Telecommunication companies (telcos) have a significant opportunity to create new value from data using Generative Artificial Intelligence (GenAI). Telcos have access to massive customer datasets, which can be used to train GenAI models and drive impact across various functions such as customer operations, sales and marketing, and network security. However, to maximize GenAI's game-changing potential, it is crucial to improve organizational readiness, including the ability to tackle the challenges associated with GenAI.
GenAI creates new content, including images, and texts, by learning from training datasets. Its ability to simulate complex behaviors, generate novel and personalized content, adapt in real-time, and remain robust even in the face of data scarcity sets it apart from traditional AI and Machine Learning (ML) models.
GenAi is a unique and valuable technology addition for Communications Service Providers (CSP) along with their existing AI and ML models. According to a survey conducted by the global consulting firm Altman Solon, sponsored by AWS, 64% of CSPs have discovered numerous innovative applications that existing non-GenAI methods cannot address. As a result, the adoption of GenAI is increasing, with Precedence Research projecting that the global GenAI market for the telecom sector will reach US$ 4,883.78 million by 2032, at an estimated Compound Annual Growth Rate (CAGR) of 41.59% from 2023 to 2032.
Building foundational capabilities
Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) are two popular generative models that are used to create data that mimic real-world data. GANs are designed to produce synthetic data that is indistinguishable from real data, while a discriminator helps differentiate between real data and synthetic or realistic data. Over time, the GAN model can learn and produce convincing artificial data, such as realistic images, for various creative purposes.
The VAE model can be utilized first to compress data into a compact form and then extract highlights to generate new data. An important use case for the VAE model is anomaly detection.
Telcos are exploring these models to create foundational capabilities that go into building business applications.
Digital twins
According to McKinsey, digital twins are virtual representations of real-world products or systems that collect and model data. In the case of network infrastructure, digital twins can simulate the entire network for predictive adjustments and maintenance, optimizing energy usage. Telcos can also use digital twins to point out profitable resource allocation and refine security measures before actual implementation. China Telecom is currently exploring the implementation of a digital twin to visualize the effects of changes in factors such as customer activity and usage of network components on overall network performance before real-world application.
Synthetic data
GenAI enables the creation of artificial or simulated data, which is not drawn from the real world but generated by an algorithm. Synthetic data is widely used for predictive modeling and anomaly detection. It speeds up the testing process and enhances security against cyberattacks. As sensitive data can be risky to use in a test environment, synthetic data is a safer option. The use of synthetic data limits the risks associated with using sensitive data in a test environment and is gaining traction among telcos. Vodafone has recently launched a proof of concept to test how synthetic data can be used for training and testing its ML models.
Churn modeling
Reducing customer churn is crucial for telecommunication companies as they strive to address dissatisfaction in network performance, customer service, and product pricing, according to Roland Berger, a consultancy firm. The churn rate is the rate at which a business loses customers annually. During churn modeling, businesses often need help addressing poor data quality and interpretability issues. GenAI outperforms traditional AI in churn modeling with enhanced predictive accuracy, personalized recommendations, real-time adaptability, and automated insights generation.
Navigating GenAI implementation challenges
Telecommunications companies face numerous challenges in implementing Gen AI, including poor data quality, high operating costs, lack of compatibility with existing systems and workflows, regulatory compliance, ethical concerns, and customer distrust in the output.
It’s about data
If the input data provided to a model is of high quality, then the output generated by the model will be good. Quality, in this context, refers to the accuracy, completeness, and consistency of not just structured data but also unstructured data. Unstructured data is the fuel that powers Gen AI engines, and hence, it brings into question the data management practices of an organization. To improve the data quality, the organization needs to capture, cleanse, store, label, classify, and bring explainability to the data.
Though it sounds surprising, GenAI offers solutions to the problems it creates. For example, consider the case of a telecommunications company that wants to label sensitive data as protected by privacy laws. With both structured and unstructured data being utilized by Large Language Models (LLM) that powers GenAI, the company needs to ensure that sensitive data is not inadvertently exposed to third-party-controlled LLMs. However, incorporating GenAI in the company’s data management and governance processes can augment and automate these tasks and enhance the efficacy of the processes.
Compatibility with existing systems
Many telcos have legacy systems that can make integration difficult and costly. AI tools that integrate well into a workflow help performance and prevent teams from resisting adoption. To successfully implement these, telcos should carefully select tools that meet their criteria for factors like speed, security, and scalability. They should also train their teams on how to optimize the tool for best results and start with small pilot projects to test and measure progress. Finally, they should fine-tune and scale these tools to maximize their benefits.
Deutsche Telekom's Ask Magenta chatbot is an excellent example of an AI-powered tool that leverages an LLM to provide more accurate responses. The chatbot uses a decision tree for 80% of its functionality and the LLM for the remaining 20%.
Dealing with the cost question
One mistake that leaders make when it comes to GenAI is to think of it as an add-on to an existing capability. This approach can lead to miscalculating the costs involved in deploying and fine-tuning the model on a sustained basis. There are different costs associated with this, such as inferencing costs for each instance when an LLM is used to generate a response, fine-tuning a pre-trained model with organization-specific data for contextualization, infrastructure costs, including cloud, and the cost of talent required to run the operations.
Boston Consulting Group recommends taking a strategic approach to pricing and choosing which of the three models—subscription-based, consumption-based, or outcome-based, best suits an organization’s business case.
Strengthening customer trust
In an EY research, 68% of telco respondents said they needed to be adequately managing the unintended consequences of AI. According to Gartner's ‘Top Three CSP Market and Technology Trends from DTW 2023’ report, trust, risk, and security management (TRiSM) are often neglected, increasing risk exposure. To address this, Gartner recommends that the C-suite establish governance procedures via a steering committee tasked with ensuring Trust, Risk, Security, Management TRiSM when exploring and operationalizing Gen AI technologies. Forrester advises that besides focusing on tackling data leakage, data lineage and observability, and privacy concerns, businesses must showcase their privacy and security measures with customer-friendly messaging. Additionally, they should educate employees and customers on privacy and security practices, such as the zero trust approach, to boost user confidence in Gen AI.
Managing regulatory concerns
AI is under increased regulatory scrutiny worldwide, which will impact telcos. The proposed EU AI Act suggests fines for violating guidelines on responsible and ethical AI, potentially reaching up to 6% of annual revenue for non-compliance. Telcos need to protect sensitive data to avoid non-compliance, using anonymization techniques like just-in-time anonymization or obfuscation. Bain & Company highlights that doing this well and quickly can build trust among consumers, employees, and investors, making ethical AI a competitive edge for telcos.
The road ahead
GenAI is set to open new possibilities for telcos and enable them to differentiate themselves in a highly competitive market. At the same time, a measured approach that considers the unique challenges that telcos face and creates a roadmap for sustained gains is needed. Technology partners with deep expertise in AI and industry knowledge can extend the support telcos need to succeed in their GenAI journey.
|
Panchalee ThakurIndependent Consultant |
