EMA Finds Strong Interest in AI for NetOps - But Only 35% Call It a Complete Win
Enterprise Management Associates (EMA) released new research on AI-driven network operations based on a survey of 458 IT professionals. Engagement is high, pilots are common, and many teams are experimenting with commercial and homegrown tools. The catch: just 35% say their AI-driven network management initiatives have been completely successful.
That gap matters for Operations. Interest without execution shows up as alert fatigue, brittle integrations, and projects that stall after the first proof of concept. The signal from the study is clear: focus on repeatable outcomes, not experiments that look good in a slide deck.
What Ops Leaders Should Take From This
- AI-driven NetOps is no longer fringe. Teams are testing anomaly detection, predictive alerts, automated triage, and config insights.
- The execution gap is real. Tools are advancing faster than processes, data quality, and org readiness.
- Success looks operational: fewer incidents, faster mean time to resolve, cleaner changes, and measurable ticket deflection - consistently.
Why AI NetOps Projects Stall
- Vague problem statements and weak success criteria.
- Messy telemetry: incomplete logs/flows, inconsistent tags, and siloed data lakes.
- "Black box" recommendations with no operator trust or audit trail.
- Integration debt across ITSM, CMDB, config management, and observability stacks.
- Skills gaps: model lifecycle, feature pipelines, and prompt quality for LLM use cases.
- No clear ROI model or executive sponsor to keep momentum after the pilot.
A Practical Path From Pilot to Production
- Pick 1-2 use cases with direct business impact (e.g., incident triage or change risk scoring). Define precise KPIs and a 90-day target.
- Fix data first: normalize device metadata, standardize tags, and ensure time-synced logs, metrics, and flow records across vendors.
- Start with human-in-the-loop. Let AI assist, operators approve. Automate only what's proven stable.
- Operationalize the model lifecycle: versioning, drift detection, rollback plans, and weekly review of false positives/negatives.
- Integrate where work happens: ITSM tickets, CMDB, chat, and runbooks - not a separate dashboard no one checks.
- Set guardrails and governance for model use, explainability, and data risk. Use the NIST AI Risk Management Framework as a starting point.
- Measure everything: baseline before, A/B during, and track costs saved per alert, per ticket, and per change.
- Upskill the team. Pair Ops with data/ML talent and create shared ownership. For structured learning, see our AI courses by job.
KPIs That Prove It Works
- MTTD and MTTR trends (median and P90).
- False positive rate on AI-driven alerts and recommended actions.
- Change failure rate and time to remediate failed changes.
- Ticket deflection and auto-closure rates with accuracy thresholds.
- SLA/SLO adherence for critical services.
- Time-to-deploy model updates and rollback frequency.
Fast, Low-Risk Pilot Ideas
- Incident triage assistant: summarize alerts, correlate events, suggest next steps with links to runbooks.
- Config drift and intent checks: flag risky deviations before change windows close.
- Anomaly detection on east-west traffic to catch noisy neighbors and misconfigurations.
- Capacity forecasting for link and device utilization to reduce surprise upgrades.
Bottom Line
EMA's study confirms what many Ops teams feel: interest is high, outcomes lag. Treat AI in NetOps like any other critical system - clear goals, clean data, tight integrations, and ruthless validation. Do that, and "complete success" stops being a small club.
If you're formalizing skills and playbooks, consider our AI Automation certification. For monitoring strategy, the SRE Golden Signals remain a solid foundation for signal quality before adding AI on top.
Your membership also unlocks:
