Leveraging Generative AI in the Technology, Media and Telecom Sector

As organizations begin to explore the enormous potential of generative artificial intelligence (GenAI), they are encountering numerous pitfalls. The risks include excessive trust in GenAI output, intellectual property and data leakage, lock-in to the wrong large language model (LLM), copyright breaches, unstable systems, reputational damage and regulatory fines, if the technology is badly applied.

But the potential rewards could be huge: GenAI systems can take on a vast array of time-consuming tasks, such as writing code and legal documents and creating images and animations, while providing a source of inspiration for creative employees. The high-risk, high-reward nature of GenAI was one of the key themes to arise from a series of interviews Deloitte conducted with the senior executives leading the rollout of GenAI in KPN, NPO, NXP Semiconductors and the Talpa Group.

GenAI “will develop very rapidly and faster than we've seen in previous technology cycles,” predicts Egon Verharen, Manager of Innovation at public broadcasting organization NPO. “But we have to realize that we're just into the development stage and we still have to find lots of answers,” with some GenAI models literally changing week by week.

GenAI is very versatile

At the same time, competitive pressures mean most organizations can’t afford to simply wait and see what happens with GenAI. In the telecoms industry, KPN has already identified more than 60 use cases for this new technology. Winifred Andriessen, Director of Advanced Analytics at KPN, believes GenAI can help to provide the telco’s customers with the best services, while automating many more of its processes. Meanwhile, the Talpa Group, one of the Netherlands’ leading media organizations, is using GenAI to create images for quiz shows, write promotions and scripts, produce subtitles, compose social media posts and much more.

Talpa has found GenAI to be far more versatile than it anticipated. As well as simply accelerating administrative processes, such as monthly reporting, planning and control, recruitment and communications, GenAI “is also very creative, so can help you discuss creative concepts, marketing campaigns, media briefings,” says Joost Brakel, Managing Director of Al and Innovation at the Talpa Network.

For NXP Semiconductors, GenAI could help deliver a major boost to the productivity of its 11,000 software engineers, thereby accelerating “the cycle time of innovation,” notes Patrick Attallah, Chief Data Officer at NXP.

Integrating GenAI into an organization

But Attallah doesn’t regard GenAI as a silver bullet. The technology should be treated like a smart intern, in the sense that it should be given “bite sized” tasks and the outcome needs to be monitored carefully, he cautions, while also pointing out that employees need to learn to craft the right prompts to get the results they are after. “Prompting is not that easy,” he says. NXP is now building prompt engineering dictionaries/libraries for different types of engineers, so they don’t have to keep reinventing the wheel.

To balance risk and reward, NXP is taking a systematic approach to implementing GenAI with a focus on addressing clearly defined use cases through dedicated centers of excellence. At the same time, the company requires every AI tool and use case to be approved by a new internal advisory council and is using sand boxes to experiment with GenAI without impacting the company’s core operations.

KPN is also taking a use case-drive approach, focusing initially on the low hanging fruit. “For the first wave we only took use cases for which there were clear business owners, were not too complex and didn’t need too much IT interface and integration, as that takes too long,” explains Andriessen. These guiding principles helped KPN select 10 use cases that would enable the telco to explore different avenues in different parts of the business, such as customer service, legal, engineering and personal productivity, and to acquire the necessary skills.

Learning by doing

The Talpa Group is taking a different approach, allowing GenAI systems to permeate the organization from the bottom-up. GenAI “sells itself: It is so clear this adds value to any department,” Brakel explains. “We no longer encounter teams or departments or project organizations where they have not yet tested this at all.” Although most GenAI tools are immature, he contends that they are “getting exponentially better.”

Talpa runs hands-on training programs in which participants see whether their routine challenges and repetitive tasks can be offloaded to GenAI. “Al is a bit like mathematics or playing the piano. You don't learn it by watching the videos,” explains Brakel. “You learn it because you are doing it. Only then will you see what its power is.” At the same time, Talpa encourages employees to maintain a critical eye and be aware of GenAI systems’ potential bias.

Protecting privacy and intellectual property

Paying close attention to the data that LLMs are trained on is important for several reasons. One of these is to safeguard both sensitive data and individuals’ privacy. KPN’s acceptable usage policy for generative AI cautions against employees from sharing data into open generative AI applications. They are also required to use a watermark highlighting output completely created by generative AI. “Doing AI responsibly is completely who we are,” says Andriessen.

NXP is considering building its own specialist LLMs to protect its intellectual property and to ensure it has unique proposition, while also avoiding lock-in to a generic LLM run by a large cloud provider.

Addressing reputational risks

Another major risk with GenAI is that a model draws on copyrighted content or the distinctive output of an individual artist to train the GenAI models to create text, images, music and software code. In the media industry, where copyright is a key pillar of most business models, this is clearly a particularly sensitive issue.

To ensure copyright is respected, Verharen at NPO believes developers should pay artists a negotiated fee before using their work to train a GenAI model. He is also concerned about the implications of GenAI for storytelling, particularly for news bulletins and other factual programming, such as documentaries. “How far can we go before people start asking whether we still tell the right stories and whether they can believe the stories that we tell?” he asks, noting that GenAI could be widely used to create satire, for example.

For NPO, transparency will be crucial to maintain audience trust, Verharen stresses. NPO flags to its audience when AI (or any software tool) has been used to create or adjust any content, such as a photo or image. It also plans to puts “a provenance stamp” on its content so that the end user can check that nothing in the transmission or in the content of the messages has been manipulated by a third party.

In future, much of the content we consume will inevitably be created by GenAI. Already capable of generating realistic digital doubles of actors that can be used to enact fight scenes or action sequences, GenAI will likely be used to scan real world spaces and then recreate them digitally. “I think in 30 years you will no longer look at a photo, but just step back into that space,” predicts Maarten Knops, Director of AI at Talpa Studios. “It will be a digital photo book that you can really step into where you can be in the moment again, then you will see yourself as a young person. You see your child walking again.”

Navigating challenges and opportunities

Building on the insights gained from our series of interviews, it becomes evident that while these organizations are advancing their exploration of Generative AI, crucial areas require further attention to fully capitalize on its potential. Far from mere experimentation, these organizations are pioneering the exploration of vast value creation frontiers through Generative AI. The pressing question now is whether they can escalate these innovations effectively throughout their operations to truly harness the revolutionary potential of this technology.

Conversely, while these firms vigorously mitigate risks concerning reputation and regulatory compliance, critical oversight looms large. The focus on cultivating trust predominantly evolved around reputational and regulatory repercussions, noticeably disregarding the crucial facets of validating the quality and reliability of Generative AI outputs. Furthermore, the intrinsic trust in the technology itself remains under-examined, potentially creating a significant blind spot in their strategic implementation of Generative AI. This oversight could significantly impede their capacity to fully realize the extensive benefits that Generative AI offers.

Equally crucial yet underexplored is the impact of these technologies on the evolving workforce. It is anticipated that most talent strategies will need to pivot towards process redesign and the upskilling or reskilling of employees, potentially leading to a short-term increase in headcount. However, this aspect was either unnoticed or not discussed in the value cases presented by the companies interviewed, suggesting a potential gap in their strategic planning. This lack of attention to the workforce implications could pose yet another barrier to achieving the comprehensive advantages of integrating Generative AI within their business models.

At Deloitte we are working alongside clients to reveal how GenAI can reimagine business models, build value, and inform their vision now and in the future. Curious to see how we can work together with your company? Explore insights on GenAI adoption, wins, and challenges throughout 2024 on our website, or sign up for our GenAI Lab Programme.