Transforming wholesale telecoms data into profitable decisions with AI

AI in telecoms: Opportunities and challenges

At ZIRA, our mission is to enrich our product portfolios and bring greater benefits to our customers. However, there are challenges to doing so.

With the huge growth of AI, it’s important to determine the most suitable field for its application. We look to strike the right balance of embracing next-generation technology without venturing into tech overkill. There’s also an ethical balance to maintain, ensuring AI implementations avoid any harm.

In the telecom sector, AI adoption has moved well beyond chatbots and network optimization. Today, leading operators apply AI across fraud detection, wholesale pricing optimization, interconnect margin protection, demand forecasting, and real-time decision automation. Yet much of AI’s untapped value remains locked in wholesale and commercial data. 

But what is the optimal approach when you look to successfully implement AI in wholesale telecoms?

Three approaches to AI in wholesale telecom:

1. The business approach

Many companies focus on the latest AI tools as a path to innovation. While adopting trending technologies can drive growth, it’s risky to invest based solely on popularity.

AI is not just one tool; it’s a spectrum that can be applied in countless different ways to address myriad challenges. Simply following trends and hype cycles—particularly around generative AI—often leads to inflated expectations and underwhelming outcomes. The real challenge is selecting AI approaches that directly support revenue growth, cost control, and decision velocity.

2. The research approach

A research-based approach carefully assesses the long-term practicality of using AI.

By researching how AI could be used, you can uncover what the right solution is. You might even realize that a certain challenge doesn’t need AI and, therefore, avoid technology overkill.

In time series forecasting, CSPs and MNOs already have the historical data needed to predict future trends, traditionally using models like ARIMA and Exponential Smoothing. While effective, AI can provide more precision by utilizing machine learning methods, such as Recurrent Neural Networks and Transformer, which handle longer data sequences for highly accurate forecasts.

Approaching AI forecasting from a business perspective, one might consider popular tools like Generative AI, but these models can fall short in accuracy compared to traditional methods and often demand higher investment. With this in mind, we leveraged the power of LLMs at the interpretation layer, using them to explain and contextualize already calculated results and present them in a clear, user-friendly way for end users.

If we instead take a research approach, rather than forcing things to fit, we can use our creativity to find how certain technologies and approaches can be applied to a situation. At ZIRA, a research-driven approach allowed us to fine-tune AI specifically for telecom needs. This involved using Generative AI as an action model to generate insights from forecasts, enhancing decision-making for our clients and providing clear, actionable results tailored to their industry.

3. The implementation approach

The cost of developing, training, and operating large-scale AI models is rising sharply, driven by compute, energy consumption, and specialized talent. This makes ROI-driven implementation and architectural efficiency critical for sustainable AI adoption. 

This keeps the message without the risky statistic.

The rising cost and complexity of AI infrastructure is reshaping how organizations think about scalable AI deployment.

Learn more

AI-enhanced time series forecasting

CSPs and MNOs have mountains of human-generated data that is full of value. But right now, most providers aren’t using it to the best of their ability. At ZIRA, we took this data and fed it into AI models. And we got great results.

By deploying closed, domain-specific AI models, CSPs and MNOs can retain full control over sensitive commercial and profitability data while still benefiting from advanced forecasting and AI-driven decision support. 

But this wasn’t enough; we wanted to go a step further. As organizations, certain data, such as profitability data, is sensitive and can’t be entered into public AI models. So, we created our own self-sustained closed AI model that our customers could use for time series forecasting without concerns over security.

We use a cluster approach, using multiple AI models to create forecasts. Why? Well, if you ask five random people a question, you’re not likely to get the same answer five times. Likewise, AI models won’t all produce the same results. By taking a cluster approach, we can use the variations in each AI model to create a more accurate result. Then, a Gen AI assistant summarizes the results and suggests the next steps. Our models are also constantly being trained to ensure they continue to deliver the most accurate results.

Conclusion

Our approach to AI in wholesale telecoms demonstrates how strategic, domain-specific AI can move beyond experimentation and deliver measurable commercial impact. By examining AI through business, research, and implementation lenses, ZIRA enables CSPs and MNOs to turn complex wholesale data into confident, profitable decisions in an increasingly competitive, data-driven market.