When the Data Consumer Is No Longer Human: Rethinking Market Data Licensing in the era of AI.
- Bill Bierds
- 3 days ago
- 7 min read
Executive Summary
Market data licensing was built for a world in which data was primarily consumed by people through terminals, screens and desktop applications. That model is now under pressure as trading platforms, risk engines, pricing tools, research platforms, data lakes, APIs, AI agents and automated workflows increasingly call, process and act on data directly.
In this environment, the core licensing question is no longer only who can view the data. It is which systems can use it, for what purpose, under which entitlements and with what level of auditability.
As machine consumption expands, market data licensing must evolve from a user based commercial model into a governance and control layer for automated data use.
We are going to explore 6 areas:
The new market data consumer is not always a person
Per user licensing struggles to describe machine consumption
Non display use is becoming the default, not the exception
Entitlements and usage control need to move toward machine identity
Audit risk increases when data usage becomes invisible
Market data licensing is becoming an infrastructure function
1. The new market data consumer is not always a person
Historically, market data consumption was easier to define. A firm could identify who had access to a terminal, who viewed exchange data, who used a desktop application and which teams required subscriptions. Licensing models could therefore be built around visible human consumption.
But machine use changes the shape of the problem.
Today, market data may flow into a risk model, a portfolio valuation engine, an order processing workflow, a research automation tool or an AI agent. The data may be queried, transformed, enriched, summarized or used as an input into another system without ever appearing on a screen.
This is not a marginal use case. It is becoming a core mode of consumption.
CME Group defines non display use as the use of information in a system, process, program, machine or calculation, rather than for display or redistribution purposes. The examples include areas such as P&L calculation, portfolio valuation, order processing, risk management and research. This definition shows that market data consumption has already moved beyond the human display model and into machine processes. (cmegroup.com)
CME Group’s Market Data Policy Education Center also treats licensing policy, units of count and non-display licensing as distinct policy areas, underscoring how important usage classification has become in market data governance. (cmegroup.com)
The key question is no longer simply, “Who can see the data?” It is also, “Which systems are using it, for what purpose and under which license?”
2. Per user licensing struggles to describe machine consumption
Traditional licensing models work best when the user is clear. A human user can be named, counted, assigned to a business unit and governed through access rights. A firm can ask how many users require a data service, which desks they sit on and whether they are consuming display or non-display data.
Machine consumers are harder to classify.
One AI agent or automated workflow may access multiple data sources, call several tools, generate outputs and pass those outputs into downstream systems. It may support a human analyst, trigger a workflow, feed another model or create derived information. In that context, a simple per user fee does not always reflect the actual pattern of consumption, value creation or licensing risk.
The licensing questions become more complex:
Who or what accessed the data?
Was it a person, an application, an agent, an API or a workflow?
Was the data displayed, processed, stored or redistributed?
Was it used for internal analysis, automated execution, compliance, risk management or model input?
Did the output create derived data?
Was that output reused elsewhere?
The rise of agentic workflows makes this more urgent. The Model Context Protocol, or MCP, is described as an open standard that allows AI applications to connect with external systems, data sources, tools and workflows. (modelcontextprotocol.io)
Research on MCP deployment also notes that while MCP can standardize external tool discovery and invocation for AI agents, production scale use requires additional mechanisms such as identity propagation, observability and tool budgeting. (arxiv.org)
For market data licensing, this matters because the consumer may no longer be a static user. It may be a dynamic chain of systems, tools and agents.
3. Non display use is becoming the default, not the exception
Non-display use was once treated by many firms as a specialized licensing category. It applied to specific trading systems, risk calculations, pricing engines or automated processes.
In a machine consumption environment, non-display use becomes much more central.
Data may be used in portfolio valuation, pricing, order routing, surveillance, compliance, research automation or model calibration. It may support human decision making without being directly displayed. It may also support automated action without a human reviewing each data point.
That means AI agents and automated workflows will often sit closer to non-display licensing than to traditional display licensing.
CME Group’s non display licensing FAQ includes uses such as P&L calculation, portfolio valuation and management, order processing, risk management, trade internalization and research. (cmegroup.com)
NYSE’s Non-Display Use Policy defines non display use as accessing, processing or consuming real time market information for purposes other than display or redistribution. (nyse.com)
NYSE’s broader market data policy materials also cite algorithmic trading, automated order or quote generation, smart order routing, investment analysis, surveillance, risk management, compliance and portfolio management as examples of non-display use. (ia801407.us.archive.org)
This creates a practical challenge for data teams. Non display use can no longer be managed as an edge case. It must be treated as a primary licensing category for modern market data operations.
The question becomes: if an AI agent uses market data to analyze, summarize, classify or trigger a workflow without displaying that data to a person, how should that use be classified?
4. Entitlements and usage control need to move toward machine identity
As machine consumers grow, entitlement management also needs to evolve.
In the human user model, entitlement is typically linked to a person’s role, desk, department, application access or terminal rights. But machine use introduces new actors: AI agents, APIs, workflows, data pipelines, models, applications and automated processes.
These actors need identity, permissioning and audit trails.
An AI agent does not simply “read” data. It may retrieve data, combine it with other information, summarize it, create an output, trigger another tool or pass results into a downstream system. That means entitlement must move beyond the question of who can view the data and toward a more granular question: which machine actor can use which data, in which context, for which purpose and under which constraints?
Enterprise AI governance frameworks are already moving in this direction. Microsoft’s guidance on AI agent governance highlights the need for data governance, compliance, data location controls and operational controls for enterprise agents. (learn.microsoft.com)
Microsoft’s Agentic AI maturity model also emphasizes governance, security controls and lifecycle management as requirements for operating agents in a stable and predictable way. (learn.microsoft.com)
Security research on MCP also points to new security and privacy risks that can arise when AI systems interact with external tools and servers. (arxiv.org)
For market data, this suggests that entitlement models must increasingly account for machine identity. A firm may need to know not only which employee has access to a dataset, but which agents, applications, services and workflows can call it.
5. Audit risk increases when data usage becomes invisible
Market data audits are already challenging for many financial institutions. Data often moves across complex technology landscapes, including terminals, trading systems, internal databases, spreadsheets, dashboards, data lakes and vendor platforms.
Machine consumption adds another layer of complexity.
When data is displayed to a person, usage may be easier to identify. When data is called by a system, transformed by a workflow, summarized by an AI agent or stored as part of a derived output, the usage trail can become harder to reconstruct.
Future audits may need to answer questions such as:
Which source data was accessed?
Which system, agent or workflow accessed it?
Was the data displayed, processed, stored or redistributed?
Was derived data created?
Was the output reused by another system?
Was the use permitted under the relevant license?
Can the firm prove that usage after the fact?
CME Group’s non display licensing FAQ is useful here because it frames machine, system, process and calculation-based usage as a licensing category that must be understood and classified. (cmegroup.com)
NYSE’s Non-Display Use Policy similarly treats non display access, processing and consumption as a distinct policy and fee category. (nyse.com)
Research on MCP production deployment also notes that production environments require infrastructure level mechanisms such as identity propagation, observability and structured error handling. (arxiv.org)
The implication is clear: auditability cannot be treated as an afterthought. In a machine use environment, auditability becomes an infrastructure requirement.
6. Market data licensing is becoming an infrastructure function
Machine consumption changes the role of licensing.
It is no longer only a legal, procurement or vendor management issue. It is becoming an infrastructure function that sits across data architecture, entitlement, observability, metadata, usage controls, audit logging and governance.
In the next phase of market data management, firms will need to do more than review contracts. They will need to operationalize licensing rules inside their systems.
That means building the ability to:
Classify machine use and non-display use clearly.
Manage entitlement not only for people, but also for agents, applications and workflows.
Track how data moves across systems and where derived outputs are created.
Connect usage rights with technical controls.
Treat audit readiness as part of everyday governance rather than as a periodic remediation exercise.
This is the deeper shift behind machine consumption. The industry is not simply moving from human users to automated users. It is moving from licensing as a commercial agreement to licensing as a control layer.
In the era of AI, the future of market data licensing will not be defined only by who can see the data. It will be defined by which machines can use it, for what purpose, under which controls and with what proof.
