Skip to content 
How Data Agent Governs Trading Data Answers
An analyst wants to know why a particular desk’s fill rates dipped on a specific set of instruments last Thursday. It’s a reasonable question, the kind that gets asked in some form every day on a trading floor. And in a lot of firms, answering it still means opening a ticket, waiting for someone on the data team to get to it, watching that person write a query against three different systems, and getting an answer back two or three days later, by which point the question has usually stopped mattering as much.
This is a story about what happens when the people who understand the business question and the people who can actually reach the data live on opposite sides of a queue, come together. Trading and market data in most capital markets firms is scattered across systems that were built at different times, by different vendors, under different assumptions about who would ever need to touch them. Getting a straight answer out of that sprawl has never been simple, and building enough SQL skill into every analyst who might have a question was never a realistic plan.
The Answer
The obvious answer, letting people ask questions in plain language and get an answer without waiting in that queue, has existed in some form for a while. Most attempts at it in regulated environments ran into the same wall. The moment you make data easy to reach, you’ve also made it easy to reach the wrong way. A natural language tool that can answer any question about trading data can also, without meaning to, hand someone a number they weren’t cleared to see, or produce an answer nobody can trace back to its source when a regulator asks how it was calculated.
That’s the actual reason self-service analytics has been slower to land in capital markets than in almost any other industry. It’s not that the technology couldn’t answer questions. It’s that answering questions quickly and answering them in a way compliance can stand behind have historically pulled in opposite directions.
What Data Agent Does Differently
This is the specific problem NuSummit’s Data Agent was built around, and it’s worth being precise about what “governance intact” means in practice rather than treating it as a phrase on a slide. When someone asks Data Agent a question about trading or market data, the query runs against the same access controls, masking rules, and data lineage that already govern that data everywhere else. If an analyst isn’t cleared to see position-level detail on a particular book, Data Agent doesn’t see it either. The natural language layer sits on top of the governance layer. It doesn’t get to route around it.
Every question and every answer also gets logged with a full audit trail, not as an afterthought bolted on for compliance reporting, but as part of how the system works. If someone later needs to know exactly what was asked, what data it touched, and how the answer was derived, that record already exists. For anything sensitive enough to warrant it, a human reviews the query path before the answer goes out. The point isn’t to slow every question down. It’s to make sure the questions that need a second set of eyes actually get one.
None of this is unique framing on our part. It’s the same principle behind the broader shift toward governed, explainable analytics that’s happening across the data platforms we build on, including the work we’ve done with Snowflake and AWS through the Power of 3 partnership, where access controls, lineage, and policy enforcement are treated as part of the platform rather than a separate compliance exercise layered on top later.
What This Changes
The analyst from the earlier example doesn’t open a ticket. They ask Data Agent directly, in plain language, why fill rates dipped for that set of instruments last Thursday. The system pulls from the governed data it’s allowed to reach, returns an answer with the relevant context, and logs exactly how it got there. If the question touches something sensitive, it routes for review instead of just answering blind. Either way, the analyst isn’t waiting three days for someone else’s queue to clear.
That shift changes more than convenience. Questions that used to get dropped because they weren’t worth a three-day wait now get asked, because asking them costs almost nothing. A lot of the insight that used to stay buried in someone’s hunch, never quite worth the effort to chase down, becomes something people actually go check.
Where This Still Needs Human Oversight
Data Agent doesn’t decide what an analyst is allowed to see. It doesn’t decide what counts as a sensitive query worth a second look. Those are governance decisions your compliance and data teams make, and Data Agent enforces them rather than replacing the judgment behind them. Firms that get the most value out of this treat it as a way to make existing governance faster to act on, not a shortcut around having governance in the first place. The firms that try to skip that step, and just want a fast answer machine with no rules behind it, tend to run into exactly the problem that kept self-service analytics out of capital markets for so long.
The realistic starting point is usually a narrow set of data domains where the governance rules are already well defined, market data and desk-level performance metrics rather than anything touching client-level sensitivity on day one. As trust builds and the audit trail proves itself out in practice, the scope tends to widen from there.
Why This Matters Beyond Convenience
Faster answers are the visible benefit, but the deeper one is what happens to the relationship between speed and control. For years, capital markets firms have treated those as a trade-off, you could have one or the other, and every self-service analytics pitch that ignored governance was really just asking a firm to accept more risk in exchange for more speed. Data Agent’s whole premise is that this trade-off was never actually necessary, it was just hard to build the alternative. An analyst gets an answer in minutes instead of days, and compliance gets a complete, traceable record of exactly how that answer was reached. Neither side gives anything up to get the other.
That’s the part worth remembering the next time someone asks a data question on a trading floor and expects to wait three days for it. The wait was never really about the complexity of the question. It was about a gap between speed and governance that, for a long time, nobody had closed properly.
