In search of truth: Avanzai’s north star

About 10 years ago, I had my first internship at Bloomberg: a networks internship where I spent most of my summer in NYC at the coolest hip spot in NYC's West Village: Bloomberg's data center. I spent most of my time there helping decommission old server racks that held an outdated identifier for instruments in favor of a recently introduced one, FIGI. At 20 years old, I learned the ins and outs of financial data identifiers and their equivalents in other competitors. In retrospect, 10 years later, I now realize that at every step of my career, these identifiers are simply a representation of finding truth, and they laid the foundation for why I built Avanzai.

I. What is truth?

Deep question for a B2B financial blog post, I know. But throughout my career, the most requested feature by all users and clients has been: how can I integrate my data with your product? The question quickly turns into how my database can communicate with a piece of information that is crucial for analysis.

In the financial data world, that question quickly becomes complex as you dive into different asset classes. Let's take corporate debt as an example. Apple as a company issues equity shares on the NASDAQ, but they also raise debt in the fixed income markets. Their shares (ID ABC) and their debt (ID XYZ) are completely separate from the entity itself (ID DEF). In financial data, we call this the entity versus instrument relationship. We call this reference data, the holy grail of truth for a financial services firm.

The problem is, most financial data vendors have their own identifiers for each. Factset bought CUSIP/ISIN, Bloomberg has FIGI, Reuters has RIC codes, and certain asset classes have dominion from a provider such as Markit with CDS. Yes, many codes are open-sourced nowadays, but the problem still persists if, for example, a fund's database is 90% from one vendor and suddenly wants to integrate a new dataset from another vendor. This becomes a massive ask for their financial data team, and all the firm is seeking is... truth.

The best way I explain this problem to non-finance folks is using cuisine as an analogy. Throughout human history, we've viewed culinary truth to fall into two buckets: absolute truths and relative truths. An absolute truth in cooking would be assuming that there is only one correct way to prepare a dish; all other methods are wrong. A relative truth would be that in some regions of the world, they prepare a dish one way, while in other parts of the world, others prepare it differently. Both regions can coexist on this Earth, enjoying their own versions of the dish.

Well, until now in the financial data space, each vendor treated their identifier as an absolute truth, balking at the idea of users wanting to integrate with different vendors. Consultants and B2B platforms have long taken advantage of this, promising custom integrations of these vendors from one to another. And in the age of AI, this gets even trickier, as these LLMs are being powered by context (RAG) and decision making (agents) that ultimately provide personalized answers to users. But most solutions fall short since the same data issue persists: users want a relative truth, but the friction is still there.

II. Why AI agents?

No, it's not because it's the hottest craze right now. Trust me on this one.

The main reason we're all in on AI agents is that they offer users two significant advantages over their current B2B software stack: agency (reasoning) and personalization (data).

Agentic software is a game-changer. It replaces the need for those cumbersome, months-long projects where you have to figure out how to migrate databases, merge tables, or normalize data. Instead, AI agents provide intermediate steps that can reason through the tools provided and generate code that's personalized to the user's context.

Put plainly, agents allow companies to think of software as their newest hire. All you have to do is provide context and procedures, and these digital employees can make decisions on their own. The personalization aspect means that these AI agents can adapt to your specific needs and data structures. They're not one-size-fits-all solutions, but rather customizable tools that mold themselves to fit your unique business processes.

In the world of finance, where data is king and time is money, AI agents represent a powerful new paradigm. They're not just a trendy add-on; they're the future of how we interact with and extract value from our data.

III. Where does Avanzai come in?

When GPT-2 came out back in 2022, it was the first time users could use natural language to output Python that could actually be executed in a Python environment. Every iteration of GPT since then has added to that use case: increased edge cases and updated Python versions (GPT-3), an understanding of your codebase and preferred coding method (RAG), and a deep understanding of the best practices within your vertical (fine-tuning). The financial services industry has realized this opportunity, and there are now multitudes of vendors providing copilots (Factset, OpenBB), LLMs (Bloomberg, Databricks), and agents (😉).

Short term, the goal is to allow financial institutions an option to create their own relative truth: combining structured and unstructured data from their vendors into one truth internally. AI agents (either created by us or the user) allow users to do this by having enough context and decision-making agency to know how to map across vendors and intelligently cite sources. Long term, the goal is simple: let users focus on investing and skip all the busywork... outside of a chatbox. We do not seek to take more screen space, we do not seek to replace your Bloomberg terminal or be your vendor of choice. You can think of us as the Zapier of Finance: connecting your favorite tools and automating your most hated workflows so you can be the most productive investor possible.

The plan is to focus on automating workflows that we believe are highly connected and are part of the day-to-day operations of many financial institutions: report generation, chart creation, and cleaning/standardizing financial data. These workflows form the pillars of many portfolio managers and risk managers across the world when it comes to portfolio rebalancing, risk exposure, and more.

IV. So where do we go from here?

Feedback, feedback, feedback: we are working to improve the product and give users what they really want. If you are interested in a demo and believe AI agents can help your company, please reach out to us for a demo here. We truly believe the future of finance is outside of a chatbox and not doing the same thing over and over and over and over again. 🙂

Avanzai users create automations by pre-purchasing credits that power the AI agents behind these automations. These automations typically create deliverables such as reports, charts, and soon, spreadsheets, decks, SQL tables, and more. This credit-based system allows users to flexibly scale their usage based on their needs.

Enterprise users uniquely benefit from our financial data AI agents, which can normalize and clean the data they upload. This ensures that their workflows are seamless and data is consistent across all analyses and reports.

Here's an example workflow we're developing with a wealth manager: a weekly portfolio holdings report generated using Avanzai:

Interested in seeing it in person? Book a demo with us, and we're happy to show you full enterprise use cases of Avanzai and how funds are currently using Avanzai at the moment.