AI ROI breaks down without governance across risk, savings and output

At a glance

What this is: This analysis argues that AI ROI should be measured across defensive, efficiency and productivity returns, not through a single pilot metric or finance lens.

Why it matters: IAM, security and platform teams need a shared measurement model because AI value depends on governance, visibility and identity-aware controls across human users and autonomous agents.

By the numbers:

• Only 39% of organizations report an enterprise-level EBIT impact from AI.
• 30% of projects were abandoned by 2025 due to poor data quality, inadequate risk controls, escalating costs, and unclear business value.
• 72% of organisations have experienced or suspect they have experienced a breach of non-human identities, 46% confirmed and 26% suspected.

👉 Read WitnessAI's full AI ROI framework for defensive, efficiency and productivity returns

Context

AI ROI is not just a finance question. It is a governance question about whether AI systems create durable value once they move from pilot into production, where security controls, operating cost, and adoption behavior all change at the same time. For identity and access teams, the issue is how to measure value without losing sight of who or what is using AI, what it can reach, and what the organization can prove about it.

The article’s central premise is that ROI needs to be measured across multiple stakeholder lenses because CISO, CFO and CTO concerns are not interchangeable. That matters for NHI and autonomous AI programmes because access, visibility and policy enforcement shape both risk and return, especially when AI is embedded in employee workflows, developer tools, and agent-driven execution paths.

Key questions

Q: How should organisations calculate AI ROI across security, finance and productivity goals?

A: Start with three separate baselines: defensive ROI for avoided loss, efficiency ROI for reclaimed spend, and productivity ROI for shipped outcomes. Assign each to the stakeholder who owns it, then connect the measurements through shared identity, usage and policy data. That prevents one department’s success metric from hiding another department’s cost.

Q: Why do AI ROI models often fail after a successful pilot?

A: Pilots usually measure activity, not production durability. Once AI moves into governed environments, data controls, access oversight, compliance logging and operating costs change the return profile. If those factors were not baselined early, the programme looks profitable in testing but underperforms in production.

Q: How can security teams prove defensive ROI from AI governance?

A: By linking AI activity to visible outcomes such as reduced breach exposure, lower compliance overhead and less Shadow AI usage. The evidence has to be attributable to identity, policy and audit controls, otherwise the claim is only theoretical. Defensive ROI is strongest when the organization can show what loss was avoided, not just what was blocked.

Q: What should leaders do when AI usage is spreading faster than governance can see it?

A: Prioritise visibility first, then classification and policy enforcement. Hidden AI use creates both unmanaged spend and unmanaged risk, so the first control objective is to discover where AI is being used, by whom, and with what data. Once that is clear, finance and security can evaluate the same usage from their own perspectives.

Technical breakdown

Why single-lens AI ROI models fail in production

Single-lens ROI models collapse because they treat AI value as either cost reduction or productivity lift, then ignore the operational controls required to sustain it. In practice, pilot metrics can show enthusiasm while production introduces data governance, monitoring, compliance, and identity attribution overhead. That means the business case often misses the real cost of keeping AI safe enough to scale. For identity teams, the key point is that measurable value depends on governed access, auditability, and usage visibility, not just model performance or user adoption.

Practical implication: baseline AI cost, usage, and control data before rollout so production ROI can be measured against real governance overhead.

How defensive ROI turns AI security into a measurable outcome

Defensive ROI measures value avoided, not value created. It includes breach cost avoidance, regulatory exposure reduction, and Shadow AI containment, all of which become measurable only when AI activity is visible and attributable. That makes identity, policy enforcement, and audit trails part of the ROI equation, because you cannot prove avoided loss if you cannot show who used what AI, with what data, under which controls. This is especially relevant where AI use touches sensitive data or external models.

Practical implication: connect AI usage logs to identity and policy records so breach avoidance and compliance exposure can be quantified.

Intent-based classification and network visibility as ROI controls

Network-level visibility and intent-based classification change AI ROI from a forecast into an operating metric. Visibility tells you what is happening across users, apps, agents and model calls. Intent classification adds context by distinguishing productive use from wasteful or risky use, which helps route requests, block unsafe prompts, and attribute spend to outcomes. For identity governance teams, this is the bridge between technical controls and business reporting, because the same signal can support security, finance and productivity measures without rebuilding the data model each time.

Practical implication: instrument AI traffic with identity-aware telemetry so usage, spend, and risk can be reported from the same control plane.

Breaches seen in the wild

• Cisco DevHub NHI breach — IntelBroker exploited exposed Cisco credentials, API tokens and keys in DevHub.
• DeepSeek breach — DeepSeek breach exposed 1M+ log lines and sensitive secret keys.

Read our 52 NHI Breaches Analysis report for a comprehensive view of breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

AI ROI is now an identity governance problem as much as a finance problem. Once AI is embedded in employee workflows, developer tools and agentic systems, the question is no longer whether adoption happened but whether it can be governed well enough to survive production. That makes visibility, attribution and policy enforcement part of the return calculation, not just security plumbing. Practitioners should treat ROI measurement as an access and control discipline, not a post-hoc spreadsheet exercise.

Single-metric ROI models understate how AI value is created and destroyed. A CISO, CFO and CTO are each measuring different forms of loss or gain, and a one-dimensional model cannot reconcile them. This is why pilot success so often fails to survive deployment: the model tracks activity, but not the governance and operating conditions that preserve value. The implication is that AI programmes need a shared measurement framework that respects stakeholder-specific outcomes.

Network-level visibility is the named concept that makes AI ROI operationally defensible. The article shows that measuring spend, risk and adoption only works when AI interactions are observable across human users and autonomous agents. That visibility is what allows intent-based classification, audit trails and policy routing to tie usage to outcomes. Practitioners should view visibility as the prerequisite for credible ROI, not as an optional control layer.

Defensive ROI and efficiency ROI both depend on the same hidden premise: AI use must be attributable. Breach avoidance, compliance reduction and spend recovery cannot be measured if the organization cannot map usage to identity, data and intent. That premise fails quickly in Shadow AI conditions, where usage sits outside procurement and governance. The implication is that AI value measurement and AI identity governance are the same programme from different angles.

Productivity ROI only holds when access is governed tightly enough for teams to use AI safely at scale. The article’s strongest point is that safety does not suppress adoption when controls are visible and usable. Instead, guardrails make AI usable in places where blunt restriction would push people to shadow tools. Practitioners should therefore measure not only speed gains, but also whether governance is broadening legitimate use or forcing it underground.

From our research:

• 72% of organisations have experienced or suspect they have experienced a breach of non-human identities, 46% confirmed and 26% suspected, according to The 2024 ESG Report: Managing Non-Human Identities.
• Enterprises that have experienced a compromised NHI averaged 2.7 separate incidents in the past 12 months, which shows how quickly one exposure can become a repeat problem.
• That is why practitioners should pair AI visibility with lifecycle controls, as outlined in Ultimate Guide to NHIs , Lifecycle Processes for Managing NHIs, before ROI claims are allowed to reach the board.

What this signals

Identity-aware AI measurement is likely to become a standard control expectation. As AI moves from experimentation to production, leaders will need a way to connect usage, spend and exposure in the same reporting model. Programmes that can attribute activity to users, agents and policies will be better positioned to defend both budget and governance decisions.

Shadow AI is not just an adoption issue, it is a measurement failure. Once unsanctioned tools sit outside the control plane, organizations lose the ability to compare cost, risk and productivity on the same basis. The practical response is to treat discovery, classification and access attribution as prerequisites for any credible ROI conversation.

With 38% of employees admitting to sharing sensitive company data with AI tools without permission, per Ultimate Guide to NHIs , Why NHI Security Matters Now, the boundary between user convenience and governance is already crossed in many environments. That makes AI ROI a control problem as much as a financial one, and it will push more teams toward identity-linked telemetry and policy enforcement.

For practitioners

• Build a three-part ROI model Map defensive ROI to security and compliance, efficiency ROI to finance and IT operations, and productivity ROI to engineering and business leaders. Use separate baselines for each so the same AI investment is judged against the outcomes each stakeholder owns.
• Instrument AI usage with identity-aware telemetry Capture who used which model or agent, what data was involved, and whether the request was allowed, warned, blocked or routed. That creates the evidence needed to connect AI activity to spend, risk and adoption.
• Treat Shadow AI as both a governance and cost problem Inventory unsanctioned AI tools, duplicate subscriptions, and unmanaged personal usage so hidden spend and hidden exposure are measured together. The same visibility layer should support procurement cleanup and security review.
• Tie production approval to measurable control signals Require audit trails, policy enforcement and data tokenization evidence before expanding AI beyond pilots. That gives risk committees a basis for approval and lets finance see when governance is enabling, rather than delaying, value creation.

Key takeaways

• AI ROI fails when organizations evaluate pilots in isolation and ignore the production controls that determine whether value survives scale.
• The strongest AI business cases now combine avoided loss, reclaimed spend and productivity gains so different stakeholders can defend the same investment.
• Identity-linked visibility is the turning point between AI as a forecast and AI as an operating metric.

Key terms

• Defensive Roi: Defensive ROI is the value an organization proves by reducing losses from AI-related risk. It usually includes breach avoidance, lower regulatory exposure and reduced Shadow AI impact. The measure only works when AI activity can be tied to identity, policy and audit evidence.
• Efficiency Roi: Efficiency ROI is the measurable spend recovered by making AI operations cheaper or less wasteful. That can include model routing, lower license waste and automated compliance work. In practice, it depends on visibility into usage patterns so savings can be separated from hidden risk or duplicated tooling.
• Productivity Roi: Productivity ROI is the output gained when AI helps people or teams ship more work safely. It includes faster delivery, better task throughput and broader adoption inside guardrails. For identity programmes, the key question is whether governance makes legitimate use easier rather than pushing activity into Shadow AI.
• Shadow Ai: Shadow AI is the use of AI tools, models or agents outside approved governance channels. It creates blind spots for security, finance and compliance because the organization cannot reliably see who is using what, with which data, or under which policy. That makes it both an access problem and a cost problem.

Deepen your knowledge

AI ROI measurement across defensive, efficiency and productivity returns is covered in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building governance for AI systems, agents and the identities that support them, it is a useful next step.

This post draws on content published by WitnessAI: AI ROI is the measure of whether enterprise AI investments are creating business value that leaders can defend. Read the original.