FAQ next gen: when AI boosts the user experience

If you have ever asked a question on an FAQ page, you know how crucial it is that you get the correct answer. With 800,000 visits a month to the help page on its website, Proximus has understood the importance of optimizing the relevance of searches. How have they done this? By introducing an interactive and intelligent FAQ.

Never mind a simple list of questions and answers, the system will soon understand the intention behind each query and will provide a summarized answer, with links to the relevant pages of the site from which the summary is taken. All of this is done with the help of generative artificial intelligence.

A call center under less pressure

"By improving the relevance of searches, we hope to have a direct positive impact on our contact center, " explains Sandro Manzo, Artificial Intelligence Lead at Proximus. This is because high-quality search results mean faster response times. "The result is fewer calls to our call center, which is freed up from a heavy workload." The call center in question can then free up resources to meet the specific and more complex needs of each customer. A considerable gain in efficiency!

Christine Cubat, Product & Solutions Analyst at Proximus

An AI as a French teacher

Before the use of AI, human intervention was essential when building FAQs. "We spent two whole days per month improving the search experience on Proximus.be," explains Christine Cubat, Product & Solutions Analyst at Proximus. "The main obstacle? Spelling mistakes and synonyms." Whereas platforms such as Google can process thousands of queries per second, Christine and her team had to enter each variation of the same word manually for the FAQ to recognize it. "Artificial intelligence means our system now quickly learns common mistakes, synonyms and anticipates customer intentions."

A question of semantics

"Our solution is based on a database that supports semantic searches. This technique is more than just keyword matching," Benoit Hespel, data specialist at Proximus ADA, explains. The FAQ now includes related terms, even if the exact word is not in the user's question. "For example, if the user has a question about their mobile phone, the program will retrieve the necessary information even if the database only uses the word 'phone' or 'mobile'."

FAQs to suit you

But even if AI optimizes the use of FAQs, creating them remains a complex process requiring real expertise. The creation of an FAQ thus involves several stages that Proximus has carefully followed:

1. Creating a database

"The way we insert data into the database varies according to its origin and format." Sometimes, the data is scattered across several platforms, or requires preliminary processing before it can be queried.

2. Prompt engineering

This stage is about the optimal instructions for the model. This could be, for example, helping a user find the best product for their needs. "This is where we also define the tone of the response (formal or tailored to the user) while providing some questions and answers to guide the model."

3. Tests and adjustments

Query results are refined through iterations. This involves readjusting the correspondence between questions and answers, and possibly modifying the data before processing it.