Industry
Communications, Media and Information Services
Connect. Curate. Comprehend: Unlocking Value Beyond Boundaries
Solution
TCS HOBS™
TCS HOBS™: Engineering experiences with personalization and innovation
Highlights
- Communications service providers (CSP) grapple with challenges in network operations, such as accurate network KPI forecasting, manual resolution of customer queries, and ensuring a seamless experience amid the complexities of handling volumes of event data.
- The network is a source of humongous data for network behavior and end customer usage. However, it is currently only utilized for monitoring network availability and performance.
- Textual data housed on structured databases (primarily SQL-based), particularly from customer support and network operations, can be harnessed through natural language queries to enhance operational efficiency.
- Harnessing structured network operations’ data into large language models (LLMs) can play a pivotal role in fault resolution, proactively preventing critical network events, and contributing to network planning and building.
On this page
Abstract
CSP (Communications service providers) operations are undergoing digital transformation, leveraging new digital technologies and processes to modernize and optimize operations, enhance customer experiences, and reduce costs.
CSPs today are capturing data from the network but they are using it only to ensure their network is up and running. While this aspect is crucial and cannot be ignored, it is imperative to realize that the insights available in this data have the potential to achieve more than just network monitoring.
Leading service providers are already running programs to create a customer-centric, efficient, and innovative organization better equipped to meet the challenges of the digital age. Many of these service providers are at advanced stages, actively running programs focusing on customer experience, orchestration, automation, and enhancement of data analytics.
Large language models (LLMs) have created a buzz because of their ability to generate responses based on large amounts of data. Different industries and segments have already started working on specific use cases. Though CSPs are optimistic about LLMs, they are also evaluating relevant use cases.
If the entire network behavior and usage data is fed to LLM models, with or without customer data, the insights derived would be incredible. They will help optimize network, improve customer experience, and enable CSPs for immediate actions as well as long term planning.
LLMs have immediate use cases for a variety of applications, such as language translation, question answering, intelligent chatbots, and analytical content generation. In this paper, we explore a few CSP network operations use cases that can either be replaced or complemented by LLM-based tools.
Network operations challenges
Expanding the network capacity to meet a growing customer base is a daunting task for today’s telcos. CSPs must strike a balance, avoiding excessive expenditures or cost-cutting, as either extreme could potentially deplete cash flow or lead to customer churn due to lack of capacity.
One of the key concerns for CXOs is network planning and design. They have to invest in specialized tools to generate capacity requirements based on data fed to them from inventory and performance management tools. However, accurate network KPI forecasting remains a challenge due to the dynamic nature of the telecom landscape and evolving customer behavior.
Another challenge pertains to manually using the knowledge base. With many equipment, each requiring unique configurations, manual troubleshooting leads to delays in problem identification and resolution.
CXOs aspire to deliver the best service experience to their customers, aiming for no outages, meeting the highest service KPIs, and having minimal complaints. However, achieving this appears to be challenging due to the procurement and use of diverse equipment from multiple vendors. These heterogeneous networks throw millions of events, making it complex to monitor, track, and act on each event.
Can LLM help?
LLMs (Large language models) excel in processing and understanding textual data. Both customer support and network operations typically generate substantial amounts of data that are housed in an SQL database. If this data can be queried like natural language, it will help optimize the operations significantly.
The key challenge is to convert user inputs already in a textual format to SQL queries which can help with interfacing with the database. Extensive efforts are underway to integrate LLM-based engines with databases, aiming to achieve results akin to how LLMs are leveraged in the textual world. However, current outcomes fall short of expectations. With the advancements in LLM, a new thought process has started integrating LLMs with SQL to help build such use cases.
Large language models (LLMs) in network operations
We suggest the approach illustrated in Figure 1 to harness the information hidden in databases, which LLMs can use to generate insights for the network operations teams. This approach uses LLM to convert the normal textual queries from the user into SQL queries, and the output generated from the SQL queries is again fed into the LLM system to generate dashboards, trends, and insights. As opposed to traditional network operations systems that have predefined dashboards and report capabilities, LLM helps generate these reports on demand.
This approach is applicable across industries where data is stored on SQL databases. Some use cases in the next section explain this approach in greater detail.
Sample use cases
- Network plan and build: Fulfilment, inventory, and assurance systems generate data based on daily orders and operations. Traditionally, this data is fed on predefined rules on the dashboard and reports. However, leveraging LLMs can help generate real-time dynamic insights from these systems. For example, an inventory system can only give us resources available in a specific region. But LLM engines can analyze sales growth to predict resource requirements over a specific timeframe. Similarly, other parameters like the economic growth of that region and demographic information can be used by LLM to forecast resource requirements.
- Prevention of critical events: There are thousands of events that happen on a network. While some critical events are managed by a system or network operations team, delving into each event to identify root causes for future prevention is a formidable challenge. Processing each event manually or through rule-based systems is a mammoth task and involves huge costs. Currently, to create a balance between effort and RoI, only critical and high/medium severity events are considered for root cause analysis. Leveraging LLM presents an opportunity to process every network event, yielding insights for preventing higher-severity events in the future. This approach can help operations save significant effort and time by running proactive LLM-based engines in the background.
Fault resolution: When a fault occurs in the network, fault management systems process, analyze, and based on the severity, raise a ticket. The ticket management system manages the lifecycle of the fault and records the steps for resolving the issue. These resolution steps and records become a huge knowledge base that can be leveraged through an LLM engine to propose a resolution immediately. This helps close issues much earlier rather than going through the full cycle of diagnostics and resolution to improve MTTR (Mean time to resolve) significantly.
- Analyzing service provisioning logs: Network assurance takes the handover of service instances from the network fulfillment team to monitor and manage services. During the configuration of services, the network fulfillment team observes issues that are resolved through automation or manual interventions. A record of this is kept in activation systems or inventory. An LLM engine can leverage this additional information to analyze the holistic cause for any issue in the network to prevent any such issue in future customer onboarding. This use case can help improve customer onboarding time and experience.
The way forward
LLM as a technology area is no longer just a passing trend. With LLM, there is a prospect of analyzing through the entire network data coupled with customer data to provide valuable insights. There is good clarity on what kind of use cases are immediate candidates and which ones would be long-term. Most CSPs have either implemented or are already delivering LLM use cases for applications like content generation and intelligent chatbots. However, the next level of use cases, which we touched upon in the above section, would require some groundwork before implementation.
First, there is a need to unify the data models and data distributed across different toolsets. This would require building a metadata model for the whole network and this data model will have the mapping of relationships between different entities across network operations. After that unification, there is a need to identify the right tool and approach to convert textual queries into SQL queries and vice versa. This is an evolving area with various frameworks and tools around. Some good examples are LangChain and DataGPT. CSPs will have to do pilots with some of the tools to work out the best strategy to handle LLM uses.
CSPs need to be diligent and focused while working on the business case for these use cases. There are both short-term and long-term uses cases. Long-term LLM-based use cases would not only help in reduced operational expenditure (OpEx) but also expand revenue opportunities by offering focused products and services to existing and new customers, basis the insights provided by LLM.
