• Pablo Fernández

With the new generation of bots, AI boosts customer relations


This article was originally published by Christine Monfort.


Chatbots are being massively adopted by brands and customers. These virtual assistants, which are based on more mature technologies, have become dynamic customer relationship supports that sort and respond to various requests, for the benefit of both customers and customer services.


"Hello, I'm here to answer your questions", "Can I help you?", "Tell me everything! These messages that punctuate the navigation in pop-up or via emoticons signal the presence of virtual assistants ready to dialogue with users. Between chatbots, voicebots, callbots, mailbots and other livebots, one could almost forget that this market has come a long way! At their beginning, chatbots were often poorly framed and not very useful. In fact, they were quite disappointing for the user and without any identified or measurable ROI for the company...


The increasing digitalization of customer relations, accentuated by the health crisis, combined with more and better adapted use cases, has revived the interest and usefulness of these conversational robots. They now rely on more mature technologies, based on artificial intelligence (AI), no-code tools and natural language analysis (NLP), a real game-changing technology for chatbot tools.


"People assume to talk to a robot. On the other hand, they expect it to answer precisely to their request", underlines Thomas Sabatier, founder and CEO of Tolk (ex-The Chatbot Factory). If you have to tell the user that he is talking to a chatbot, the bet is successful "when you don't feel that you are talking to a machine", adds Nadia Amal Larfaoui, DigitalCX manager for France at CM.com, a cloud software editor for conversational commerce. Organizations are increasingly supported in their efforts. "We no longer put a tool into production without having conducted full-scale tests, which are the best way to gather feedback and build a powerful tool," she continues.


Like all learning tools, chatbots need to become more proficient. "AI feeds as much on consumers' questions as on customer service answers. An autonomous learning ecosystem allows it to progress from real contexts and the brand's DNA, to provide a truly tailored and personalized response," notes Steve Desjarlais, CEO and co-founder of Heyday (acquired by Hoosuite this summer), a Canadian company specializing in customer experience that deploys conversational solutions.


The right alliance between the bot and the consultant

Trained to analyze freely formulated requests, these virtual assistants can go much further than first-generation chatbots built on decision trees that replicate FAQs and other interactive voice servers (IVR) in bot. "Chatbots today can understand long, complex or ambiguous questions. The conversation gradually removes ambiguities and sometimes shows the necessary adaptations, for example when we realize that the user does not understand the jargon used by the brand or organization," observes Nadia Amal Larfaoui. The chatbot's verbatims set up by CM.com for the Caisse Nationale d'Allocations Familiales (CNAF - National Family Allowance Fund) showed that, for some of the recipients, the notion of income used by the institution was not necessarily assimilated to that of salary.


What uses for bots?

- The messaging solutions and the KT bot, implemented by Heyday for Decathlon Singapore, reproduce the in-store experience and mobilize all sales staff. They have made it possible to handle 50% of incoming customer requests, with an "excellent" satisfaction rate of 74%. In January 2021, telephone and e-mail were abandoned in favour of chat.


- The bots developed by CM.com for the CNAF and CNAM made it possible to absorb the flows (more than 350,000 interactions per week) and to maintain the sites of the two institutions, even during the announcement of the reform of housing benefits or the news about the Covid vaccination (1.5 million interactions). Agents were able to focus on high value-added issues.


- Thanks to Tolk's bot, implemented in December 2020, the insurer Entoria has reduced its inbound calls by 56% and was able to manage the January peaks on updating third-party payment cards, schedules... The bot resolved 95% of the requests from nearly 15,000 users with a 94% satisfaction rate.


- At the end of 2020, Voxygen developed the synthetic voice of Bouygues Telecom's Santa Claus, used in a WhatsApp chatbot mixing deep learning, computer vision and speech-to-text. A CRM mailing offered customers to write a message on WhatsApp, read by Santa Claus in a video synchronizing his lips and the text. 2.964 million messages were created, three times more than the expected target.


These virtual assistants 2.0 are capable of sorting conversations, especially between those related to customer service or sales. They are particularly well suited to answering questions that require you to look for information on the merchant's website, in its databases, at a carrier's, on Google My Business records... "By bringing up contextual elements from the navigation or information already collected, the bot frames the theme and directs the customer to the right advisor. One will answer a request for advice before buying a perfume or a cosmetic product, another will know if the product is available or if it contains allergens," explains Thomas Sabatier.


Conversations related to sales rely more than any other on the alliance between man and machine, says Steve Desjarlais: "From the design stage, we have to build a kind of choreography that makes a complete loop, in which the AI that sorts the requests and ensures the first interactions, gives the answer or refers to an advisor, takes over again if the customer has other questions, asks for the level of satisfaction... And, if the latter is not satisfied, refers again to a human." The chatbot developed for Decathlon, a brand Heyday works with in eight countries (but not in France), understands if the customer is looking for an electric bike or a badminton racket, then directs them to the right expert in store for advice on the product. The customer can then proceed with their purchase online and the AI continues the conversation, for example regarding delivery. "At the moment of payment, if a chatbot detects that the customer is taking time to finalize his purchase, it can intervene to offer help and avoid a cart abandonment," says Nadia Amal Larfaoui.


Powerful and responsive tools

The chatbot has become an effective and very responsive customer relations tool: "We have reached levels of automation of over 80% and satisfaction rates of over 90%. It brings up within the hour a trend that allows to realize a problem of delivery or payment," says Thomas Sabatier. Before the summer, his company launched a new platform, Tolk Agents, which also makes it possible to manage the performance of customer service: "When a question is a little technical, instead of putting the call on hold, the advisor immediately sees if an answer has not already been given on the subject by a more expert agent. The response time is shorter for the customer and the agent's productivity can increase by 30%. Having access to this knowledge in real time enhances the work of agents and erases the effects of staff turnover," he says.


Voicebots are becoming increasingly popular. "Connected speakers have opened up a lot of uses. Businesses have seized on voice as a new channel for communicating with their customers and have decided to switch to more dynamic proposals than IVRs," explains Olivier Rosec, Chief Operating Officer of Voxygen. In fact, it was by observing the potential of connected speakers that BNP Paribas approached this voice synthesis specialist to improve the conversational experience of its voice assistant Telmi and HelloïZ, the Hello Bank! By giving each of them a voice - quite classic for Telmi and more cheerful for HelloïZ - the bank wanted to humanize the customer relationship and strengthen the commitment in the conversation. "When a voice carries the brand's values, its sound identity can become as strong as a logo," he says. The brand's voice can then be expressed on all broadcasting channels, as the SNCF understood with the voice of Simone Hérault synthesized by Voxygen.


"A voicebot is not just the voice transposition of a chatbot. You have to redesign the experience because, in this case, it's the user who decides if they want to interact. This implies proposing a good dialogue design, with a welcome sentence, the right information delivered with the right tone, a message if the person hasn't understood correctly...", develops Olivier Rosec. For the past three or four years, voice synthesis has no longer aggregated pieces of text to make sentences, but relies on neural networks to generate speech. The synthesized voice can also be synchronized in personalized videos for a shared experience.



New avenues to explore

Image recognition, on the other hand, offers an interesting potential to explore the wealth of information contained in a visual. "If a user spends a certain amount of time on a product, the AI will look for items that resemble him or that go well with him. This requires attaching the right data to the right product, which no brand does correctly," says Steve Desjarlais. In decoration, Castorama uses the solutions of the Israeli start-up Syte, which also has many clients in fashion.


Other evolutions are to be expected: "If virtual assistants allow to recreate online the experience of the store, brands will be able to launch stores dedicated to virtual sellers, which will become common in the future", adds the CEO of Heyday. A better connection with CRM, ERPs and business tools will advance transactional exchanges to declare a change in status or with a commercial purpose. "We are not yet at the transactional level. Cognition is also underway, to recover conversations and tend towards suggestions," notes Nadia Amal Larfaoui. On this point, AI is not yet at the level of the human brain!


1 view0 comments