White Paper: AI-Native Requirements for Modern Networks

WHITE PAPER

AI-Native Requirements for

Modern Networks
By Bob Laliberte, Principal Analyst
Enterprise Strategy Group
January 2024

This Enterprise Strategy Group White Paper was commissioned by Juniper Networks
and is distributed under license from TechTarget, Inc.

© 2022 TechTarget, Inc. All

Juniper Business UseRights
Only Reserved.
 White Paper: AI-Native Requirements for Modern Networks

Contents
Modern Network Environments Are Complex ............................................................................................................... 3
Challenges to Implementing AI ...................................................................................................................................... 4
Technical Challenges ................................................................................................................................................. 4
Cultural Challenges .................................................................................................................................................... 4
Process Challenges ................................................................................................................................................... 6
5 Key Criteria to Drive AI Adoption ................................................................................................................................. 6
Juniper Networks AI-Native Networking Platform .......................................................................................................... 7
Organizations Need to Embrace AI ............................................................................................................................. 10

2
 White Paper: AI-native Requirements for Modern Networks

Modern Network Environments Are Complex

There have been multiple drivers in play that have led to the creation of modern network environments. Businesses
need to be more agile and responsive to rapidly changing market demands and are increasingly deploying cloud-
native application architectures in public clouds and private data centers. Plus, edge computing environments are
growing in order to gain real-time business insights, and hybrid work initiatives remain popular. As a result, the
network required to support these initiatives has become increasingly distributed and complex. In fact, research
from TechTarget’s Enterprise Strategy Group (ESG) shows that almost three-quarters of organizations (73%) stated
their network environment has become somewhat or much more complex than it was just two years ago (see
Figure 1).1

Figure 1. Network Complexity Over Time

In your opinion, which of the following statements best describes your current
view of the network environment? (Percent of respondents, N=374)

52%

21%
18%

9% 1%

Network environment Network environment Network environment Network environment Network environment
has become much has become somewhat is no more complex has become somewhat has become much less
more complex than it more complex than it than it was 2 years ago less complex than it complex than it was 2
was 2 years ago was 2 years ago was 2 years ago years ago

Source: Enterprise Strategy Group, a division of TechTarget, Inc.

Essentially, the modern network must support every connection across this environment. This would include
connecting multiple private data centers, all campus locations, multiple different public cloud providers, as well as
potentially hundreds or thousands of branch/edge locations and home offices. As a result, network operations
teams are inundated with network data and increasing alarm storms, resulting in alarm fatigue, delays in
remediating issues, and poor experiences. Unfortunately, these modern networks create so much data it is virtually
impossible to manually correlate and understand what is happening in real time.
This is where network teams need to leverage artificial intelligence (AI) and machine learning (ML) tied to
automation to drive operational efficiency, create better experiences, and shift from reactive to proactive or even
predictive management. It is important to note that this isn’t about AI/ML replacing the network team but about the
network operation team embracing the technology to be more effective. When built correctly, it will dramatically

1 Source: Enterprise Strategy Group Research Report, A Network Perspective on SASE and SD-WAN, November 2023.
 White Paper: AI-Native Requirements for Modern Networks

reduce, if not eliminate, alarm fatigue and accelerate the transition to proactive and predictive management of the
network environment.
The hype and interest around generative AI (GenAI) is certainly driving more attention to the use of AI, with
TechTarget website properties showing a 909% increase in search activity for the topic in 2023.2 While this can be
seen as a “rising tide” to help drive adoption, there is also a lot of confusion related to this topic and the large
language model (LLMs) being used to support it.
Similar to the rise of public cloud computing and the cloud-native environments (e.g., microservices architectures,
Kubernetes), we are now seeing the pioneers in the AI space for network operations deliver AI-native solutions. This
is an important distinction: AI for network operations should not be confused with the network infrastructure used to
support GenAI LLMs (see Ultra Ethernet Consortium3). The latter is about high-performance networking to enable
models to be built, while the former is about leveraging AIOps in the network space.

Challenges to Implementing AI
While the benefits of leveraging AI/ML tools are appealing, there are still a number of challenges that can arise in
creating the models and limiting adoption in customer environments. The challenges are grouped into three main
categories related to technical, cultural, and process considerations.

Technical Challenges
• Some of the most important parts of an AI model are the quality and quantity of the data. Abundant, high-
quality data is crucial for training AI models. Unfortunately for the network operations teams, the network
operations data has been siloed in customers’ on-premises management systems historically and cannot be
shared.
• Adding to the challenge of siloed network operations data is the fact that many network management solutions
have not been unified, so data center network data is separate from campus networks, creating more data
puddles of unshared data. Without access to end-to-end data, it will be difficult to isolate network events in
context to the domain in which they occur. As a result, network vendors might not have a large volume of
quality network data to use to create models, despite being in the business for a long time. Even more
important is that the vendors are focused on the right data. This goes beyond just “up” or “down” status and
being able to detect changes in experience.
• Many organizations are also very concerned about leveraging the benefits of public GenAI models without
exposing themselves to risk. While the open nature of the public model enables access and drives growth, it
might not be appropriate for sensitive or proprietary data, so organizations require solutions that don’t expose
their data.
• Lack of convergence of AI/ML and network automation is another obstacle. While network operations teams
have been leveraging network automation solutions for some time, the rise of AI/ML creates a technical
challenge of ensuring the technologies can work together across all the network domains. According to ESG
research, just over one out of 10 organizations (12%) have tied network AI solutions to their automation
capabilities.4

Cultural Challenges
• One of the biggest challenges with AI/ML technologies is a cultural hurdle. After becoming experts in
correlating network data and determining the root cause of problems over years, many network operations
team members are reluctant to trust these models. Much of this is ingrained into the conservative nature of

2 Source: TechTarget, 2024 Media Consumption Study: North America, December 2023.
3 Ultra Ethernet Consortium, 2023.
4 Source: Enterprise Strategy Group Research Report, End-to-end Networking Visibility and Management, April 2023.
 White Paper: AI-Native Requirements for Modern Networks

operations teams. When the company is relying on the network to conduct business, the appetite to deploy
innovative new technologies will always be met with healthy skepticism. As a result, operations teams need
time to become comfortable with the technology and validate that the AI/ML results are in line with their trusted,
albeit manual, methods and years of experience.
• If the AI/ML technology is not transparent about how it derived its results, that could further delay full
acceptance. Inherently, operations teams don’t want to trust the “magic happens here” approach. So, the lack
of an explainable AI solution (i.e., one that explains how AI arrived at an alert or recommendation) could delay
adoption and the time to become comfortable with the technology.
• Closely tied to the lack of an explainable AI is the lack of a closed-loop control system. Essentially, this makes
the network operations team part of the vendor’s AI team. Without a closed-loop system, experienced network
operators will not be able to provide feedback in a timely manner. This is problematic for the network
operations team, as not having this system in place would call for the creation of separate tickets that require
more time and effort to resolve. It would also slow down the ability to refine algorithms in a timely manner or,
conversely, the ability to confirm and validate that an algorithm is correct. Plus, this type of system is critical for
demonstrating operations have returned to normal, regardless of who fixed it.
• AI adoption is progressing slowly. Organizations are beginning to understand the value but want to stay in
control of remediation efforts. As ESG research shown in Figure 2 highlights, organizations want to leverage AI
for alerts and recommendations, but full reliance on AI to automate a fix is still limited.5 (Note: As highlighted
above, organizations are in the early stages of integrating AI/ML into automation. This creates a potential
intermediate step, where operations teams can manually engage automation to fix a known issue prior to
letting it be changed automatically.) Interestingly, organizations interviewed by ESG also cited a lack of AI/ML
maturity as the top challenge to using more network automation.6
Figure 2. AI Adoption Preferences

Which of the following statements aligns with your organization’s preference or

expectations in terms of leveraging network intelligence/automation capabilities?
(Percent of respondents, N=339)

48%

34%

18%

We get alerts, but prefer retaining full We expect the management software We want the management software to
control so that our staff analyzes and to provide alerts and serve as a automatically detect, analyze,
defines what should be done recommendation engine based on recommend, and apply changes as
learned behavior from automated real- needed, with the ability of the staff to
time and/or historical data analysis, review outcomes and make changes if
but staff should be responsible for required
implementation of the
recommendations

Source: Enterprise Strategy Group, a division of TechTarget, Inc.

5 Ibid.
6 Ibid.
 White Paper: AI-Native Requirements for Modern Networks

Process Challenges
• These highly distributed and complex network environments mean more frequent and larger alarm storms
when an error occurs. As a result, many operations teams or individuals are suffering from alarm fatigue,
resulting in the shutting off of alarms, which ultimately can impact experience and create a more reactive
environment.
• Many legacy network management systems require knowledge of CLI commands to initiate problem resolution
and identify the root cause of a problem. This requires vendor-specific training to become proficient and
knowledgeable for each vendor/domain.
• Skilled IT resources are in high demand, and the network domain is no different. Organizations struggle to find
skilled resources or lack the time and/or money to get them trained.
• Virtual assistants that leverage natural language processing (NLP) instead of CLI commands are in very limited
supply. While industries such as contact centers have been working on improving the technology for years, it is
still early for network operations to leverage these tools.
Organizations need to overcome these challenges and work with AI-native network vendors that can deliver real
value today.

5 Key Criteria to Drive AI Adoption

For organizations actively researching and investigating AI operations solutions for their network environments, it is
important to understand which criteria should be evaluated to determine the efficacy and value of a solution.
Here are five key criteria that should be considered when evaluating AI-native network solutions:
1. Cloud-based and cloud-native. While technically these are two criteria, their value is amplified when
leveraged together. The network management system should be cloud-based, not on premises. This enables
network vendors to collect all the necessary network data (both real and synthetic) across all network domains
and all customers (anonymously). It also provides vendors with a wealth of the right data, as opposed to an
LLM that scrapes the internet for data, thereby supplying network data that can be trusted and used to create
more accurate algorithms. Being cloud-native is important, as it shows the solution is not a lift-and-shift
operation from on premises to the cloud. Having a management solution that is cloud-native ensures rapid
innovation and scale, in addition to relieving operations teams of the burden of lifecycle management, security
patches, and bug fixes.
2. A platform approach to provide end-to-end context. This is also very important, as the value of the AI
solution will increase with each new technology domain added. Having a platform approach eliminates data
silos and puddles, ensuring that all the data collected from wired and wireless technology, software-defined
WANs (SD-WANs), switches, and routers deployed across the business environment (including data centers,
home offices, as well as all campus, branch, and cloud networks) can be used to get both end-to-end visibility
and, more importantly, context across these environments and beyond just the network (see criterion No. 5).
3. Virtual assistant with conversational interface. Given the shortage in skilled resources and the need to
focus on strategic initiatives, network operations teams should understand how virtual assistants with
conversational interfaces can provide benefits, including reducing the time and money required to learn
vendor-specific CLIs and the ability of virtually any team member to query information quickly, easily, and
accurately with efficient human language communication. In addition to assisting network operations team
members, these virtual assistants enable organizations to drive convergence with security, helping developers
to enable secure connectivity and accelerate development velocity. Plus, C-level executives can quickly get
status reports.
 White Paper: AI-Native Requirements for Modern Networks

4. Use of trusted data at the granular session layer or application layer. This also has two components, as
AI models require not only a significant volume of relevant data but also that the right questions are asked of
the data. The first part is empowered by cloud-based management and ensures that network AI solutions are
using data collected from real user network environments and across all domains of wired and wireless
environments. Because the vendor is using the telemetry data anonymously, there is no risk to the business.
Next, it is imperative that algorithms are focused on the right problems—for example, not just “up” and “down”
status but digging deeper into connection times and other factors that could affect a user’s experience. This
requires contextual data and, in some cases, might even require a combination of real and synthetic data
generated by a network vendor to ensure optimal configuration.

Another key factor to ensure AI generated results are valid is the ability for experienced users to provide
feedback. This would include incorporating a closed control loop to help refine algorithms based on feedback
from operations teams. Ideally this would be more than just a “yes” or “no” response and enable skilled
network members to describe the actual root cause of the problem as well as the verified fix. This is important,
as it shows that the network vendors want to work in partnership with the operations team and value their
feedback to improve the system. Ultimately, the faster an operations team becomes comfortable with the
recommendations the AI solution is generating, the faster it can transition to tightly integrated and automated
solutions.
5. A bidirectional API ecosystem. Having context across the entire network is important, but that value can be
amplified by collected data from adjacent domains, such as collaboration applications or even private 5G
networks. Both of these can be accomplished by creating APIs that allow for bidirectional data flows. This not
only could help to improve the experience when using voice and video apps but also be used to drive even
greater operational efficiencies by tying into workflow automation and security tools.

Juniper Networks AI-Native Networking Platform

Juniper is driven by its mantra of “experience-first networking,” and it has recognized the importance of AI and
automation in delivering the best possible experiences across an organization’s end-to-end network.
To accomplish this, Juniper ensures that any network is ideated, architected, and built to drive optimal outcomes by
leveraging AI. These are the foundational elements that comprise being AI-native.

Figure 3. Juniper Mist Cloud AI-Native Architecture

Source: Juniper Networks

 White Paper: AI-Native Requirements for Modern Networks

As shown in Figure 3, to deliver against this vision, Juniper has brought together several key cloud-native
innovations and technologies to market, including:
• The Mist-ification of Juniper. By making the strategic Mist acquisition several years ago, Juniper acquired
the talent and technology that would become the building block for a cloud-native, cloud-based AI
management platform. In addition to the Wi-Fi technology, Mist now has seven-plus years of experience with
AI engines and virtual network assistants (VNAs). Marvis, which is the product name for their VNA, provides
Juniper with NLP expertise and extensive network data to build and refine its AI models.
• A comprehensive, end-to-end, AI-Native Networking platform. The cloud-native, cloud-based platform
started with wireless access points (APs) and quickly incorporated Juniper wired switches, firewalls and SD-
WAN technology. It is now expanding to include data center networking, with plans to cover all routers in the
WAN environment. This platform has a number of benefits, including:
o The ability to collect trusted network data from each additional network domain. This provides
operations teams with additional context and value. Essentially, with each new domain added, the AI model
becomes more effective at isolating root causes and ultimately delivering even better experiences. Juniper
refers to this as the flywheel effect. Juniper’s AI engine covers data center, campus, branch, SD-WAN, and
(soon) WAN environments.
o A customer-focused culture. The Juniper Mist approach is to have the customer success team tightly
integrated with the data sciences team. This ensures that all major issues are pushed to the Marvis AI
platform and the data sciences team, ensuring they spend their energy and focus solving customers’ top
issues. This “experience-first” culture ensures the most important problems are always addressed in a timely
manner.
o The use of both real and synthetic network data collected from on-premises and cloud
environments. This includes data collected via routers, switches, firewalls, APs, and other sources. The
data is analyzed in a cloud-native platform capable of rapidly scaling to analyze massive amounts of data,
deriving real-time insights and ensuring enhanced experiences:
▪ Granular data collection goes beyond up/down network status, including application- or session-level
data to ensure users have the best experience for each interaction. This reflects Juniper’s focus on
quality of experience versus simply quality of service.
▪ Such granular data collection enables organizations to shift from reactive to proactive or, in some
cases, predictive management, finding and fixing faults before they create issues. For example, a
large public university in Massachusetts shared how its incumbent wired and wireless solution would
generate over 200 trouble tickets a semester, leaving the operations team in mostly reactive mode.
However, after deploying Juniper, that number dropped to only two tickets per semester.
Furthermore, it shared that the AI engine was finding and fixing issues that neither the students nor
staff were aware of, ensuring optimized experiences and a far more proactive approach to managing
the network. The staff learned of these changes when the closed control loop shared what it found
and the actions it had taken, as well as validating that it was running in an optimized state.
▪ Juniper is now expanding its comprehensive data collection capability by adding Marvis Minis.
Deployed virtually within its wireless APs, this technology simulates users or devices to test the Wi-Fi
environment in order to expedite problem resolution or ensure it is ready for business and delivering
an optimized experience. Leveraging the synthetic data from the Marvis Minis, organizations can
accelerate the adoption of a self-driving network that takes advantage of AI and automation. Juniper
Minis are effectively digital twins of users and devices that enable the system to understand the
impact of changes without affecting actual users.
 White Paper: AI-Native Requirements for Modern Networks

• Simplicity of operations. By incorporating the data center and eventually the WAN, Juniper will provide its
customers with a single portal and cloud-based AI platform to cover the end-to-end network environment. This
could dramatically simplify network operations, as it eliminates swivel-chair management of multiple different
portals and reduces errors from manual correlation. Plus, the AI engine provides rapid insights with context to
expedite problem resolution. In addition to the single portal, the Marvis Virtual Network Assistant has also been
extended to the data center, reducing the learning curve, and enabling operations teams to take advantage of
its conversation interface (see Figure 4). Organizations can now get an end-to-end view and ensure optimized
experiences from the application to the user. To further aid network operations teams, Juniper has incorporated
its technical documents and manuals into a GenAI model integrated into the portal to further assist users in
getting up to speed. This enables Juniper customers to use natural language to find out how to perform
specific tasks and provides a simplified experience for the operations teams, reducing the amount of time
required to become proficient and enabling them to focus on strategic initiatives.
Figure 4. Marvis Portal

Source: Juniper Networks

 White Paper: AI-Native Requirements for Modern Networks

Organizations Need to Embrace AI

Modern IT and application environments are driving more complex and distributed network environments. As a
result, operations teams can be quickly overwhelmed by alarms and remain stuck in reactive, firefighting mode. To
overcome this complexity, organizations require end-to-end visibility and contextual intelligence to transition to a
more proactive approach and drive greater operational efficiencies.
Juniper Networks has steadily improved its cloud-native, cloud-based platform to incorporate additional network
technologies and domains to drive greater context and efficacy for its AI engine. Today the platform covers data
center, campus, branch, and SD-WAN environments, with more planned. It encompasses wired and wireless
technologies and leverages both real and synthetic data. The virtual network assistant Marvis, AI engine, and ability
to leverage a single portal for end-to-end visibility will dramatically simplify operations and enable operations teams
to harness the capabilities of AI and network automation in order to become more proactive and predictive.
Juniper’s experience-first approach to technology acquisition and development has enabled it to deliver a cloud-
native, cloud-based platform that leverages AI-native capabilities in order to ensure that both operators and users
have optimized experiences. Juniper’s AI experience and expertise can help any organization accelerate the time to
adopt, validate, and recognize value using its comprehensive AI-Native Networking Platform. Organizations need to
embrace AI now to streamline network operations, deliver enhanced experiences, and add value to their business.

©TechTarget, Inc. or its subsidiaries. All rights reserved. TechTarget, and the TechTarget logo, are trademarks or registered trademarks of TechTarget, Inc. and are registered in
jurisdictions worldwide. Other product and service names and logos, including for BrightTALK, Xtelligent, and the Enterprise Strategy Group might be trademarks of TechTarget or its
subsidiaries. All other trademarks, logos and brand names are the property of their respective owners.

Information contained in this publication has been obtained by sources TechTarget considers to be reliable but is not warranted by TechTarget. This publication may contain opinions of
TechTarget, which are subject to change. This publication may include forecasts, projections, and other predictive statements that represent TechTarget’s assumptions and expectations
in light of currently available information. These forecasts are based on industry trends and involve variables and uncertainties. Consequently, TechTarget makes no warranty as to the
accuracy of specific forecasts, projections or predictive statements contained herein.

Any reproduction or redistribution of this publication, in whole or in part, whether in hard-copy format, electronically, or otherwise to persons not authorized to receive it, without the
express consent of TechTarget, is in violation of U.S. copyright law and will be subject to an action for civil damages and, if applicable, criminal prosecution. Should you have any
questions, please contact Client Relations at [email protected].

About Enterprise Strategy Group [email protected]

TechTarget’s Enterprise Strategy Group provides focused and actionable market intelligence, demand-side research, analyst advisory services,
GTM strategy guidance, solution validations, and custom content supporting enterprise technology buying and selling. www.esg-global.com
Juniper Business Use Only
3200161-001-EN