NOS

The Portuguese telco group NOS needed better insight into the pain points and root causes of its customer calls. The AI voice transcription and text summarization tool (Speech Analytics), put in place by SoftwareOne (formerly Crayon), automates what had been a cumbersome manual process for call auditing, yielding significant cost and time efficiencies. The SoftwareOne project also paves the way for NOS to elevate its customer experience and create better opportunities to sell products and services.

The Portuguese media group NOS Communications offers the latest generation fixed and mobile solutions for television, internet, voice, and data for all market segments.  It is a leader in pay TV, and with the fastest 5G network in Portugal, NOS is growing rapidly and recently posted Q2 revenues of €412 million. In 2022 the telco demonstrated the power of 5G with the first autonomous car journey in Portugal. NOS has more than 11m subscribers, and handles calls through an extremely busy contact center.

“We audited these calls manually,” says Joana Aguiar Santos, transformation project manager at NOS. “Not all of them – that would be physically impossible – but even summarizing a small sample took up a lot of resources and was very time-consuming. So two years ago, we started on the road of leveraging voice-to text-transcription and AI technology to do a lot more with our call data.

“At the time, the openness of NOS to commit to this kind of technology project was much lower than it is today,” Santos explains. “The first step, before asking for an investment, was to demonstrate that AI would be effective in summarizing voice-to-text transcriptions. We initiated a PoC with Microsoft who invited SoftwareOne to be a partner on it.

“We had a lot riding on this PoC because customer calls are one of the biggest sources we have to identify efficiency opportunities and also to tell us how and in what ways we can enhance the NOS customer experience.

“It’s about listening and finding out the main pain points of the customer, and why we are able – or not able – to solve their problems. Before the project, the only way we could approach this was to talk to the operations team, who had a general feeling about what people were telling them, but nothing detailed or specific.

“Then we listened to calls manually, which is a huge undertaking as I said. The calls are eight minutes long on average, and difficult to follow because people are not very clear, they talk over each other, the recording may be poor, and so on.

“It would take up about an hour to summarize a typical call, so it was impossible to monitor a lot of them. We were not getting a true picture,” says Santos, “and this imperfect process absorbed a lot of time and money. So becoming more efficient would have a huge impact on our team.”

The PoC was developed in the summer of 2023. “We had speech-to-text from Azure and AI comes in to summarize the conversation in two or three pages of text of two people speaking, giving us the main points of the conversation: what the agent did, if the problem was resolved, and if the customer was satisfied.”