The Head of Investor Relations' Guide to AI Agents

Investor relations meets AI from two sides. Internally, the preparation and monitoring work that fills an IR team's week is largely assembly and analysis, which agents handle well. Externally, investors now ask what the company is doing with AI and increasingly want evidence rather than ambition. This guide covers both, with the disclosure discipline that keeps the second from becoming a problem. It is general guidance, not legal advice.

Where do agents help IR internally?

Task	Today	Agent work
Peer and sector analysis	Manual reading of transcripts and filings	Extracts themes, guidance changes, and question patterns
Question bank	Spreadsheet maintained after each call	Maintained continuously with prior answers and sources
Briefing materials	Assembled manually before every meeting	Drafted from internal data, prior disclosures, and recent coverage
Analyst and sentiment tracking	Ad hoc	Monitored continuously with changes flagged
Shareholder analysis	Periodic	Continuous signal tracking from available data
Disclosure data assembly	Manual gathering across functions	Assembled with completeness checks
Post-call follow-up	Manual	Drafted, tracked, and chased

All of it is internal preparation. The output is reviewed by people before anything external happens.

What stays human, absolutely?

External communication. Every statement to investors, analysts, or the market carries securities law implications regardless of who or what drafted it, and the standard of accuracy does not change because a model produced the first draft. IR professionals author, legal reviews where applicable, and the company stands behind the words. An agent that sends an external communication autonomously is a disclosure incident waiting to happen. The AI guardrails explained for executives piece covers the gate design.

What must never enter an external AI system?

Material non-public information. Draft earnings materials, unannounced guidance, transaction information, and unreleased financial data should not be processed in consumer or unprotected AI services. Options: enterprise arrangements with contractual protection and no training on inputs; private or restricted deployments; or keeping the work out of AI systems entirely. Set the rule explicitly, train the team on it, and confirm the approach with counsel and the CISO. The private AI explained for executives piece covers deployment options.

What do investors ask about AI?

Increasingly specific questions: what is in production rather than planned; what measurable effect it has had, with baselines; what it costs to run; what the risks and dependencies are, particularly single-provider concentration; and how it changes the competitive position. Companies that answer with a roadmap and adjectives invite follow-up; companies that answer with two named processes, baseline-to-actual numbers, and a run cost differentiate themselves immediately.

Preparing those answers is an IR function working with the AI program owner and finance. The how to brief investors on AI guide covers the construction; the AI value realization whitepaper covers producing the evidence.

Where is the disclosure risk?

In the gap between the investor narrative and internal reality. Overstating AI capability or adoption has attracted regulatory enforcement in several jurisdictions, and the evidence used is usually the company's own internal reporting. The discipline is simple: public claims must be specific, supportable, and consistent with what management actually reviews. Never describe a pilot as a deployment or research as a product.

A practical control: before any AI claim goes into a script, deck, or release, confirm it against the monthly management review. If the review would not support it, it does not go out. The agentic AI for public company executives guide covers the wider disclosure picture.

How should IR handle an AI incident?

As a potential disclosure matter with a prepared process: know who assesses materiality, what facts are needed, how quickly they can be obtained, and what the holding statement says. Agent incidents can involve customer data, service failures, or erroneous communications, and the assessment should not be improvised. The what executives should do in the first hour of an AI incident guide covers the response sequence.

What should IR measure?

Preparation time per event and per meeting; question bank coverage of questions actually asked; time to produce disclosure data; analyst question themes over time, including AI questions; and internally, hours returned from assembly to relationship and analysis work.

What should heads of investor relations ask?

• Could we answer an analyst's AI question with production numbers today?
• Does every AI claim in our materials match what management reviews internally?
• What rule governs material non-public information and AI tools, and does the team follow it?
• Who assesses materiality if an AI incident occurs, and how fast can we get the facts?
• How much IR time goes to assembly rather than investor relationships?

How can FISTA Solutions help investor relations teams?

FISTA Solutions builds internal preparation and monitoring AI agents with strict data controls, private deployment options for sensitive material, and human authorship preserved for anything external, and works with executive teams through its AI enablement practice to produce the measured evidence investors now ask for. Since 2017, FISTA has delivered 150+ projects for 50+ companies across 12+ countries.

To prepare evidence-backed answers before your next earnings call, talk to FISTA on WhatsApp, or read the CFO's guide to AI and agentic AI.