Why IT Operations Haven’t Accelerated After Decades of Automation and AI

About the Author:
Vinod Kumar is Head of IT & Engineering Services for Akkodis North America. He works with enterprise leaders to modernize IT operations and engineering environments through AI, automation, and next-generation delivery models.

Why IT Operations Haven’t Accelerated After Decades of Automation & AI

IT leaders have invested in automation for many decades. The tooling has evolved quickly, moving from RPA to machine learning, then AIOps, then generative AI, and now agentic AI. On paper, that progression should have reduced ticket backlogs, shortened resolution times, and made IT operations feel lighter.

In practice, many teams still feel stuck. Ticket volumes keep climbing, escalations remain common, and major incidents still pull the same people into the same late-night decision loops.

The core issue is not entirely due to an AI limitation. It’s that IT operations still run on an operating model where humans remain the decision engine at almost every step. AI can speed up individual agents, but it does not speed up the overall system when the operational workflows still require human interpretation, validation, routing, and approval before anything meaningful happens.

The Automation Progression Outpaced the Operating Model

 The technology story is easy to track:

1. RPA helped automate repetitive clicks and form updates.
2. Machine learning improved detection, classification, and prediction.
3. AIOps reduced alert noise and helped connect signals across tools.
4. Generative AI improved search, summarization, and draft responses.
5. Agentic AI is beginning to promise broader action and execution.

A lot of value has come from this, especially in analyst productivity and time saved per ticket.

The sticking point shows up when you look at the end-to-end flow. Most teams are still running the same workflow shape, with new tools layered on top. As a result, the gap between AI capability and system-level performance keeps widening.

The Main Bottleneck Is Human Latency in the Decision Chain

A typical incident or service request often moves through a series of checkpoints that are still human-dependent:

• An alert fires and someone decides whether it is real or noise.
• The issue is validated, enriched, and translated into a ticket.
• The ticket is routed, escalated, or reassigned based on judgment.
• A remediation step is proposed, reviewed, and approved.
• The fix is executed, monitored, and closed out with confirmation.

None of these steps are unreasonable. The problem is that, together, they create systemic delay. Even when AI reduces the time to summarize logs or draft a response, the overall clock keeps running while humans wait on context, align stakeholders, and approve next actions. 

This is why teams can feel busier even after adopting AI automation. AI increases throughput at the individual level, but resolution time stays stubborn when the workflow still depends on humans for every decision point.

Why Ticket Volumes and Backlogs Keep Growing

Leaders often ask why work keeps piling up when so much automation is in place. The answer usually lives in the space between tooling and execution.

Alert and Ticket Inputs Keep Expanding 

Monitoring is broader, environments are more distributed, and dependencies are more complex than they were ten years ago. Even a well-tuned system can generate more signals than a human team can triage quickly.

Fragmented Knowledge Forces Human Interpretation

When the information needed to act is scattered across runbooks, tribal knowledge, old tickets, and multiple monitoring tools, humans end up acting as the integration layer. Generative AI can help summarize and search, but teams still hesitate to act unless they trust the source-of-truth and the path to remediation.

Google’s SRE guidance on reducing toil highlights a related point. Automation needs safeguards and should default to humans when conditions are unsafe, which is a practical reminder that trust and safety checks have to be designed into the workflow, not bolted on at the end.

Triage Automation Is Not the Same as Resolution Automation

Many organizations have done a solid job automating triage steps such as deduplication, categorization, and ticket assignment. Those are helpful, but they do not remove the slowest part of the chain, which is decision-making and execution.

A ticket that is routed faster is still a ticket. The system only accelerates when the workflow can safely move from “identified” to “resolved” with fewer human gates for routine issues.

Why AI Improves Agent Productivity but Not Resolution Time 

If you only measure productivity per person, AI often looks like a win. People write faster, find answers quicker, and spend less time on repetitive tasks.

If you measure mean time to resolution, the picture is different. The end-to-end timeline is governed by the slowest points in the workflow, and those points are often human approvals and escalations.

This creates a familiar scenario:

• AI helps generate a strong recommendation.
• A human still needs to validate it.
• Another human still needs to approve the change.
• Another team still needs to execute it in a separate system.

The system did not become faster. It simply produced better inputs into the same workflow shape.

What Must Change for Agentic AI to Deliver Autonomy

Agentic AI can add real value in IT operations, but only when the operating model evolves to support autonomous decision-making for routine work. The goal is not to remove humans from IT ops, but to remove humans from routine decision loops so they can focus on exceptions, complex issues, oversight, and governance.

That operating model shift usually requires a few concrete changes.

Define Which Decisions Can Be Delegated

Not every ticket should be autonomous. Teams need clear boundaries for what AI can decide and execute, especially for low-risk, repeatable issues.

Build Guardrails That Make Autonomy Safe

If the workflow does not include confidence thresholds, rollback paths, and safe fallbacks, autonomy becomes risky. Effective autonomy always includes robust guardrails, following AI ethics and responsibility best practices.

Connect Workflows Across Tools 

Agentic execution breaks down when the workflow is split across disconnected systems and handoffs. Modern incident management practices already emphasize structured ownership, escalation, and lifecycle flow, and those same ideas become even more important when automation starts taking actions.

Design Governance Up Front

Governance works best when it’s part of the workflow, not a separate review board that slows everything down. The goal is fast routine execution with strong oversight on exceptions.

A Case Study in Effective Workflow Redesign

Akkodis supported a major retailer with an AI-enabled IT service desk modernization that combined workflow redesign with AI capabilities. Within the first year, the modernized service desk achieved a 35% increase in first-call resolution, a 40% reduction in call abandonment, and 25% faster response times. 

Those outcomes matter because they reflect an operating model shift where routine work becomes easier to resolve quickly and consistently. 

How Akkodis Can Help

Your IT ops team is overwhelmed because humans remain the decision engine in almost every workflow, and that design creates unavoidable delay at scale.

Agentic AI can transform IT operations when organizations redesign operational workflows to support AI-driven decision-making and autonomous execution for routine issues, while keeping humans firmly in the system for exceptions, governance, and accountability.

We help organizations modernize IT operations, apply practical AI automation safely, and build the operating model needed to scale AIOps and agentic capabilities without increasing risk.

To learn more about how we can support your organization, connect with our experts to assess where autonomy can be introduced safely and where workflow design needs to change first.