从洞察到行动扩展AIOps：自主网络需要新技术和新工作方式

Autonomous networks require both new technologiesand new ways of working CONTENTS RCR Tech Takeaways Introduction4 The current state of AIOps — data rich, action poor Climbing the network maturity ladder one use caseat a time9 Building the technical architecture for system-levelthinking (and doing) Change management is part of the stack Conclusion Acknowledgements RCR Tech Takeaways The next phase of AIOps is about converting insight into governed, cross-domain operational action. That shift depends on CSPs have made real progress in observability, correlation and AI-assisted decision support, but most of that value still Operators do not reach higher levels of autonomy through wholesale transformation; they get there by selecting high-value Scaling AIOps requires a governed data layer, fit-for-purpose models, policy guardrails, a standardized execution platformand feedback loops that continuously refine performance. In practice, intelligence only creates value when it can be System-level AIOps scales when workflows, governance, trust models and decision rights are redesigned to let intelligenceparticipate in execution safely and repeatably. That means change management is a central condition that determines whether AI AIOps is moving from a visibility and analytics tool into an execution capability, and that changes what success looks like.The operators that win will be the ones that can turn AI-generated insight into governed, repeatable, cross-domain action by INTRODUCTION governanceand workflows needed toturn AI-driven insight into reliable actionacrossnetwork domains.Fragmenteddata pipelines, siloed operating models,brittle automations and limited execution Telecommunicationsnetworks arebecoming more software-defined,distributed complex.As communications serviceproviders (CSPs) push deeper into cloud-nativearchitectures,disaggregatednetworksandincreasinglydynamicserviceenvironments,the operationalburdenisrisingwhiletolerance major part but not themajority.” The next phase of AIOps maturity shouldbedefined by the move from pointintelligenceto system-level execution.That does not mean a sudden leap frommanualoperations to full autonomy.Itmeans climbing the maturity ladderinstages,moving from visibility andinsight toward assisted action, boundedautonomyand,eventually,coordinatedcross-domainorchestration.Achieving — Ahmed Abdelaziz, Vice Presidentof Automation and Transformation, That investment is producing real results.AIOpsis helping operators correlateeventsacross domains,reduce alarmnoise,supportrootcauseanalysisandidentify degradations before theybecome service-impacting failures. Newergenerative and agentic AI solutions are insight to action is as much a change-management challenge as it is a technicalone. The starting point for that discussionisthe current state of the market: Thatgap between insight and actionisnow the central issue.In manyorganizations, the limiting factor is not THE CURRENT STATEOF AIOPS — DATA RICH,ACTION POOR World in April, for instance, the debateamongoperators was not whether AImatters, but how fast and how far CSPsneedto move to rework their internalplatforms and operating models aroundit.The differences were mostly aboutpace and posture, with more radical AI- G AIOps right. The good news is they’re notenteringthis new era from a standingstart.Operatorshavespentyearsmodernizing their networks, virtualizingfunctions,shifting toward cloud-nativearchitectures and building the data andautomationlayers needed to operatemoredynamicenvironments.That “The uncomfortable truth is the averagetelecom operatorsits on an enormousvolume, variety andvelocity of data.” The problem Nar described is about bothquantityand dispersion.Operationaldataspans radio,core,OSS/BSS andotherenvironments where it sits indifferent formats, under different teams Atthe same time,the importanceofAIOps has moved well beyond an — Fatih Nar, Distinguished ChiefArchitect, CTO Office, Red Hat andprediction havenottranslatedcleanly into improvements in execution.AsRafael Ballesteros,head of businessandtechnology with TUPL,put it,“Repair speed is the ultimate measure ofoperationalsuccess.”By that standard, data-richin theory but operationallyconstrainedin practice because theinformationneeded to support reliable Even so, it would be wrong to frame thecurrentstate of AIOps as immature orunderdeveloped. Real value has alreadybeen created at the insight layer. AIOpsmethodologiesare helping operatorscorrelate events across domains, reducealarm noise, support root cause analysis insights that aredisplayed in beautifuldashboards, but Ballesteroscapturedthatimbalanceparticularlywell when he described a He then made the problem even clearer:“We’ve seen that AI…mostly sticks in theanalytics side and not on the operations.”Thisis the defining paradox of thecurrentAIOps moment.Operators are The market is also moving beyond theoryinselected domains.Qualcomm,forinstance,is positioning a