A Promising New App in Any Asset Manager's Tech Stack
I have recently joined OpenBB as Head of (Quant) Research after my previous fintech firm Visible Alpha was sold to S&P Global. It has been almost a decade since I graduated as a PhD economist. Since then, my career involved a great deal of buying and selling datasets and technology in capital markets.
I am still a baby if you ask industry veterans but hey, I learned a thing or two about a thing or two. I’m here to argue that OpenBB is the most promising solution to many problems the asset management industry faces today. It feels like staring at the groundwork of a new skyscraper that will be the talk of the city over the next decades. It feels refreshing. I said “many problems”, plural, but let’s make it simple: it is the cost of data and infrastructure problem to run a fund (or a brokerage, an advisory,…).
The road to Alpha is paved with good infrastructure
Tomes have been written on asset returns already and if you have ever studied financial economics, even at college 101 level, you have probably heard things like the [Efficient Market Hypothesis](Efficient Market Hypothesis). People love to debate whether alpha exists, where, how much, and for how long, etc. But ask anyone who has been professionally involved in capital markets and they will agree that there is no get-rich-quick scheme out there. You don’t hear much about the cost side of it all though. Cost is so deadly dull, and tedious, and stuffy, and boring, and desperately dull as Monty Python’s infamous Accountant sketch once exclaimed:
It is very expensive to run a professional financial services firm in a cutthroat capital market. A large chunk of that cost is the data and technology you need to buy or build. You employ highly skilled people in short supply, and you can’t easily outsource it to another part of the world.
I hope you see where I am going with this: the future success of your firm in capital markets is not likely to come from a breakthrough discovery of a magical investment strategy no one has thought through before. However, if you can somehow achieve the same returns for half the vendor costs and without hiring additional armies of analysts and portfolio managers you poach from rivals with eye-popping pay packages, you may just have a shot at being a Wall Street legend. Deadly dull ideas from good ol' fashioned economics; I love it.
OpenBB’s all-in-one solution
I joined OpenBB because I see the seedlings of ideas here that can give any asset manager that magical shot: a sleek, versatile, and comprehensive interface to financial data, designed to deliver you the productivity boom Generative AI promises, at a fraction of your existing “terminals”, “desktops”, “workstations”, or “platforms” you are cajoled into buying.
I can hear you think this is too good to be true. The secret is in the word "open" in OpenBB (and not the BB which refers to BlackBerry as our co-founder explains here).
Let me explain with a three-legged stool metaphor. For full disclosure, I asked Dall-E to draw one for this blogpost and it kept drawing four-legged stools, claiming it has three legs… So I gave up on eating my own dog food and did an image search to find a nice looking three-legged stool created by a named human being. 😄
1. Data
We are not selling you data.
We want you to purchase the best datasets, at the best price, from the best vendors. We might want to offer you some data partnerships eventually but it is optional to use them.
We are equally agnostic about how you want to warehouse them: cloud databases, on-prem, any flavor of SQL, or a data lake (or two). You do you. And it never stops there, does it? There are scores of websites you visit often, PDFs you receive, internal data points you reference, RSS news feeds you subscribe to, spreadsheets you work on, your meeting notes… I can go on and on.
Now imagine that there is one app you need to login to interface all of the above; one interface to rule them all. That’s what OpenBB wants to be: an interface to your unique data zoo.
We can’t achieve that by being a walled garden of OpenBB-approved datasets only. We want to be versatile enough to integrate any dataset you throw at us instead. This is a fundamental difference in product philosophy between OpenBB and the conventional wisdom in the industry.
2. Dashboards
We don’t sell you pre-packaged interfaces.
There are no thousands of identical pages to sift through, no keystrokes to memorize to get there quickly, no pre-built experiences that take 50 meetings with customer success and product managers to pass your feedback so they let you customize it a little in six months. 😅
At OpenBB you design your own window to financial data using the sleek, modern, and versatile widgets infrastructure we provide for you. You are at the steering wheel, not OpenBB product managers!
We do offer you a starter widget library: economic data release calendar, price charts, news screener, etc.; but we plan to open-source widget templates so our users can create and integrate their own to OpenBB.
Sometimes our users share their dashboards on social media (or directly with us), and surprise, surprise: no two OpenBB experiences look the same!
Take a look at these two crypto-focused dashboards shared to us:
If we tried to design these “experiences”, it would take us ten meetings to lock down the specs, negotiate compromises to what to do if we are missing this or that component, then another ten to gather client feedback, and another ten to iterate on version 2, and we would still end up with half-happy clients. No such thing at OpenBB; we give you the infrastructure, you pour your data and creativity into solving your own problem.
3. Generative AI
I am certain there are enough articles on the internet already speculating about how fruitful AI will be in financial services so let’s instead talk about all the nuts and bolts of the infrastructure needed to harvest those fruits. This is where I feel the skepticism over AI’s potential in finance is rooted because there is enough hyped-up talk over where we want to be and not nearly enough substantive talk over how we get there.
- Agent workflow ✅
ChatGPT described it as:
An agent workflow refers to a framework where a task-specific agent operates within a defined process to achieve goals. Unlike general-purpose large language models (LLMs), which are designed to be broad and flexible, agents in a workflow are tailored to specific tasks, domains, or use cases.
Financial analysis is a collection of highly specific tasks that an investor would demand high precision for. If all that a vendor is offering: “here is OpenAI summary of this financial document for you”, run away and don’t look back! They haven’t thought through how LLMs would integrate into investor workflow just yet.
We liken OpenBB Copilot to a symphony orchestra: when you send your prompt to OpenBB, our current conductor OpenAI takes a first look, breaks it down to a bunch of sheet music (tasks), then picks the right instrument (agent) to play those assigned sheets, switch and adjust the tempo to complete the composition. Did someone ask for a chart? Call in Ag Grid in the end! Web search? Bring in Perplexity API at the start! We haven’t launched our code interpreter yet but hopefully in 2025, the conductor might call Claude the Pythonista to write Python code, but ask GPT-4 model for SQL. This framework adapts and scales with every new tool and gismo coming in to the market to do better and better at specific tasks relevant for some working with financial data. We believe it is indispensable for OpenBB’s success at institutional level adoption at a very demanding industry.
- Context management ✅
It is the subject of every cringe LinkedIn personality’s humor post. Yes, LLMs make stuff up if you ask a loaded, nuanced question and give it no context to answer it. They are not good at saying “I don’t know”. Neither do 80 million+ Turkish people according to Cem Yilmaz, a famous comedian (turn on closed caption and enable auto-translate!), so LLMs are not alone in this vice.
Enter OpenBB’s magical + button on widgets.
If you are an investor looking at a data or chart widget in your dashboard, this is how you invite Copilot for help. OpenBB Copilot works with the context you provide and you control what goes in and out of context within the chat thread.
Even ChatGPT, Claude, Gemini apps themselves don’t provide such extensive control: the conversation you are having is influenced by the context of the past thread so if you had some document or link at the beginning that you now want the LLM to ignore, you are better off deleting the thread and start anew! This is understandable as these apps are not designed for an investor working with plethora of connected datasets and the context needs to adjust to the ebbs and flows of the research agenda.
Suppose you don’t point to a specific widget for context. OpenBB Copilot doesn’t rush to answer the question but starts with searching what widgets you have in your library that it can use as context to answer your question:
At the end, the Copilot will always cite the data source it referenced, so you can easily double check it.
OpenBB widgets come with descriptions and title (auto-generated even if you don’t type anything!), and user-input metadata if you want to give your Copilot a hand.
OpenBB Copilot searches this metadata to pick the most relevant widget, cites the source on the screen for you to audit, and proceeds to answer within that self-decided context!
Copilot simply stops to ask for directions if the widget search doesn’t yield results; it is more useful to say “please give me more context” than throw in the towel and say “I don’t know.”
We believe disclosing the step-by-step reasoning behind OpenBB Copilot to the investor like an audit trail is going to become the industry standard for building trust to work with LLMs in high stakes finance world!
- Analysis ready outputs ✅
Chat interface naturally gravitates towards textual output to textual input.
The world of finance is one giant Excel spreadsheet for analysts, portfolio managers, and investors sometimes so there is a missing UI layer to make LLMs immediately useful in finance. Vendors' go-to “export data as CSV” is not a thoughtful solution; the user still needs to open this data in another application, complete the analysis, and figure out a way to combine the new data with all the other ones. It is not a solution when it requires several more steps of work to be remotely useful!
OpenBB Copilot outputs a table or a chart from the responses that turns into another widget on your dashboard directly. This enables iteratively building up complex workflows in OpenBB where you start at one widget, derive something useful from it, make another widget that becomes a context to the next analysis, and repeat.
Take a look at this recorded earnings update workflow with OpenBB.
- I asked the Copilot to extract specific management guidance from a transcript, convert it to numeric-only data with metrics on rows, and the low, mid, high end of the guidance range the management communicated.
- I added this widget to my dashboard so I no longer need the textual context.
- I use the guidance widget to scan for the metrics in the 10-Q filings and append the reported numbers next to the guidance.
- I can then compute percentage surprise from the management guidance and keep augmenting my dashboard.
- I asked the Copilot to calculate the market reaction to the earnings announcement using the historical price chart, then put the surprise and market reaction together to offer me some hypothesis what might have driven the stock prices.
- One hypothesis made sense to me, so I concluded the analysis with scanning the news to back up this hypothesis.
At the time of the recording, the Copilot couldn’t generate charts directly but now I can ask some of these charts directly as my end goal.
- On-prem Deployment and Custom Copilots ✅
Now comes the biggest white elephant in the room when you are doing business with a large institutional investor: they don’t want their vendors peaking into their workflow and getting ideas.
In the old days, the investor would be concerned about the vendor logging which ticker or metric the analyst is pulling data for and reverse-engineering the strategy behind the data requests. LLMs are much nosier than some harmless usage logs.
Now I can sit down and enjoy the stream of exact phrasing of what questions are asked to which datasets because OpenAI’s public cloud service is going transmit that information from my application to OpenAI and taking a quick peak at those juicy little nuggets of insights is just too irresistible. I would not expect the investment industry to be eager to adopt any LLM solution using this privacy nightmare of a setup.
OpenBB offers two solutions.
First, we can be deployed to a private cloud with ease because a large part of the OpenBB ecosystem is already open-source. Does your private cloud of choice enable OpenBB Copilot’s APIs? Even better, you can continue to use OpenBB Copilot in your cloud and have full control over what gets logged and who gets to see it.
It gets a little trickier if OpenBB Copilot loses access to key APIs it needs in your private cloud. But we got you covered: you can add your own custom copilot regardless of where OpenBB is deployed! This is a great solution when institutions develop their own LLM products in-house but don’t want to deal with creating the entire infrastructure to make it accessible across the organization or when they have a specific preference to do business with one LLM provider over the other, or just want to use an open-source Copilot they host somewhere.
The same philosophy with data and dashboards applies to OpenBB Copilot: you can bring your own and you are not obliged to use ours!
In a nutshell
This blog turned into a much longer write-up than I originally anticipated but hey, I am excited about OpenBB.
I hope I made a compelling case for anyone launching a new fund or revising an existing data infrastructure of an asset manager. You should give OpenBB a serious look. It is not a one-trick pony. It is a comprehensive solution at an unbeatable price.
