Top 5 Practical Challenges & Considerations with AIOps

Author Tejo Prayaga
Published March 3, 2020

With rapid advancements in AI and Machine Learning technology and the widespread availability & affordability of high-performance compute (CPU/GPU), enterprise leaders are beginning to take big bets on AI, to succeed in their digital and cloud transformation initiatives. Application of AI for IT Operations (AIOps) has been a top initiative for many enterprise IT leaders who have already secured budgets and are actively evaluating multiple vendors by running proof-of-concept (POC)/pilot projects and planning to objectively evaluate vendors on several key business outcomes and improvements observed over current operational processes.

Most IT leaders are approaching AIOps with high expectations and have firm requirements and objectives set out for AIOps implementations, due to the broad applicability and integral nature of AIOps in the overall IT ecosystem. Following are the top 5 Enterprise AIOps challenges and considerations to watch out for:

Challenges	Opportunities & Considerations
1. Interoperability with existing tools and data Legacy tools tend not to be integration friendly. Tickets or incidents originated by service desk are left out. Data in AIOps is not accessible – almost like black hole Still dependent on underlying tools for deeper analysis	Comprehensive data ingestion – legacy & Modern Ingest incidents from service-desk as these incidents have direct impact on customer/stakeholder experience AIOps platform with in-built data lake for anytime data access
2. Service & Asset Interdependencies Dependency on clean CMDB, which is rare Extended time to value due to tools instrumentation, tagging, and scripts. Lot of resource overhead cost	Automate service and application dependency mapping (ADM) using self-learning and ML models Automate alert enrichment using context extraction and resolution AIOps platform to have built-in asset intelligence
3. Embracing Process & Culture Change Difficult to change the system of engagement. Hard to gain trust of AIOps decisions. Black box platforms are not flexible or customizable	Continuous collaboration and bi-directional integration with current system of engagement tools Open box ML models & fully customizable algorithmic behavior Platform should give ability to validate algorithmic decisions
4. Support for outcomes and use cases for multiple stakeholders Most tools are KPI or metric driven missing the outcome notion to drive results and sponsorship. Use case seem to be focusing only on ITOM or DevOps	Outcome driven – business & operational Use cases should cover across multiple IT stakeholders such as DevOps, ITOps, ITSM and IT Planning
5. On-premises option available for security/compliance Most tools are SaaS only or On-premise only, but not both. Some vendors retrofit legacy architectures for enterprise deployment. Some vendors over-provision infrastructure to address scale (increases capital expenditure and Opex) even though the deployment matures only over time.	Support for On-premise, Cloud and fully managed SaaS Modern application with cloud native architecture Uses Microservices/Containers architecture Distributed platform Horizontal and vertical scalability