Paul-Henri Chabrol AI program Director, AXA REV

Should we pull the plug on chatbots?

April 2016: Facebook Messenger announced it would open its instant messaging platform to chatbots. By 2017, 100,000 chatbots had already joined the network – providing yet another example of the tremendous enthusiasm for these conversational robots powered by artificial intelligence. But what value do they offer to users? To learn more, let’s take a closer look behind the bandwagon. Crafting tomorrow's insurance
Apr 17, 2019

When I hear my granddaughter ask Siri or Alexa to play a YouTube video – even though she cannot yet read, much less peruse the instructions – I realize we have entered a new era in our relationships with machines, which have become more immediate, intuitive and even natural. And for good reason, since the countless advances in artificial intelligence have now made their way into our living rooms, offices, and all of our daily activities, in the form of conversational modules known as chatbots.

25%
of customer service departments worldwide
will use chatbots by 2020. In 2017, only 7% did so.
61%
of businesses report an improvement in customer satisfaction
when they combine AI and human interaction in their customer service

Conversations that drive engaging relationships

Chatbots have made enormous strides in the space of just a few years. Initially based on simple decision trees with clickable buttons, chatbots can now understand and answer questions, direct users towards an appropriate service, submit an order, etc. All that is made possible by the natural language processing technologies perfected by big techs – Google, Facebook, Microsoft, Amazon and IBM primarily – and a rich ecosystem of startups.

Chatbots are now gaining a new appeal among companies, both in terms of boosting their operational efficiency and enhancing their customer experience. In fact, chatbots facilitate and enrich this experience by creating new interactions and new services. Available at a moment’s notice 24 hours a day, these chat representatives provide immediate and personalized answers. By storing your chat history, they will recognize you each time you visit and adapt their questions from previous conversations, thus establishing a continuous and attentive relationship with each user. This is the type of service we believe in at AXA, especially in light of the added value chatbots can offer when we strike the right balance of technology and human expertise.

Have we overestimated chatbots?

It’s no surprise that chatbots have met with such an enthusiastic response. But are chatbots really all they are cracked up to be? Is it possible that we have overestimated their potential? Already in 2016, some predicted that chatbots would replace apps. Today, that scenario does not seem any more likely to occur. The bandwagon effect should not obscure a trend’s limits or setbacks. For example, chatbots still have trouble grasping sentences containing multiple intentions (“I want to buy a plane ticket and reserve a table at a restaurant”) and also struggle to manage digressions. Everyone knows what it’s like to get stuck with a chatbot that cannot understand their request, and instead repeats the same irrelevant reply on loop. In these situations, what was meant to optimize the customer experience simply leads to frustration and a lot of wasted time.

Some epic fails have even gone viral. For example, Microsoft pulled the plug on its Tay bot just 8 hours after its launch, as a group of users put its learning technology to the test by teaching it to express racist ideas and hate speech. Then there is the virtual assistant used by the Human Resources Department at L’Oréal, named Vera, which was designed to preselect job applicants, and instead fired itself after only a few months of testing. In these cases, the teams clearly failed to strike the right balance between digital analysis and a human touch.

However, these limits and setbacks do have one silver lining: they offer crucial learning opportunities.

At AXA, we have identified three keys to developing a successful chatbot:

  1. Offer a clear service to users: In our view, each bot should respond to a specific use case so that users can understand exactly what it does (tracking claims, advice on a home policy, information about a health policy, etc.) – and what it doesn’t do! – in order to save them some disappointment and frustration.
  2. Provide adequate error management: If the bot does not understand the request, it should immediately transfer the conversation to a representative who might be able to help the customer. To offer a seamless and stress-free experience, we have expanded our teams to include new skills like “bot trainers”, whose job it is to improve machines based on use data and retrain them following errors.
  3. Never disconnect machines from people! This is clearly the most important point for us: a bot – no matter how smart or “learning” it is – should never be left without human oversight. In our view, the idea of machines replacing humans and handling interactions on their own should stay in the realm of science fiction. A chatbot’s value lies in its ability to enhance the experience by offering an additional service – not by emptying the relationship of its irreplaceable human touch. Chatbots can also perform certain repetitive tasks and thereby free up more time for representatives to process more advanced requests or interactions requiring greater empathy. A complementary relationship between chatbots and representatives is essential.
"A chatbot’s value lies in its ability to enhance the experience by offering an additional service – not by emptying the relationship of its irreplaceable human touch", says Paul-Henri Chabrol, AI Program Director at AXA REV

From enthusiasm to building skills

In terms of our concrete initiatives at AXA, we embarked on the chatbot adventure at the start of the innovation wave in 2016. A community of enthusiastic users has also developed around the world. Since then, we have compiled multidisciplinary teams to explore the topic, we are forming strategic partnerships to boost our skills – most notably with Microsoft – and we play an active role in the innovation community, especially through open-source projects.

For the past three years, we have followed the learning curve, first by offering fairly simple modules, then by expanding our approaches with natural language recognition. Imagine a customer at an optician who wants to know how much of the cost of their eyeglasses will be covered. This is now possible for our German customers, who can ask questions about their health policy directly through the chatbot on our www.axa.de website. No more waiting for a personalized reply from a representative. Since its launch in early 2019, 600 to 700 requests are processed every day.

These new opportunities also make it possible to develop unique services. This is the case for the newest addition to our family, Maxime, launched by AXA Legal Protection. Developed through an innovative participatory approach combining customers, employees and startup partners, it provides free answers to legal questions pertaining to housing on the website www.dailydroits.fr. Are your neighbors too noisy? Do you want to make sure you get your security deposit back? The chatbot will point you in the right direction, give you the essential information and even provide model letters. And if it doesn’t know the answer, it will connect you with a representative. Since January, Maxime has already started 5,000 conversations.

Welcome to maturity

At the moment, we at AXA are working on more than thirty chatbots worldwide – compared to just 5 or 6 projects launched in 2017. But we are also fully aware that we are only at the beginning of a movement. Including natural language generation, optimizing the use of voice, ultra-personalized responses and more, the technological advances to come are colossal. They will make it possible to create reliable assistants for every situation in life, whether you are at home, in the office, on the go, or in transit.

Finally, among all these marvelous prospects, I am convinced that technology alone will not provide the only key. We will need to draw even more deeply on the human sciences – and notably cognitive science – to continue to adapt our services more closely to individuals and their natural behaviors.

The bandwagon effect is fading away, and we are all the happier for it. Gone are the chatbots designed solely as marketing ploys. The future will bring us a much more mature approach to chatbots, while all the enthusiasm surrounding their promise will remain perfectly intact.