By 2026, 1 in 4 interactions will be AI-enabled, according to Gartner predictions. As far as TP is concerned, that is already becoming reality, said Paul Joustra, SVP Transformation & Global Head of Innovation at Teleperformance. Together with keynote speaker Maarten Verschuere, he addressed the question of how generative AI will contribute to improving CX for a select audience.

 

The presentations to TP's guests were part of the TP event 'High Tech meets High Touch' which took place in an appropriate environment: the 17^e floor of Adam Tower in Amsterdam. A good view is essential, also when it comes to keeping up with and interpreting new technological developments. Both keynotes were accompanied by an excellent dinner and many table discussions.

Generative AI: increasingly specialized applications

After GPT2, several other LLMs have emerged from multiple parties, he said Joustra. In subsequent generations, increasingly larger datasets are used and the models are trained on more and more variables. Unlike chatbots, generative AI is pre-trained and is perfectly capable of creating content that can play a role in serving customers. For example, by contributing to knowledge management in which generative AI converts available information into customized content. This prevents a document being presented as a search result in which the employee still has to look for the answer himself.

A second example: providing customers with tailor-made advice, as is possible with the Klarna plugin (specialized in purchasing advice). The ecosystem of plugins around generative AI is growing fast anyway: more and more specialized functionalities are being added. According to Joustra, the chance that consumers will exchange a traditional search using Google for a targeted question to a specific AI solution is increasing.

A platform for applications

TP works, as Joustra previously stated in an interview to Ziptone, been working closely with OpenAI for some time now. One of the results is that a platform for generative AI has now been realized, which recently also accommodates other LLMs such as PaLM 2 from Google or on-premise LLMs for companies from the financial sector, for example. The platform is equipped with a 'finetune layer' that limits the chance of hallucinating. In addition, these types of applications will work with a 'human in the loop', but the employee will become more distant as systems improve.

On top of the finetune layer, TP currently has a handful of applications placed. The LLM is linked to external systems such as ERP or CRM, which makes it possible to realize context-dependent and personalized automated services or to support the agent in this.

Realized applications

TP now has several applications live. One is summarizing emails and generating a suggested response (in use since February for a car brand, resulting in, among other things, higher quality on outgoing communication. The second is supporting agents – by listening in real time and relevant knowledge, afterwards provide the employee with coaching feedback about the conversation, and summarize the conversation so that it can be written to the CRM system after a check by the employees.

Another application is StoryfAI, also in production: real-time multi-lingual translation of e-mail, messaging, chat or speech. The application can be integrated with your CRM solution. TP also sees opportunities in the field of real-time compliance checks: the agent can see in real time whether all mandatory parts of the conversation have been discussed.

Generative AI is 'the new bike'

Martin Verschuere, founder of Cleverland.io (a customer service chatbot platform with a strong focus on CX) and 'generative AI evangelist' was the second speaker. “In companies we always work with yesterday's 'shit'”, but that is not how you create future value,” says Verschuere, “you do that with solutions of the future.”

If you look at the distance that humans can travel on the basis of a standard meal, that is not substantially more than any other mammal. But equip people with a bicycle and the distance becomes a thousand times greater. In other words, after the internet and the smartphone, generative AI is man's next new bicycle, says Verschuere. The big question is what all this will bring within reach.

New competencies needed

Verschuere also emphasizes that generative AI is a platform where solutions can land, comparable to Google. Plugins provide specific application areas with specific data, with which you can then enter into the 'conversation'. Doing this well requires new competencies, just as we learned to use Word, Excel, Powerpoint and Outlook in the past. Anyone who learns to work with generative AI will have an incredible number of new possibilities at their fingertips, including scalable services that are precisely tailored.

Of course there are concerns that need to be addressed, such as fake news and privacy issues. However, Verschuere is positive about the upcoming AI act, which the EU is working on. At the same time, he says, it's too late to fear AI; tech companies are already engaged in a tech race.

Verschuere's message is therefore: watch out, but be ready. And Verschuere recommends starting with your organizational issue and not with the new applications that are rapidly becoming available. If you still want to explore what opportunities have arisen for each sector, you can consult a report by McKinsey, among others. But that report is likely due for an update soon. (Ziptone/Erik Bouwer)

Featured, Technology Teleperformer