QUODD’s Scott Argyres explains why firms evaluating market data providers should look beyond price and accuracy to A.I. licensing, provenance, flexibility and tokenization.
Financial institutions have spent decades refining how they evaluate market data providers. Accuracy, reliability, latency, coverage, price. Those criteria are well understood, straightforward, and defensible in
Scott Argyres, Head of Product, QUODD
front of a decision-making committee. They’re also increasingly insufficient.
The challenge is that they answer a narrowing share of the questions institutions face today. The next generation of financial infrastructure is being assembled around use cases most firms cannot yet fully specify. AI agents are beginning to consume data without a human in the loop. Tokenized markets are compressing trading, settlement and compliance into programmable workflows. Global platforms are pushing market access closer to continuous operation.
These developments create real opportunity, but they also raise unfamiliar questions about licensing, entitlements, provenance, governance and accountability. That changes the criteria for choosing the right partner.
The gap between using AI and integrating it
The scale of that gap is now measurable. AutoRek’s 2026 Investment & Capital Markets research, drawn from 250 senior operations, finance and technology leaders across the US and UK, found that 98 percent of firms report using AI in some form, while only 14 percent have actually integrated it into operations. The same study found 85 percent of firms saying their operational processes already struggle, or would struggle, to keep pace as volumes grow.
The distance between those numbers has little to do with technology. Firms have bought the tools. What they haven’t resolved is the foundation underneath: fragmented data, legacy pipelines, and asset classes the original architecture was never designed to carry. Digital assets were named the single greatest operational challenge by roughly 39 percent of respondents, up from 27 percent a year earlier, the sharpest single-year move in the study.
A market data provider sits directly on that foundation, which makes the selection decision considerably more consequential than a feed comparison.
Data for AI workflows is not the same as data for display
An AI system doesn’t consume data the way a terminal user does. It requires consistent schemas across asset classes, metadata rich enough to establish context, lineage that survives transformation, and entitlements granular enough to consume data in a compliant manner.
That last point is where most institutions discover the gap late. Training, inference, derived outputs and automated redistribution each carry distinct licensing implications. Many data agreements were drafted before those categories were meaningfully distinguishable, and a fair number still treat “non-display” as a single undifferentiated bucket. A firm can build an agentic workflow that works beautifully in development and turns out to be unusable in production because nobody asked what the license actually permits.
Surfacing use cases and contractual rights early in the discovery process matters more than ever as automation evolves. The cost of discovering an entitlement problem during a compliance review is materially higher than the cost of raising it during diligence.
New venues expose old assumptions
On-chain and tokenized markets are frequently described as another delivery channel, which understates what they actually introduce: continuous operation, programmable execution, and dependencies that reach outside the institution’s own perimeter.
Ostium, a decentralized perpetuals platform for real-world assets and a QUODD client, offers a useful illustration. Rather than rebuilding orderbook liquidity on-chain, it sources pricing from traditional markets as an input to its oracle infrastructure, using institutional equities, FX and commodities data to price instruments that settle on entirely different rails. The relevant observation for this audience has less to do with that specific architecture than with what it demands of the data. Institutional-grade content increasingly needs to travel into environments the original infrastructure was never designed to serve, and behave correctly when it arrives.
Tokenization is better understood as an operational restructuring than a technology upgrade, touching trading, settlement, custody, compliance and risk. Market data is the connective tissue across all five. If it cannot cross the boundary cleanly, neither can the workflow.
Six questions worth asking before you sign
The practical exercise is less about predicting the future than about testing whether a provider can move with you into it.
- Licensing adaptability. Can the provider articulate, clearly and in writing, how display, non-display, AI training, inference, derived data and redistribution are each treated? Language that stays vague at signing tends to get resolved against you later.
- Provenance and auditability. Can any given output be traced to its source, its transformations and the entitlement under which it was delivered? If a regulator asks how a number was produced, the answer shouldn’t require an archaeology project.
- Architectural flexibility. Can the provider support APIs, streaming, cloud delivery and formats that don’t exist yet, without requiring another infrastructure program in three years?
- Operational transparency. How are anomalies detected, escalated and corrected? Where is AI used in that process, and where do humans retain authority? A provider that can’t describe its own exception handling is asking you to trust a black box.
- Reversibility. This is the question buyers most often skip. What does leaving look like? Can you extract your historical data, your mappings, your derived logic, in a usable format and on a defined timeline? Is any of that written into the contract? A partner confident in its own value should be able to answer without discomfort, and where it can’t, you’ve learned something worth knowing. An exit right you can’t actually exercise is just a clause that made everyone feel better at signing.
- Partnership under ambiguity. Can the commercial, technical and operations teams help define an emerging use case, or can they only quote against an established product?
That last test is the differentiator, and the hardest to score. No buyer needs a vendor that claims to know the entire future. They need one willing to work through it in the open, including the parts where the answer is currently “we don’t know, but here is how we would find out.”
What you are actually buying
The temptation in any procurement cycle is to optimize for the requirement in front of you. It’s measurable and defensible, and it closes the file.
But the institutions that move fastest over the next few years won’t be the ones that bought the cheapest feed satisfying a 2026 specification. They’ll be the ones that preserved their ability to act when the 2028 requirement arrives, with licensing that stretches, architecture that adapts, and a provider whose first instinct when asked something unfamiliar is curiosity rather than a change order.
The best market data decision is no longer the one that satisfies today’s requirements at the lowest cost. It’s the one that leaves you the most room to move.
Is your company featured in this article? Contact us about reprints or licensing.
Leave a Reply