One message came through clearly from today’s “AI/ML Adoption by SPs” panel at MPLSSD & AINETWORLD2022 in Paris, France: The value of AI/ML-powered network operations is clear, and it is time to get AI/ML out of the labs and into production.
Service Providers stated that AI/ML-powered network operations can:
- Detect anomalies not detected by traditional tools
- Reduce false positives for alarms
- Predict future defects
- Reduce mean time to detect and repair (MTTD & MTTR)
- Provide benefits for both infrastructure and SD-WAN use cases
Specific benefit examples stated included
- 30 million savings per year from initial use cases
- 50% of anomalies detected were not detected by existing tools
- Reduce level 1 to level 3 support time from 40 minutes to 5 minutes
- Predict cell radio battery failure
- Reduce truck rolls through anomaly prediction
- Paradigm reversing effect of MTTD going from a positive to a negative number
Service Providers emphasized that AI/ML-based prediction works, and it is important to get out of POCs and give AI/ML-based platforms access to production data and get started. There are organizational and privacy challenges in implementing holistic data analysis, however, do not wait for the perfect environment, get started with the data that is accessible, and achieve benefits today.
There continues to be a mix of Service Providers primarily pursuing “do it yourself (DIY)” approaches and/or partnering with vendors. Partnering with vendors can help better define the use case and AI/ML-based modeling requirements.
Both infrastructure and SD-WAN use cases are benefiting from AI/ML-powered operations. SD-WAN solution data and AI/ML analytics can be provided to benefit both managed service providers and SD-WAN customers, providing application experience insights, and tuning recommendations.
AI/ML is not an all-knowing, all-powerful approach. However, AI/ML is already delivering real benefits to Service Providers. The ability to analyze large quantities of data, in a rapid fashion, to detect existing anomalies and predict future ones, reduces the load on L1 through L3 support, reducing MTTR, and increasing network uptime and quality of experience. It is time to get AI/ML of the labs and into production.
