From the idea to the research question

More and more companies are using artificial intelligence or large language models (LLMs) at various touchpoints of the customer journey to enhance the customer experience and retain customers in the long term. Always with the aim of increasing sales. However, if you look at Pangaro’s research (for example: Luigi’s Pizza: A Parable), the question quickly arises as to how sustainable and ethically or morally justifiable these new customer journeys actually are. This question led us to a completely new research idea. We wanted to find out how the use of LLMs in the customer journey can be ethically sustainable. During our initial research on this topic, we noticed the connection between the ethical use of LLMs and the overall success of LLM-supported campaigns, so we see our research not only as a meaningful contribution to the field of ethics research, but also as relevant to all research in the field of customer service, customer experience and customer centricity.

(The project was financed by an innovation check in cooperation with the Swiss Inisights Association ).

The use case

To work on our research idea, we needed a concrete use case. This use case should be technically easy to implement and at the same time be usable out-of-the-box for as many companies as possible. It was also important that the use case actually involved an exchange of dialogs between an LLM and a human. This time we didn’t opt for chatbots or voicebots, but for a newsletter campaign. However, classic chats would also have been conceivable.

The technical implementation

We used the AI Suite from BSI for the technical implementation. The BSI team and, above all, their technical solution are very flexible and can be easily adapted to different use cases with the help of the BSI AI experts. This enabled us to technically implement the first prototype within a few weeks.

Our expert interviews confirmed that we should definitely apply penetration theory in addition to the planned methods of conversational theory. This means that a customer or user only gains trust in the course of the relationship with the company and is therefore only willing to disclose more and more personal data over time. We used these results and trained our LLM to gradually ask for more details from the customer, but never all at once.

Specifically, we used BSI to generate a sequence of different newsletters in which the majority of the content was created by Generative AI.

While the first newsletter had no conversion goals in principle, but was merely intended to collect the customer’s initial preferences, this was followed by individually generated newsletter sequences with further conversion goals. The previous newsletter always served to find out more and deeper information about user preferences. These findings were then used to train the LLM and to further refine the newsletter content. The individualization of the newsletters was mostly done automatically by the LLM. The creation or definition of the conversion goals was created with regard to the sender of the newsletter, i.e. the company, which will ultimately use the newsletter sequence for itself.

Below are some pictures of our newsletters, as well as the graphic representation of the LLM sequences in the background.

The result

The users or newsletter recipients are addressed individually, which generally leads to a greater success of the newsletter campaign. However, according to the penetration theory, this individualization of the user approach only takes place gradually. Users should not be put off by obtaining too much information at once. Our LLM or newsletter campaign only asks for a few features per mailing, moving from unimportant features to important features.

In our example, a newsletter sequence was used. This is just one of many examples. It is also possible to transfer our results to chatbot or voicebot dialogs or to implement social media campaigns instead of newsletter campaigns.

How can other companies use this?

We are currently still in the prototype phase and would like to deepen our ideas and test them with other target groups, media and senders. We work independently of the industry. We would like to test how different senders, use cases or media affect the success of our LLM. So if you would like to take part, please just click on the button and let me know if you are interested.