Four requirements of AI-driven automation for 5G mobile operators

Centralized management isn’t always the best business strategy, but it certainly is for telecom industry automation that leverages artificial intelligence (AI). Mobile network operators that choose a centrally coordinated approach to AI projects enjoy a 53% increase in their likelihood of success¹.

This makes sense: in the past AI was quite robotic and scripted, but it is now decisional and adaptive. Only through centralization can the usefulness of AI-driven automation be leveraged to make better decisions faster at lower cost.

Let’s examine four aspects of successful AI-driven automation, starting with the way monitoring is performed and the visibility this provides into network and service performance.

1 - Base AI-driven automation on a foundation of visibility

Just as AI has changed, so too have the network and service monitoring technologies that now underpin it. In the past, real-time monitoring and analytics were nice to have or even unnecessary. Now, say 91% of service providers², these are essential for automation to succeed. It’s not just a matter of knowing what to automate, but also verifying that automation is working as intended.

Over the next few years, networks will become statistically driven systems. The inputs to these systems are critical. That’s why monitoring is built into 5G from the start, rather than being an afterthought as often happened with previous generations of mobile networks.

But the inability of various monitoring systems to share data is becoming a big problem.

2 - Fill AI-driven automation visibility gaps with pebbles, not rocks

Right now, operators are living with significant visibility gaps in existing monitoring systems, and this is hindering their progress with AI-driven automation. Case in point: 65% of CSPs said they are held back by systems lacking ubiquitous, real-time data capture³.

While big data will continue to play a role, it’s no longer enough. What’s needed to fill the gaps in a complementary way is "small data”—streaming, real-time detection with machine learning built in. This processing happens upstream of big data systems, allowing AI-driven automation to use the data right away and act on it within seconds.

About 60% of CSPs don't have an established automation framework for service assurance⁴. One reason for this may be the difficulty of using existing networking telemetry, monitoring and analytics tools—which 91% of CSPs say are essential for network automation⁵—in a unified way. Since these were designed to be one-way systems to service a certain problem, it’s not surprising they lack features like APIs that enable automation.

But, integrating these systems is still a requirement.

3 - Use many sources with AI-driven automation to solve performance problems

Working from data silos is not only inefficient but also increasingly ineffective. What’s the root cause of streaming video degradation? Why are IMS issues recurring? Answering questions like these requires looking at a lot of different layers to figure out what’s going on. This is only feasible using systems that are automated together. And it requires systems with the ability to talk to each other and to orchestrators.

Misaligned or inaccurate data in existing inventory systems is a significant challenge for CSPs. They call this out in various ways. For example, 25% said their most important technical buying criteria when evaluating passive virtualized probing solutions is end-to-end visibility by integrating data from multiple sources⁶.

What tends to happen now is that each group within a CSP has their own data sources, and during troubleshooting they must attempt to manually correlate these sources. While this works to some extent in a traditional network setting, when it’s time for automation things fall apart.

Over half (56%) of CSPs are unable to share data between service assurance systems because these tools lack APIs⁷. In a similar vein, over half (59%) of mobile operators point to multiple, siloed data systems as a barrier to using AI and automation⁸. Contributing to this is the fact that 44% of CSPs lack a unified topology view of services, network and subscribers; they can’t see dependencies between layers, locations, users and infrastructure⁹.

Yet it still falls on CSPs to ensure that these systems integrate with each other somehow, using accurate, contextualized data—perhaps not immediately, but very soon. If they can’t, the closed-loop, AI-driven automation required for 5G will fail.

4 - Think globally, act centrally

Which returns us to the point that central management for AI-driven automation is the path to success. This should be approached first from a business standpoint, rather than an operational focus.

Many near-term automation projects within CSPs will be carried out as pilots or small-scale efforts. That said, they can still be centrally coordinated, with common goals in mind. A unified roadmap makes it possible for these projects to be funded and championed at a higher level in the organization, increasing the likelihood of success and driving more value overall.

This does a few things. First, it allows for an organization-wide focus on service-oriented architecture that flows through into automation decisions. Second, having a “chief automation officer” coordinating an overall vision ensures that RFIs and RFPs to vendors include requirements that benefit the entire organization. Plus, this drives vendors to develop capabilities needed for future automation.

Conclusion

Technology systems and the processes that leverage them may not be fully up to speed yet when it comes to supporting AI-driven automation. But they are evolving fast and CSPs are in a good position to take control and steer things in the right direction.

Watch the full podcast for a more in-depth look at the challenges CSPs face with automation integration projects.