What Evidence Do Investment Firms Need to Support Their AI Claims?
Investment firms may be tempted to make broad claims about how AI improves their operations. “AI informs our investment process” can make them sound like a more sophisticated partner. But no AI claim should be broader than the evidence a firm has to support it..
Before approving a statement, decision makers should ensure they can identify the capability in use, show how it performs in specific workflows, and assign human responsibility for keeping the statement accurate as workflows change. If the evidence supports a pilot rather than a deployed capability, the language should say so.
This is a practical standard for any firm that wants to speak confidently about its technology. It is particularly important when firms present AI as a reason to trust their investment process, an operational capability, or the firm itself.
Why do AI claims require evidence?
In a March 2024 press release, the Securities and Exchange Commission announced settled charges against two registered investment advisers over false or misleading statements about their purported use of AI. According to the SEC, Delphia did not have the AI and machine-learning capabilities it described in its investment process.
The case raises serious questions about how firms should discuss AI use. A technology team may be testing a tool while a business team discusses its future potential. A vendor may describe what its platform can do, while the firm uses only one feature. Public language can blur those distinctions, even when the people writing it believe they are describing a genuine initiative.
The best safeguard is a reliable connection between the statement and the operational reality. A firm should be able to move from the words on a website or in a presentation back to the workflow those words describe. If that connection is weak, the firm should review the claim before using it.
What is the firm claiming?
The first step is to determine whether a statement describes an intention, a current practice, or a result. Each calls for a different kind of support. For example:
• “We are exploring AI to assist our research team” describes an initiative. • “Our analysts use AI to summarize approved research materials” describes a current workflow. • “Our AI improves investment decisions” goes further: it makes a claim about an outcome.
Consider an AI assistant that summarizes public filings for an analyst. The firm may be able to show that the tool is deployed, that analysts use it, and that its summaries are reviewed before they inform further work. That supports a carefully worded description of research assistance. It would not, on its own, establish that AI selects investments or improves portfolio performance.
This distinction gives executives a useful way to examine language before approving it. Rather than asking only whether a statement sounds accurate, ask what a reader would reasonably understand it to mean. A narrow, well-supported claim can be more credible than a broad one because readers can understand the role AI actually plays. It also leaves room for the description to develop as the capability matures.
What evidence supports a defensible AI claim?
There is no single document that can validate every statement about AI. The evidence depends on what the firm says and how consequential its use. The evidence should match the claim and the role AI plays in the underlying workflow. Leaders can apply a practical standard without treating every use case as identical.
Show that the capability exists
The firm should first be able to demonstrate the process it describes. A successful demonstration in a test environment does not establish that the capability has become routine practice. The evidence needs to reflect the firm’s own use, not simply a vendor’s description or an implementation plan.
That means identifying where the AI capability is used and what people do with its output. For example, a demonstration of an AI research assistant might show an analyst providing an approved filing, reviewing the resulting summary, and using it as a starting point for further research.
Show that it performs as described
The firm must have tested the capability in a setting relevant to the claim. A tool used to summarize filings should be evaluated for the quality of those summaries and for the way analysts use them. A test showing that it produces readable text would not, by itself, substantiate a claim about better research judgment.
The International Organization of Securities Commissions issued guidance for securities regulators concerning AI and machine learning used by market intermediaries and asset managers. Its guidance emphasizes oversight of development, testing, deployment, monitoring, and controls. For investment firms, the practical implication is that evidence should come from how a system performs in its intended workflow.
Keep an owner close to the claim
A well-supported statement can become inaccurate without a human in the loop to update it when circumstances and use cases change. An accountable owner should be able to confirm the current use and bring material changes to the attention of whoever approves external language. That responsibility cannot sit entirely with a communications team. It requires a dependable route back to the people operating and evaluating the system.
For the research assistant example, a change from summarizing approved documents to generating draft recommendations would warrant a fresh review. Any new claim about the expanded capability would need evidence suited to that new role.
Why do claims drift from reality?
An AI initiative often crosses organizational boundaries. Business leaders define the opportunity; technology teams implement the tool; analysts or operations staff use it; another team describes it to outside audiences. Each group may understand one part of the picture without seeing the whole workflow.
This division of work is normal. Problems arise when one part of the organization has only a limited view of the workflow. That perspective alone may not be enough to support a statement about the firm’s investment process.
Firms need enough operational visibility to identify active AI tools and assign clear ownership for them. In addition to better messaging, leaders gain a more reliable picture of where AI is contributing value and where expectations have moved ahead of implementation. The broader operational lesson is that a firm needs a way to test its own words against its own practice.
The practical issue is straightforward: language should evolve with the capability. If the system improves and the evidence supports a more specific claim, the description can become more ambitious. If the process changes or the evidence weakens, the wording should change accordingly.
What should leaders ask before approval?
Executives do not need to review every technical record behind an AI tool. They do need confidence that a business owner can demonstrate the AI working in the process described and that the relevant team can support the specific claim. The more consequential the claim, the more carefully leadership should examine its meaning and evidence.
Credible claims start with operational evidence
Investment firms should be able to speak confidently about useful AI capabilities. Confidence comes from knowing precisely where those capabilities operate, what they contribute, and where their limits remain.
That discipline supports external communications; it also helps leaders distinguish a promising experiment from an established practice and a functioning tool from a demonstrated result. When those distinctions remain visible, claims can grow more specific as evidence accumulates, without getting ahead of the work itself.
How Option One Technologies can help
Accurate claims still require judgment from the firm’s business, communications, and legal or compliance teams. A dependable operating environment gives those teams a clearer account of what is actually in use.
Option One Technologies helps investment firms strengthen the technology foundations around emerging AI workflows. Through consulting and implementation, managed IT, cloud services, and cybersecurity, Option One can help firms improve visibility into deployed systems and support more consistent operations as those systems evolve.
Executive FAQs
What counts as an AI claim by an investment firm?
An AI claim is a statement about how the firm uses AI, what an AI-enabled capability does, or what results it produces. The key distinction is whether the statement describes a plan, a deployed workflow, or a demonstrated outcome. The evidence should support the meaning a reasonable reader would take from the claim.
Can a firm describe an AI pilot publicly?
Yes, provided the description accurately presents it as a pilot. The firm should avoid wording that suggests the tool is already deployed broadly or has produced results it has not established. Whether a particular statement meets applicable legal requirements depends on the firm’s circumstances and should receive appropriate review.
Is vendor documentation enough to support an AI claim?
Vendor documentation can help explain what a product is designed to do. It does not, by itself, establish how an investment firm has configured the product or used it in its own workflow. A claim about the firm’s current practice should be supported by evidence from that practice.
