Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

What your security team knows about AI that your board doesn’t

Four steps to bridge the gap between AI spend and AI savings.

[Photo: Getty Images]

Profile photo of Chris Schueler

CEO · Cyderes

Security leaders have mostly stopped debating whether agentic artificial intelligence works. The question now is what they’re getting back. That’s where board conversations get uncomfortable: 46% of security leaders using agentic AI report savings of less than $1 million.

The gap between AI spend and results is where the security team lives. Agentic AI needs data and context to work. In most environments the tools are poorly integrated with the rest of the stack. Without that contextual tissue, agents can’t produce accurate results.

Another failure point is poor governance. Agentic AI needs the same identity controls we apply to people. But most organizations still manage AI identities ad hoc, with 86% not enforcing access policies for AI identities. If this is left unaddressed, the agent meant to strengthen security becomes one of the biggest risks in the environment.

WHY AI RETURN ON INVESTMENT IS DIFFICULT TO TRACK

Boards often discover agentic AI underperformance late because nobody defined what success looked like before spending. When you don’t state what success means up front, measuring it later becomes subjective.

To the security team, success might mean cybersecurity outcomes: better alert filtering, stopping lateral movement, or flagging dormant accounts before exploitation. But these results don’t always land with a board. That outcome mismatch is why underperformance drags on.

The solution is straightforward: Define what success looks like in terms the board recognizes before the investment, not after. To do that well, security leaders need to understand the technology they’re buying well enough to know which business outcomes are realistic.

Closing the AI ROI gap is both a technology and leadership issue. These four steps can get the board and security leaders there.

1. DEFINE ROI METRICS BEFORE DEPLOYMENT, NOT AFTER

When leadership stays involved, AI investments return more value. The board sets the vision, success metrics, ownership structure, and reporting cadence. Those decisions govern how the AI gets used, managed, and judged.

Financial return is usually the first thing a board asks about. But much of what these AI agents deliver never shows up in a revenue line. For example, an agent prevents a breach that saves a cost that never came.

Capturing that invisible return is what a model like the Factor Analysis of Information Risk standard is built for. FAIR translates cyber risk into financial terms, breaking it into the probable frequency and likely magnitude of loss over a set period, expressed in dollars. Its parameters cover how often a threat actor is likely to act; how likely an attempt will succeed; the primary loss from response, replacement, and downtime; and the secondary loss from regulatory fines, reputational damage, and customer churn.

Those parameters answer questions a board can act on: How much financial loss do AI agents reduce? Which capabilities drive the biggest drop in loss frequency or severity? What’s the dollar value of preventing an agent failure, whether that’s misuse, privilege escalation, or data leakage?

Instead of defining AI ROI with vague language, such as “high risk,” FAIR would present a board-friendly probabilistic evaluation: “We estimate a 90% chance that annual loss from this scenario falls between $2 million and $8 million, with a most likely value around $4.5 million.”

2. SOLVE THE INTEGRATION PROBLEM

The 2025 MIT NANDA report debunked the myth that AI models are the primary constraint on delivering value. The real hindrance is tools that don’t learn your infrastructure or integrate poorly into your workflows. Cerby’s 2025 research put a number on it: 58% of organizations run identity operations on disconnected tools and inconsistent workflows.

Layer an AI tool onto a poorly connected environment, and it has nothing to learn from, so it can’t produce results that fit your infrastructure. That’s especially true in security operations, where understanding the environment is what tells a real threat from noise.

This is what contextual data in AI looks like in security operations center practice: If your AI tool notified you about a login session, it may not raise eyebrows. But it would force an investigation if the AI were connected to other SOC tools and centralized the alert like this: “a successful login from a blacklisted IP address using an unknown device without any multi-factor authentication.” That integration difference is work only the security team can do—wiring AI agents into the tools, workflows, and processes around the job.

3. IDENTITY SECURITY FOR AI IS COMPULSORY

This one also lands on the security team. Weak security around AI agents is a steady drag on ROI. It comes back to access. These agents already reach into organizational data and make their own decisions. Their scope must be defined, managed, and monitored. Leave it open and an agent in the wrong hands becomes one of the fastest ways in.

Enterprises can solve this by including AI agents in their identity management program. This ensures AI agents can be scoped to specific resources and subject to basic identity security principles, such as least privilege, privileged access management, and lifecycle management. Those aren’t the only controls you’ll need, but they’re what give you the documentation, oversight, and audit trail to keep an agent operating inside established boundaries.

4. INVEST IN THE HUMAN LAYER THAT MAKES AI WORK

Without investing in skilled professionals, the gap between installing AI tools and getting anything back only widens. Talent shortage is the top barrier to AI readiness. Only 35% of security leaders say they have properly equipped employees for AI roles.

Hiring AI-skilled professionals is one way to solve the talent crisis. Another way is to invest in training and adjacent skills, such as data analysis and interpretation. Statistically, employees who benefited from the latter approach report improved AI performance.

THE LEADERSHIP IMPERATIVE

Security leaders are under pressure to prove their AI spending is paying off in terms the board understands. But the returns are lagging because AI needs leadership involvement to set the success metrics and the right people to run it. Without that, agentic AI investments will keep underdelivering.

Chris Schueler is CEO of Cyderes.