Case Study

Lumen’s Network Automation Journey: From Manual Workflows to AI-Assisted Autonomy

Five years. 350+ live workflows. 1 billion alerts collapsed to 57,000 actionable incidents. The operating model behind one of the world’s most-peered networks evolving toward 80% machine-to-machine operations – governed, measured, earned.

Challenge

A heterogeneous, acquisition-built network across IP, optical, and metro – with fragmented scripts, slow human-gated change pathways, and a need for governed, observable automation that could safely scale toward autonomy.

Solution

Itential as the orchestration engine and policy layer. Selector as the AI brain for observability. ServiceNow as the front door. Modular, atomic actions composed into runbooks – with pre/post-checks, approvals, and earned autonomy by design.

Impact

350+ live workflows. 1B alerts collapsed to 57K actionable incidents. Less than 2 minutes of automation downtime YTD. 1,000 workflows identified, 700 on the path to autonomous execution. A cognitive NOC where machines handle routine, engineers focus on innovation.

The Challenge

From Fragmented Scripts to a Governed Operating Model

Lumen operates one of the world’s most interconnected networks – AS3356, ~340,000 fiber route miles, ~163,000 on-net buildings, 350+ Tbps of backbone capacity, and 4 million Quantum Fiber locations across 16 US states. Built through acquisition and merger, the network is deeply heterogeneous, with legacy and modern domains spanning IP, optical, transport, metro, and voice.nnEarly automation efforts ran into integration and scale challenges – stitching together service activation and inventory stacks, internal portals like Global Change Requests, and service-assurance instrumentation so automation outputs could be controlled by consistent guardrails. Slow, human-gated change paths sat outside the operations team’s control. Scripts were fragile. Adoption was uneven. The team needed to re-platform from fragile scripts to governed, observable automations that could be exposed as services – and eventually as API-based MCP tools – without losing control of safety, auditability, or customer experience.

This wasn’t a tooling decision. It was an operating model decision – framing automation as a skills evolution, enforcing engineering discipline, measuring outcomes in business terms, and moving toward autonomous AI only when proven safe and valuable. Itential as the governed execution layer was the choice that made the rest possible.

I want the primary way to be that AI/ML interacts with the orchestrator to do closed loop automation. Natural language is a secondary way to interact with the network.

Greg Freeman

VP Network and Customer Transformation, Lumen

Why Itential

Why Lumen Chose Itential

Lumen’s vision required a platform that could be the governed execution layer beneath observability and AI – not another point tool, not a script library. The decision came down to integration depth, modular design, and the operational discipline needed to move toward autonomy safely.

The Governed Orchestration Layer Built for Autonomy

After early proof-of-concept work demonstrated rapid integration across diverse systems, Lumen scaled Itential as the orchestration engine and policy enforcer. Seven principles shaped the partnership – from clear swim lanes with Selector to the discipline that made earned autonomy possible.

Explore the Platform

Clear Swim Lanes: Brain + Hands

Selector serves as the AI brain – consolidating signals across IP and optical, surfacing probable causes, proposing or triggering remediation when confidence is high. Itential serves as the hands – ensuring AI-driven proposals translate into safe, auditable network actions through governed execution. Each tool excels at its job without stepping on the other.

Credibility Established Early

Itential’s executive proof point came in hours, not weeks: integrate quickly across diverse systems and prove orchestrations could be adapted when a system changed. That early demonstration established platform credibility and opened the door to broader service-assurance use cases the team scaled internally.

Modular Designs That Anticipate Autonomy

Atomic actions – read-only diagnostics like ping and trace, interface checks – composed into reusable runbooks. Pre- and post-checks, error handling, and rollback validation built in. Orchestrations engineered as if no human would be present. The discipline that paid off as automations graduated to autonomous execution.

Inclusive Builder Model: Low-Code & High-Code

Both low-code and high-code paths welcomed network engineers, software developers, and the rare unicorns who do both. Engineers were treated as citizen developers and automation architects – turning skeptics into contributors and accelerating output without forcing a culture war between code styles.

Governance & Discipline at Every Stage

Every workflow required PDD → SDD review, peer governance, naming standards checks, adapter validation, and AAA-scoped permissions. Five years of CI/CD evolution moved from three-hour manual maintenance windows in 2023 to fully automated zero-downtime deployments – the discipline that made headless execution safe.

ServiceNow as the Front Door

Internal automations exposed through ServiceNow as the consumption layer; selected actions exposed externally to customers through Lumen’s customer portal. Infrastructure leaders own orchestration. IT preserves governance. Both charters met without blocking delivery – fewer shadow IT tools, faster consumption, better telemetry on what’s used.

MCP u0026 Agentic Patterns Built on Governed Execution

Itential’s MCP server prototyped on the platform preserves all the guardrails while enabling tool-calling from northbound AI. Lumen’s MCP-based proof-of-concept now connects 90+ tools across multiple LLMs. Earned autonomy, governed exposure, agentic readiness – built on the operating model that came first.

When you’re operating infrastructure at Lumen’s scale, the question was never whether AI could help – it was whether we could trust it in production. FlowAI answered that. Our teams were building production-ready agents in minutes, within the same governance and access controls we already rely on. As we build the next digital backbone for AI, this is the next evolution in our journey with Itential – and it’s redefining how we operate networks at scale.

Greg Freeman

VP Network & Customer Transformation, Lumen

The Solution

Selector as the Brain. Itential as the Hands. ServiceNow as the Front Door.

Lumen built a three-vendor architecture with explicit swim lanes: Selector consolidates signals and proposes actions, Itential executes them under policy, and ServiceNow brings the work to engineers and customers in the systems they already trust. The same governance applies whether a human clicks run or an agent makes a call.

Atomic Actions Composed Into Runbooks

Engineers built diagnostics and configuration workflows from atomic actions composed into reusable runbooks – callable from the tools people already used. The repeatability built trust as operators saw the same action run the same way every time, with evidence and rollback baked in.

Pre/Post-Checks & Earned Autonomy

Pre- and post-checks, AAA-scoped permissions, code reviews, PDD/SDD gates, and stage environments – all applied identically to human-initiated and agent-initiated execution. To protect ourselves from ourselves, as the team put it after an early UAT-against-production incident hardened their stage gates.

Closed-Loop Remediation

Closed-loop scenarios where Selector detects, Itential remediates, ServiceNow tickets get updated, and confirmation flows back to Selector – keeping operators in one pane of glass with end-to-end evidence. Customer-facing example: a one-click guardrail-checked orchestration replaced a manual call-and-wait revert process.

ROI Metrics Streamed to Splunk

OpEx saved (replacing manual tasks) tracked separately from value created (running safe automations far more often than humans could). Streamed into Splunk dashboards. Workflow count, transaction volume, error rates per workflow – evidence that justified expanding autonomy where success conditions were proven.

CI/CD Pipeline Evolution Over Five Years

Five years of CI/CD evolution: from manual enterprise admin deployments to a 2.0 pipeline with two-approver requirements, comprehensive guardrails, and the elimination of deployment windows in mid-2024. Three-hour manual maintenance windows in 2023 became zero-downtime deployments.

MCP u0026 Agentic Operations as the Next Step

Itential’s MCP server preserves all guardrails while enabling tool-calling from northbound AI. A multi-agent proof-of-concept connects 90+ tools across multiple LLMs – powerful but not yet in production, precisely because determinism and QA need stronger guardrails. Earned, not assumed.

The Outcome

Production Impact at Lumen Scale

Numbers that compound across IP, optical, and service assurance domains – backed by the operational discipline that made earned autonomy possible.rn

350+

Live Workflows

Live workflows in production as of November 2025 – with 700 more identified as candidates for autonomous execution.

57K

Actionable Alerts (from 1B+ Raw)

Initial AIOps work collapsed 1+ billion raw alerts to 57K actionable incidents via 2,700+ ML patterns and 100+ expert rules – the prerequisite to closed-loop autonomy.

<2 Min

Platform Downtime YTD

Year-to-date downtime for the automation estate – evidence that the orchestration plumbing is itself highly available.

1,000+

Workflows Identified

Workflows identified for automation across IP, optical, and service assurance – with 700 on the path to autonomous execution.

80%

M2M Goal

Greg Freeman’s North Star: machine-to-machine operations across the network, measured, governed, and customer-facing where safe.

Earned Autonomy

The cognitive NOC where machines handle routine, engineers focus on innovation, and customers experience fewer events with fast, safe access to services.