Conversational Assistants: go to scale

by Patrick Meyer - Artificial Intelligence Senior Architect & Technical Director, Sopra Steria
| minutes read

74% of French companies consider chatbots as a lever for digital transformation and more than a third have already deployed one. By 2020, 80% of them could use a chat assistant. A massive deployment that echoes consumer habits: 69% prefer the bot to a human exchange. The main perceived advantage: saving time. This is a significant advantage in the face of the demand for real time imposed by markets that are increasingly under pressure. To meet this expectation, companies need to accelerate the automation of their business processes. But beware of the pitfalls when going to scale.

64% of companies in France make automation a priority. Similarly, 74% of entities familiar with Robotic Process Automation (RPA) have even already started a reflection or an integration project. Applied to IT, financial or HR departments, RPA would thus make it possible to reproduce manual processing on increasingly large volumes of data. However, these figures need to be put into perspective, as nearly one company in three (29%) still has less than 50% of its processes automated.


Business Automation - NLP: the genesis of the virtual agent

While the technology is not a recent concept, it is now mature enough to foster a high level of understanding of user needs. How can this be achieved? Thanks to Natural Language Processing (NLP). This makes it possible to interact with the machine and to create a new technology: the conversational assistant. At the key: the ability to run RPA scripts and to automate a growing number of questions, like ClaudIA, the virtual assistant for Chorus Pro users by the AIFE (Agence pour l'Informatique Financière de l'Etat). 

Launched in 2017, the chatbot covered all the topics encountered by users the following year, and proved capable of handling two-thirds of the requests. Target for 2020: to reach 90%. It seems a long time ago that the Swedish online bank Nordnet had symbolically "fired" its bot, Amelia, for incompetence... or more precisely its inability to meet customer expectations. The fault lies in a lack of learning. As a result, the bot quickly proved unable to answer the questions it was asked.

How to create the best conversational assistant?

This is where the complexity of a conversational assistant lies: poorly designed or insufficiently "educated" beforehand, it can quickly turn out to be counterproductive. This is why it is essential to model the dialogue, to anticipate the different questions and to teach it how to answer them.

In this sense, most solutions on the market are based on the concept of the decision tree. Thanks to the learning machine, the system detects the user's intention, the class, identifies the "entities" (i.e. the important words contained in the sentences) and generalises the answer to be given, either directly or by asking other questions aimed at obtaining more details. It is on this approach that the dialogue is built.

From then on, five types of wizards appear: e-commerce bots, support bots, information bots, form bots and search bots. What are the elements to be taken into account to create your virtual assistant? First step: find out what the most frequently asked questions are - via FAQs, support tickets for example. Once sorted and categorized, you need to automate 20% of them (20% of the questions covering on average 80% of the total volume of requests).

Then, it is necessary to model the answers: do I need additional information or can I answer directly? Does the sentence already correspond to a detected intention or is this a new area? Finally, is this question punctual or out of context? Whatever the request, it is important to follow a guiding principle: provide a short answer.


From experimentation to scaling up

Once these first two steps have been completed, you are ready to scale up. Nevertheless, be sure to anchor your system in a process of continuous improvement. Indeed, the system cannot live on its own and must constantly be optimized in order to enrich the dialogue and provide increasingly precise answers. The three golden rules of your conversational assistant?  Define a persona, name your bot and indicate its scope of intervention as soon as you connect.

A global vision that allows us to respond to the increasingly high challenges of proximity and customization that companies are facing, especially in times of crisis. In the years to come, virtual assistants will become the preferred companion of consumers. A "virtual clone" capable of knowing precisely the user's profile and context.

Technology alone cannot solve the adoption of virtual assistants. It is through use that they will spread and become indispensable. This is why companies must set up bots capable of communicating with all domains and industrialize the experience acquired. They must then decide upstream of their project to separate each system or to deploy a master bot that will act as a router capable, depending on the issue, of directing the user to the business robot concerned. This is a critical point that you must anticipate when scaling up in view of the diversity and volume of questions envisaged, and which will determine the success of your project... or its failure.

Discover our offer Sopra Steria Alive Intelligence



Related content

Related contents

How will artificial intelligence transform industry?

Whilst there is no set definition of artificial intelligence as of yet, experts are in agreement that AI can simulate human cognitive capabilities such as perception, reasoning, action, and learning. AI now promises to completely transform the industrial sector – one of its primary applications.

Related contents

HoloCare is one of the #EUvsVirus Hackathon winners with its solution for a digital doctor-patient relationship

Sopra Steria is proud to announce that HoloCare won the #EUvsVirus Hackathon efficient remote working challenge. Read more.