Generative AI for Continuous User Experience Improvement With Proactive Fine Tuning of Networks
Can you imagine managing networks for hundreds or thousands of customers where guaranteed performance and user experience mean everything? That’s our mission here at Nile: to deliver unified wired and wireless networks that are capable of supporting thousands of customer networks and hundreds of thousands of devices, as-a-Service, without overwhelming IT teams with a flood of alerts and events.
To meet this challenge, we’ve developed Nile Experience Intelligence (NXI), a generative AI-driven solution that analyzes network traffic and events across our managed customer environments, allowing Nile to proactively resolve issues and deliver a more reliable, high-performance experience. It’s what allows us to optimize user experience at scale and stand behind the industry’s only financially backed performance guarantee.
Because we’re managing such a large and growing environment with 1000s of variables, this poses the following challenges where Nile AI plays a critical role:
- Unscalable User Experience Monitoring: Inability to monitor, predict and proactively address user experience issues at scale
- Performance Challenges: The impossible task of guaranteeing uniform performance metrics everywhere
- Alert Fatigue: The time and resources required by network operations teams to analyze alerts
For current and future customers, this means always-on monitoring, proactive issue resolution, uninterrupted user connectivity, and seamless IoT operation—without the reactive guesswork typical of legacy solutions.
A Measurable and Predictable Network User Experience
Nile’s AI Automation stack collects telemetry data in the form of time series from the following network entities:
- Nile Zero Trust Fabric devices – APs, switches, sensors and headends
- events: RSSI, rate, latency, system events including errors, etc.
- All connected Wired and Wireless devices
- events: device behavior, wireless state transitions, errors, etc.
- Users and their access patterns
- events: Location, logins, authentication/authorization, etc.
- Physical environment characteristics – of the site, devices and users
Ex. events: interference, WiFi air quality, etc.
Fig. 1: Nile AI Automation Stack & Nile Services Cloud
Nile Services Cloud collects billions of events per day and processes 5TB+ of data generated by everything listed above. Nile AI models are trained on this extensive time-series dataset. The models predict the next sequence of events based on historical events. If the next sequence of events deviates from the predicted sequence, the deviation is considered an error. The error is aggregated across all events, resulting in the aggregated site experience.
A Reduction in Mean Time To Resolution (MTTR)s
Fig. 2: Mean Time To Resolution (MTTR)
Reducing MTTR in three steps:
- Reduced Mean Time To Detect (MTTD)
- Aggregated Site experience is continuously tracked by Nile Cloud Services. The models detect any deviation from the baseline site experience considering all the context and seasonalities and mark the deviation as anomalies.
- This continuous, proactive and automatic anomaly detection ensures that any network issues are captured even before they are reported by users.
- Reduced Mean Time To rootCause (MTTC)
- Models automatically identify the event i.e., root cause that contributed the most to the deviation in site experience.
- Models also identify all the “surrounding” events that correlate with the root cause event.
- This automatic correlation ensures that both the symptoms and root cause are captured as when the a deviation in site experience is observed
- Reduced Mean Time To Action (MTTA)
- Now that the anomaly and root causes are identified, Nile AI employs LLM models to identify the correct runbook to address the root cause.
- Nile AI then calls our Automation Center’s Softbots to execute the actions as mentioned in the runbook.
- Softbots capture the pre and post correction conditions to ensure that correction action did not cause any adverse user experience issues.
Nile AI Model versus Traditional Approaches
Here’s where Nile differs compared to rule based approach:
| Nile AI | Other Approaches |
| Generative Based | Static Rules Based |
| Standardized network design with devices from a single vendor enables a unified model for different networks across different customers | Non-standardized design with network devices from different vendors requires different rules for different networks across different sites |
| Built on machine learning infrastructure – model development, deployment & monitoring | Requires streaming analytics infrastructure that can process rules |
| Continuous learning capability enables models to automatically adapt to changes in network conditions | Lack of continuous learning means the technology can not adapt to changes in network conditions |
| No rules to maintain, automatically learns conditions on its own | Continuous maintenance of rules is required |
| Scalable with the volume of data (TBs of data) and variety of data (1000s of network events) | Not scalable |
| Guarantees performance because of closed loop issue resolution | Can not guarantee performance because of lack of closed loop issue resolution. Requires manual intervention |
The Benefits of Refined Optics
We believe that user related issues are brought about due to a lack of end-to-visibility in many competitive solutions today. Nile AI provides the following benefits for our customers, IT administrators, and end users.
- Guaranteed Performance: seamlessly integrates anomaly detection, root cause analysis and issue resolution, thus ensuring that any network issues that have even the remote potential of impacting user experience are automatically resolved without any involvement of end users or IT administrators. This enables Nile to deliver a 99.95% performance guarantee.
- Enhanced User Experience: provides detailed metrics and analyses about how devices interact with the network, offering a clear view of the user experience. This helps in understanding whether the connectivity issues users face are within the norm or indicative of a larger problem.
- Zero Network Alerts: proactive approach to analyzing and resolving network issues, eliminates the enormous amount of network alerts that are raised to IT administrators.
Conclusion
Nile’s AI models continuously track a site’s experience, automatically root cause of the events that contributed to anomalies, and work in close conjunction with AI Automation Center’s Softbots to automatically resolve those issues. This sets Nile on a path to deliver the highest level of network autonomy in the industry.
It’s not easy to provide service guarantees focused on end-users for a network that spans across millions of square ft spread across 1000s of customer sites is by no means an easy endeavor. Nor is identifying issues from 1B+ events that have the potential to adversely affect service guarantees is tantamount to finding a needle in a haystack. We believe Nile AI and standardized architecture is why Nile is the only vendor offering a financially backed performance guarantee.
I’d like to thank my colleagues Ebrahim Safavi and Ganesh Sundaram for their insights and contributions to this blog.
