Intelligent Ops: Why AI Agents Are Rewiring the Way Businesses Run

78% of organisations now report using AI in some form, up from 55% just a year ago.^[1] Yet only 1% describe their AI rollouts as mature.^[2] Only 10% of organisations using agentic AI report significant, measurable return on investment.^[3] And Gartner predicts that more than 40% of current agentic AI projects will be cancelled by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls.^[4]

The gap between adoption and impact is not a technology problem. It is an operations design problem.

Intelligent Ops is the operating model that closes that gap. It is the discipline of redesigning how work flows across a business: not deploying AI tools on top of existing processes, but rebuilding processes around human-agent collaboration. Organisations that get this right will run faster, make better decisions, and build a structural advantage that compounds over time. Organisations that skip this step will accumulate expensive pilots and declining confidence.

What Intelligent Ops Is, and Why the Distinction Matters

AI Adoption Is Not the Same as Operational Transformation

The first wave of enterprise AI was largely about access: giving people better tools, copilots, assistants, search, summarisation. That wave is effectively over. The majority of organisations now have some form of AI in their environment.

The second wave is agents. AI agents can plan, retrieve information, take multi-step actions, and complete end-to-end tasks with limited human intervention. They are not assistants waiting to be prompted. They are systems that work alongside, and increasingly ahead of, human operators.

This shift changes the nature of the challenge. Adding a copilot to an existing workflow is a technology decision. Introducing agents into operations is an operating model decision. It requires rethinking who does what, how authority is distributed, where human judgment is non-negotiable, and how performance is measured when part of the workforce is software.

McKinsey's 2025 global research makes this concrete: workflow redesign has the largest measurable relationship to earnings before interest and taxes (EBIT) impact from generative AI, larger than the AI tools themselves. Yet only 21% of organisations say they have fundamentally redesigned even some workflows. Most are deploying AI on top of old processes and wondering why the returns are modest.^[1]

Intelligent Ops is the answer to that question.

Digital Transformation, Not Technical Delivery

Intelligent Ops sits firmly in the tradition of operational transformation: the same discipline that drove lean manufacturing, service design, and business process reengineering. The underlying technology changes; the operating principle does not. You do not improve a system by automating its inefficiencies. You improve it by redesigning the system.

This is the distinction that most AI programmes miss. Technical delivery teams focus on models, APIs, and integrations. Intelligent Ops focuses on workflow maps, queue structures, decision authority, handoff points, and the human roles that need to change when agents take on routine execution.

The questions Intelligent Ops asks are not "which AI tool should we use?" They are:

• Where does work slow down, pile up, or degrade in quality?
• Which handoffs exist because of human bandwidth limits that agents could remove?
• Where does decision quality matter more than decision speed, and who should own those decisions?
• What does the operating model look like when agents handle triage, drafting, analysis, and routine execution, and humans own judgment, escalation, and exception?

These are operations questions. The technology answers them; it does not ask them.

Augmentation Before Automation

The strongest evidence from real deployments points toward a consistent principle: start by expanding human capacity, not replacing it.

A study published in the Quarterly Journal of Economics covering 5,172 customer-support agents found that AI assistance improved productivity by 15% on average, with gains concentrated among less experienced and lower-skilled workers, where issues resolved per hour increased by approximately 30%. Customer sentiment also improved. These results came not from removing humans from the process but from giving them better support within it.^[5]

Klarna's experience illustrates the risk of the opposite approach. After an aggressive push into AI-driven automation that included significant reductions in headcount and vendor relationships, CEO Sebastian Siemiatkowski publicly acknowledged the misstep. In a Reuters interview ahead of Klarna's US initial public offering (IPO) in September 2025, he said the company had "over indexed a little bit" on cost-cutting and was now trying to "course correct," shifting focus back to improving services and products.^[6]

The durable operating model is human-led, agent-supported. Agents handle what is repetitive, measurable, and reversible. Humans own what requires judgment, accountability, and trust. Intelligent Ops is the framework for deciding which is which. The next question is what that framework runs on.

Why AgentOps Is the Foundation of Intelligent Ops

The Gap Between Deployment and Production

Most organisations treat AI deployment as a project with a go-live date. Intelligent Ops treats it as a managed capability with an ongoing operational discipline. The difference is AgentOps.

AgentOps is the set of lifecycle practices for running AI agents in production: observability, performance evaluation, drift detection, incident classification, rollback procedures, and change governance. It is to AI agents what DevOps is to software: the production discipline that makes deployment sustainable rather than fragile.

The case for investing in AgentOps is not theoretical. McKinsey's survey found that 27% of organisations review all generative AI output before use, while a roughly equal share check 20% or less.^[1] A significant proportion of organisations have agents acting in production with inconsistent or minimal human oversight. Gartner's warning is the logical consequence: by 2028, the average Global Fortune 500 enterprise could have more than 150,000 agents in use, yet only 13% of organisations believe they have the right governance in place.^[7]

Without AgentOps, scale creates risk, not advantage.

AgentOps as the Implementation Layer of Intelligent Ops

Within the Intelligent Ops framework, the relationship between the operating model and AgentOps is clear.

Intelligent Ops defines the what and the why: which workflows to redesign, how to structure human-agent authority, what outcomes to target, and how the operating model evolves as trust is built and scope is expanded.

AgentOps delivers the how: the technical and operational practices that make agents reliable, observable, and controllable in a live business environment.

An organisation that runs good AgentOps without Intelligent Ops will have well-monitored agents doing the wrong work. An organisation that designs good Intelligent Ops without AgentOps will have sound operating intentions and brittle production. Both are required.

The key AgentOps practices that Intelligent Ops depends on:

These practices are not overhead. They are what separates a pilot from a production capability.

The Compliance Dimension

Governance is no longer optional. The EU AI Act is in phased application, with prohibited practices and AI literacy obligations already in force and the majority of rules applying from August 2026.^[8] The UK's Information Commissioner's Office has been explicit: meaningful human review of automated decisions requires reviewers with the knowledge, authority, and independence to actually challenge outcomes, not ceremonial sign-off on decisions already made.^[9]

The National Institute of Standards and Technology (NIST) AI Risk Management Framework and its Generative AI Profile push organisations toward continuous monitoring, structured feedback loops, and mechanisms to disengage systems whose behaviour becomes inconsistent with intended use.

For operations leaders with regulatory exposure in financial services, healthcare, insurance, or the public sector, AgentOps is not just good practice. It is the operational infrastructure that makes compliance achievable at scale. Intelligent Ops integrates this from the outset rather than retrofitting it after incidents occur.

What Intelligent Ops Delivers: The Metrics That Matter

Intelligent Ops is not measured by AI adoption rates or number of agents deployed. It is measured by operational outcomes. The metrics that matter to COOs and operations leaders fall into five categories.

Throughput and Cycle Time

The most direct measure of Intelligent Ops performance is how much faster qualified work moves through a process: from intake to resolution, from request to decision, from submission to outcome.

• Cases or tickets resolved per agent hour: Baseline vs. post-deployment, segmented by complexity tier.
• End-to-end cycle time for target workflows: median and 90th percentile, tracked weekly.
• Queue depth and age: Reduction in backlog volume and time-in-queue for high-priority cases.
• First-contact resolution rate: The proportion of interactions resolved without escalation or rework.

Quality and Accuracy

Speed without quality is not improvement. Intelligent Ops tracks quality as a first-class metric, not an afterthought.

• Accuracy rate on agent-handled decisions: Measured against human-reviewed ground truth, updated regularly.
• Reviewer disagreement rate: The proportion of agent outputs that human reviewers override or correct, used as a leading indicator of drift.
• Rework rate: Cases returned for correction or escalated after initial agent handling.
• Customer satisfaction scores: Tracked separately for agent-handled vs. human-handled interactions to identify experience gaps.

Cost Per Outcome

The business case for Intelligent Ops is built on cost per unit of output, not total cost reduction. This framing resists the temptation to cut headcount as a primary objective and instead focuses on what it costs to deliver a unit of service at quality.

• Cost per case resolved: Total operational cost divided by cases resolved, tracked over time as volume and agent scope expand.
• Cost per decision: For workflows with discrete decision outputs such as approvals, classifications, and recommendations.
• Agent return on investment (ROI) payback period: Time from first production deployment to the point where accumulated operational savings exceed deployment and operating costs.

Control and Compliance

For regulated industries and functions with meaningful compliance exposure, Intelligent Ops delivers control metrics that satisfy audit and regulatory requirements.

• Coverage of meaningful human review: The proportion of agent decisions subject to qualified human oversight before or promptly after action.
• Incident rate by severity tier: Frequency and classification of agent failures, broken down by impact level.
• Time to detection and time to remediation: How quickly drift or failure is identified and corrected.
• Governance checklist completion rate: The proportion of agent deployments with documented authority limits, rollback plans, and review cadences in place.

Adoption and Capability Maturity

Intelligent Ops is also a capability-building programme. The organisation's ability to deploy, govern, and expand agent use over time is itself a strategic asset.

• Workflow coverage: The proportion of target workflows with at least one production agent deployment.
• Time to first production deployment: From workflow selection to live operation, as an indicator of internal delivery capability.
• Expansion ratio: The rate at which successful initial deployments expand in scope or replicate to adjacent workflows.
• Manager and reviewer confidence scores: Self-reported measures of how effectively frontline leaders understand and can act on agent performance data.

That last metric matters more than most. An organisation that can measure its own agent performance honestly has already cleared the bar that most cannot. The metrics above are not a board deck. They are the operational feedback loop that makes the next deployment cheaper, the next workflow redesign faster, and the next governance decision easier. Build the measurement system before you need it.

Why Intelligent Ops Creates Strategic Agility

The Operating Model Advantage Is Cumulative

Organisations that implement Intelligent Ops early are not just running operations more efficiently today. They are building an operating infrastructure that compounds. Each workflow redesign produces a reusable pattern. Each governance decision produces a documented standard. Each agent deployment produces evaluation data that makes the next deployment faster and lower-risk.

This is the structural advantage that is hardest to imitate quickly. A competitor can procure the same AI technology in weeks. They cannot replicate two years of workflow redesign expertise, governance frameworks, production observability, and operator capability.

Responding to Market Conditions Without Rebuilding from Scratch

Intelligent Ops also changes the economics of operational change. In a conventional operating model, responding to a significant market shift, a regulatory change, a competitive disruption, a demand spike, or a new product line typically requires hiring, retraining, process redesign, and technology procurement. Each takes time. Together they take months.

In an Intelligent Ops model, a significant portion of operational capacity sits in agent-handled workflows that can be reconfigured, redirected, or scaled without changing headcount. A new regulatory requirement that changes how decisions must be documented does not require retraining every operator. It requires updating the agent's action thresholds and audit logging. A demand spike in one channel does not require emergency hiring. It requires routing more volume through existing agent infrastructure while human capacity focuses on complex cases.

This is not theoretical. Across the deployments where early returns have been clearest and most durable, a consistent pattern holds: the operating model was built before the technology scaled. Workflow redesign first. Human-agent authority defined early. Governance in place before volume grows. The result is operational flexibility that compounds into competitive advantage.