The task was to create a virtual agent that understands the nuances of a particular branch in the business world, and have it respond appropriately to the inputer's requests. Koïos Intelligence already had promising expertise in this field, as they had designed conversational agents for the insurance sector. Today, these agents are even able to talk to customers over the phone and offer them quotes based on their needs.
"It's a bit like training human beings," explains Mohamed Hanini, CEO of Koïos. In elementary school, we learn simple syntax and conjugation. Then, as we go along, we add finer details, tone, and nuances. In the case of a conversational agent, we'll feed it with an additional layer of data that will enable it to understand a context such as insurance, for example."
Malheureusement, créer des assistants intelligents capables de ces nuances s’avérait fastidieux et coûteux. Comme les domaines des assurances et du marketing sont en constants changements, être capable de les mettre à jour facilement s’avérait essentiel. Les deux entreprises ont donc uni leurs forces pour faire essentiellement deux choses.
The first was to develop an "interpreter" tailor made for the world of marketing. This is, in a way, the engine of a virtual agent: the part that authorizes the passage between natural language and machine language. Wittycloud took on the task and produced a kind of "marketing bible" to feed the algorithms. "We relied mainly on content from our users, but also on our own expertise," explains Philippe Larose Cadieux.
On the other end, Koïos intelligence worked in R&D to optimize the pre-training and tuning phase of their deep networks to integrate Wittycloud's new layer of information economically and efficiently. In other words, they made qualifying and updating a voice assistant easier and cheaper.