “This allows us to quickly analyze complex documents such as IC notes and credit agreements.”

Reeson Price told me that about 60% of Hebbia’s users are on the buy side, with the rest made up of banks, law firms and insurers.

Mehta added that Hebbia’s search capabilities are deeper than a chatbot. You can access and compare results generated by multiple large language models, including OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini.

Agent Features

The next frontier of artificial intelligence will be agent AI, autonomous robots that can execute complex tasks from start to finish with minimal human guidance. These tools can generate complete documents, such as a comprehensive credit agreement, independently.

Hebbia’s pre-built “agents” (templates for generating MoUs, updating earnings summaries, and reviewing credit agreements and other routine materials) impressed me. Clearly, this is much more than a disguised version of ChatGPT.

The product receives regular updates and expanded offers. Recently, it launched Drafts, a feature that generates Word, PowerPoint or Excel files in a company’s template, automating the boring formatting work that can eat up young bankers’ time.

The big questions

The arrival of tools like Hebbia raises new questions: Are schools that teach finance prepared for what’s coming? What about banks’ analyst programs? The train has already left the station and they will have to catch it quickly.

But Mehta says some young entrants are ahead of the curve; At an annual workshop she has led for the past few years for early-career investment bankers at a company that uses the product, she was impressed by the growth of what every progressive class seems to know. “Every year I see them making more progress in their learning and their comfort with the tools,” he said.

For Hebbia founder Sivulka, this is a must. “If you don’t learn to use AI, you will necessarily become obsolete,” he told me.

One thing’s for sure: it’s only a matter of time until we all find out if he’s right.

Read the original article on Business Insider