Meeting data is the most underused asset inside a financial institution because it holds the reasoning behind decisions, yet almost no firm treats it as data. Emails, documents, trade records and transactions all have a system of record. The investment committee debate, the credit discussion and the client call usually do not. They end up in someone's notes, a recording nobody reopens, or one person's memory. Firms that capture those conversations, keep control of them, interpret them for different users and share them within proper boundaries can turn a perishable activity into durable institutional knowledge.

Capture → Own → Interpret → Share

Why meetings are different from every other data source

Most finance executives would not tolerate a trading desk that failed to log its executions, or a credit function that lost its loan files. Yet we routinely accept that the conversations behind those executions and loans disappear when the call ends.

On a recent Stack to Street episode, Dmitry Grankin, founder of the open-source meeting intelligence project Vexa, put the contrast plainly. He listed the data types institutions already manage well: emails, documents, execution logs and transactions. All of it, he said, is "persistent and it all has its own place." Then he named the gap: "meeting data doesn't have its own place." He added that those conversations "are lost if uncaptured," even though they are arguably the most valuable data a firm produces.

I think that framing is right, and it matters in finance in particular. Our industry runs on judgment. A credit memo records the conclusion, but the meeting where the committee pushed back on the collateral, questioned the sponsor or debated the covenant package holds the reasoning. When I was in banking, the important context was almost always in the conversation and rarely in the final document.

There is also a reason this data will not go away as AI matures. Grankin's view is that "meetings is where people build trust" and that this has been true for thousands of years. Later in the conversation he went further: "meetings is something you cannot automate… We cannot automate human responsibility, human authority." If that is correct, and I believe it is for the relationship and decision-heavy parts of our business, then meetings will become relatively more important as other workflows are automated. More of the institution's real judgment will be concentrated in the conversations.

Capture: give conversations a home

The first step is the simplest to describe and the one most firms skip. Conversations need a defined place to land, just as email has an archive and trades have a book of record.

Today, capture tends to be ad hoc. A managing director might let Zoom transcribe a call, save the file to a folder, and possibly upload it to a CRM. That is personal capture, not institutional capture. It depends on individual habit and produces scattered artifacts in inconsistent formats.

Platform fragmentation makes this worse. Grankin pointed out that Google Meet, Microsoft Teams, Zoom and physical meeting rooms each have their own vendor, and they "are not compatible with each other." Most large institutions use more than one of these, often because clients and counterparties choose the platform. A capture strategy tied to a single vendor's built-in transcription will miss part of the picture by design.

The executive question here is not which transcription tool sounds best. It is whether the firm has a consistent, cross-platform point of capture that feeds a single repository.

Own: keep the raw data inside your control

In financial services, capture without control is a non-starter. The most valuable conversations are also the most sensitive. They include client financials, investment views, deal terms and material business decisions.

Grankin described the target state as conversations that are "stored, secure, air gapped, self-hosted inside your organization." He also framed it in terms banks will recognize: "you own the software so it works inside your stack you don't rent software you don't put your precious data outside."

He acknowledged the tension too: "you want to own it, but you don't want to babysit it." That is an honest description of the build-versus-buy trade-off every technology committee faces. Self-hosting gives control but creates an operating burden. In practice, many firms will look for supported, self-hosted deployments rather than either extreme.

There is a client-facing side to ownership as well. In the episode I raised the awkwardness of a third-party bot joining a sensitive financing discussion. It is a very different conversation when you can tell a client that the bot is our bot, running inside our environment, functioning like an assistant. Grankin noted that institutions can self-host the language model layer as well, and that the data can be deleted if needed. Whatever architecture a firm chooses, the principle holds: own the raw record, not just a vendor-generated summary of it.

Interpret: one meeting, many perspectives

This is where most current approaches fall short. The typical pattern is to take the transcript, run it once through a language model, and store the summary. Grankin's critique is that this approach is isolated. The model "knows nothing about you". It sees only the transcript.

He also made a point I found clarifying. Different people take different things from the same meeting. A CEO, a CTO and a risk officer in the same room will each focus on something different. As he put it, "you have five people on a meeting, you need at least five summaries." That is why owning the raw data matters: you can reinterpret it later, for a different person or a different question, so that you "can interpret it all the time."

Consider a credit committee. The relationship banker wants the client commitments. The risk officer wants the concerns raised and how they were resolved. Compliance may want to know what was discussed and by whom. A single generic summary serves none of them well. Interpretation should be repeatable and tied to the role of the person, or the authorized AI agent, asking the question.

Grankin described the goal as building something like a "Wikipedia of company knowledge" rather than a pile of isolated summaries. Whether or not that exact model fits a given firm, the direction is sound. Meeting data becomes valuable when it connects to everything else the institution knows.

Share: respect natural boundaries

Personal knowledge repositories are not enough either. Grankin noted that there are always shared layers. Some knowledge is firm-wide and should be available to every employee and agent. Some belongs to a group, such as a recurring weekly meeting whose context should build over time. And "every organization's got natural boundaries for how their information is available."

In financial institutions, those boundaries are not optional. Information barriers, client confidentiality obligations and need-to-know principles shape who can see what. A meeting knowledge layer has to inherit those boundaries rather than flatten them. The opportunity is real. Picture an authorized agent preparing a credit committee pack that retrieves not only financials but the relevant reasoning from prior discussions of the same borrower. That opportunity only works if permissioning is designed in from the start.

What finance executives should ask

If you want to test whether your institution is treating meetings as an asset, these questions are a reasonable starting point:

  1. Where does meeting data live today? If the answer is "in individual folders and vendor clouds," you do not have a system of record.
  2. Does capture work across every platform we use? Clients choose their tools. Your capture layer should not depend on one vendor.
  3. Do we hold the raw record, or only summaries? Summaries lose detail. Raw data preserves the ability to reinterpret.
  4. Where is the data processed? Know whether transcription and language models run inside your environment or outside it, and whether that matches your risk appetite.
  5. Can interpretation be tailored by role? A banker, a risk officer and an AI agent need different views of the same conversation.
  6. Do access controls mirror our existing boundaries? Information barriers and client confidentiality must carry through to the knowledge layer.
  7. What are the retention and deletion rules? Treat meeting data with the same discipline as other records, consistent with your applicable regulatory and policy obligations.

The larger point

There are real governance, security and permissioning questions to work through before this becomes routine. I do not want to understate them. But the underlying idea is simple: conversations inside financial institutions are data, and right now most of that data evaporates.

The firms that build a disciplined path from capture to ownership to interpretation to sharing will retain more of their own judgment. In a business where judgment is what clients pay for, that may prove to be a meaningful advantage.

This is the kind of connection between builders and the front office that I explore on Stack to Street and in my work at Path66. If your institution is thinking through how to treat conversations as an asset, I would welcome the conversation.

About the author

Michael McLaughlin has more than 30 years in structured finance and private credit. At Macquarie Group from 2008 to 2023, he built the credit markets business from inception into a portfolio financing platform that averaged a $5 billion portfolio and generated more than $1 billion in cumulative net revenue. He has advised Hebbia on AI research software for financial services, and builds his own finance applications with AI coding tools and open-source software.

Through Path66, Michael works with financial institutions and investment firms on asset finance, liquidity, capital raising, data-driven due diligence and the practical application of emerging technology in finance. He hosts Stack to Street, where he interviews the developers building the software that will change financial institutions. Full bio.

Source

Your Firm's Most Valuable Data Is Trapped in Meetings, Stack to Street with Dmitry Grankin, Founder of Vexa. Full transcript on the episode page.