By Michael Knott

Generative AI (Gen-AI) is set to transform all industries, and that is not a superlative statement. Many may have predicted the Metaverse or Web 3.0 to lead the digital transformation, but it will be Gen-AI, simply because the use cases are more obvious, and impactful.

The new code in Telecoms

It would be logical to assume the benefits of Gen-AI to be somewhat marginal in Telecommunications as the industry has traditionally been an early adopter of emerging technologies. However, there are many impactful uses for Gen-AI across different functions of Telecom providers (Telcos) that span network infrastructure, back-office operations, marketing and product management, and customer service. The aim of this article is to simplify what is meant by Gen-AI and showcase its power through select use cases, deep diving into how it can revolutionize customer touchpoints and aid in preventing churn.

To demonstrate the capabilities of Gen-AI, we have included imagery related to our select use cases and concluded with a section authored by Gen-AI — a process that has been an interesting learning experience.

In this article, we explore how Telcos can integrate Gen-AI across their business, with a focus on enhancing customer touchpoints and preventing churn.

Charting the AI genome: AI 101

While Gen-AI may seem like a 2023 innovation, it has been around since the 1980s. Primitive applications existed in video games such as "Rogue" and "Elite," which used procedural generation to create random environments for players. And this ability to generate new data is the core of Gen-AI. For example, new data can include text, imagery, video, voice, and other synthetic data. The Gen-AI that has been propelled into the spotlight has only been made possible thanks to the technological advancements of the past two decades, which are important to cover.

The development of deep learning and machine learning (ML) models throughout the 2000s and 2010s enabled a move away from rigid, deterministic models. This evolution led to the development of Generative Adversarial Networks (GANs), which pits two models against each other: one aims to generate realistic data, while the other tries to distinguish real data from that which has been generated. It is akin to a jeweller distinguishing a genuine diamond from a lab-grown one. The adversarial strategies of GANs has produced data often indistinguishable from real-world data and made it the cornerstone in the evolution of Gen-AI.

While these AI innovations are remarkable, they have only been made possible by major developments in computing power, data availability, and the on-demand scalability of the cloud. To illustrate this, Open AI published analysis that shows computing power doubling every 3.4 months from 2012. This is equivalent to ~300,000x growth. To put this into perspective, Moore’s Law—a doubling every year—would only be 7x. In terms of data availability, when you look at the parameters these models have been trained on, you get the magnitude of development. For example, AlexNet (March 2012) was trained on 62 million parameters, Neural Machine Translation (2015) on ~210–380 million, meanwhile ChatGPT-3 (2022) has 175 billion, and ChatGPT-4 (2023) has 1.76 trillion. From AlexNet to ChatGPT-4 that is a ~28,000x increase that has enabled large language models (LLM) to be more accurate and informative. We believe that there are five key facets of Gen-AI that set it apart from traditional AI and ML models which are:

• 1. Simulation of complex behaviors: Gen-AI models can create new instances of data that closely resemble an existing training set, allowing models to capture more intricate patterns.
• 2. Novelty: They can generate entirely new, never-seen-before content through their large training sets, learning patterns and built in randomness.
• 3. Personalization: They can provide highly personalized content tailored to individual behavior, preferences, and linguistics.
• 4. Real-time adaption: They can modify output in real-time based on random inputs or evolving conditions.
• 5. Robustness to data scarcity: These models can provide synthetic data in the absence of real data for model training.

Beyond conventional AI: Gen-AI disruption in Telecoms

The Telecommunications sector is set for Gen-AI disruption across its operations and consumer relationships. The noted use cases are a small sample of the possibilities where Gen-AI can have a transformative impact, and as Gen-AI continues to develop further opportunities will emerge. We have selected four use cases, one from each of the typical Telco functions: predictive maintenance, legal drafting, customer segmentation for ad creation, and a combination of chat bots and churn modeling.

These use cases have been chosen as we believe they are examples of where Gen-AI could have the biggest impact, given its current capabilities. In our full article, for each use case, we offer an overview, Gen-AI’s visualisation, and then delve deeper into the use case with the highest impact on customer touch points—chat bots and churn modeling.

Unravelling the tangle: Decoding customer complaints and preventing churn

Telcos invest considerable time, money, and effort to lower churn, both proactively and reactively. Proactive solutions might involve building the network for better and faster network coverage or rewarding customers before dissatisfaction arises. In contrast, reactive strategies address issues once dissatisfaction has set in, and aim to swiftly and efficiently transform negative experiences into positive ones through best-in-class customer service.

When it comes to customer service, current chat bot technologies fail to effectively reduce wait times, and sometimes even contribute to churn. This is because—for the most part—they are rule-based systems (i.e., not generative) that try to match a user question to a pre-programed response. Even the more advanced chat bots which follow a multi-layer decision tree structure, struggle with queries outside their programed scope. These chat bots are unable to grasp context or linguistic nuances, and their rigid responses make them less effective in handling complex customer queries. Crucially, they do not scale as queries increase in complexity and they lack any ability to learn.

A new solution is needed, and Gen-AI has the potential to transform both the front- and back-end processes of churn prediction and customer complaint handling, offering significant advantages over existing chat bot technologies.

The final code: Interpreted and authored by Gen-AI

The Telecommunications sector is continually changing, and Gen-AI presents a significant opportunity for advancement. It is set to improve customer engagement by providing faster and more nuanced interactions. Gen-AI could also revolutionize churn prediction in Telecoms through its advanced data analytics and predictive capabilities. This could provide Telecom companies with deeper insights, allowing them to detect potential churn more accurately and implement strategies to increase customer loyalty. While this article focuses on churn prevention, Gen-AI can be applied to various aspects within the Telecommunication domain. It has the potential to improve operational efficiency and customer interaction, paving the way for significant advancements in the sector. Although Gen-AI is still emerging, it is expected to have a substantial impact on the trajectory of the Telecommunications sector. It could lead to increased efficiency, precision and innovation. This transformation brought about by AI is not just a theoretical idea—it is a tangible change that could redefine the industry.