Stefan TrockelonApr 25, 2019

“Next Level Dialogue Experience“ for Chatbots and AI Assistants

When we as humans engage in conversations with other humans, we do not use canned sentences or responses that we input into the conversation following pre-defined conversational patterns. To the contrary, we know the context and situation in which the conversation takes place, we have own goals, we understand the goals of the people we interact with, and we have knowledge about the people we talk to.

These features of human-to-human conversation are exactly those that the platform developed by tries to mimic by using cutting-edge AI methods. Using the platform companies can develop their own corporate conversational intelligence, allowing customers to interact with it as if it were a friend that is available around the clock. This creates a completely different customer experience and contributes to establishing long-term relationships with customers.

The technology can be used to improve customer support, for marketing campaigns, to improve brand value perception, or simply to inform about products or services offered in an interactive manner. The use cases that can be realized with such conversational assistants and with the platform are numerous.

The conversational bots that can be developed with the platform differ from those of other providers in their ability to carry on a relatively natural dialogue. This is achieved by the fact that the algorithms that drive bot behaviour decide for every input provided by the user on the best fitting dialogue move or behavior by taking into account the context and goals of the interaction but also knowledge about the user. Just like humans would do.

Concerning the contextualization of behavior, the bots remember what has been said in the conversation so far. This allows them to not simply react to the last input by a user, but to consider the whole conversational context to decide which next dialogue move or response fits best. A further question to narrow down the search for a product or service? A question for general preferences by a user? A first recommendation? Or rather a joke that resolves a stalled situation with a bit of humor? As a result, bots can handle vague or incomplete input such as „Give me another one“, which can be correctly interpreted in the context of the overall conversation.

Beyond this, the bots developed using the platform have a diverse repertoire comprising different dialog strategies which drive the behavior of the bots and are adapted to specific KPIs and goals. distinguishes four main types of dialogue strategies: generic dialog, search, information dialog and form filling dialogue. The latter one can save cumbersome filling of forms by users. The bot behavior is defined by these strategies and optimized to support the KPIs and goals associated with the bot. Also, several dialog strategies can be combined within one bot — a unique feature of the platform.

The bots by collect information about users but do this in a maximally compliant and transparent way, building on the explicit consent by users. A user can inquire at all times about the information that the bot has acquired about her, and even request the deletion of information. The bots in turn use this information to tailor their behavior to specific customers. This leads to a perception of an individualized interaction.

Contrary to, other provides of conversational interfaces define the content and sequence of dialogue moves a priori at design time of the bot. Bots designed in this way rigidly follow a given behavior pattern and have no way to deviate from this given pattern. This makes the dialogue with such bots brittle and unnatural. Some vendors make use of dialogue tree structures that support limited variation by making pre-defined choices along the tree. But even those providers that learn the tree structure in a data-driven way suffer from the fact that the dialogue remains pre-structured and feels less natural.

The platform developed by has been developed in terms of the above mentioned features to support a fluid, natural and contextually and individually tailored dialogue management approach that in contrast to other provides provides a „next level experience“ in the field of conversational interfaces.

Stefan Trockel