The two American groups advance their pawns on a whole new chessboard: that of the analysis of audience of the conversational agents. For everyone, the stakes are high.
For a little over a year, the market for bots is in full boiling. Intelligent agents to manage the reception on websites, chatbots of customer support, decision support in the bank, consultation of content in the media, voice skills … Projects are multiplying and ads are raining. Agencies have specialized in supporting businesses on this new field. This is the case in France Clustaar, The Chatbot Factory or The Social Client.
Publishers are also beginning to offer analytics tools to measure the performance of these applications, whether they are deployed in instant messengers or voice assistants. Position on this Google (with Chatbase) and Facebook (which provides via Facebook Analytics the possibility of monitoring bots on Messenger), not to mention some pure player in the forefront of which Bot Analytics.
Chatbase like Facebook Analytics delivers a whole host of indicators of bot analytics: evolution of the volume of active users and messages consulted, average duration of exchanges (over a given period, according to time slots), share of conversations included and not -comprises (or blocked) … “Whether at Facebook or Google, behavioral analysis of visitors remains limited,” says Nicolas Chollet, co-founder and head of product development at Clustaar. “It will be difficult or impossible to trace a specific event from end to end.In the case of the push of a carousel of news for example, we will certainly know the volume of Internet users who have seen, but not the share among them having clicked on one of the links put forward … These tools not being able to integrate correctly with the solutions of web analytics to make the link with the corresponding page views. ” On their side, the pure player of the field would not do better.
And Thomas Sabatier, CEO of The Chatbot Factory, to drive home the point: “These offerings are content to analyze the decision tree type bots.” By this mean conversational agents to “frozen scenarios”, guiding the user through pre-formatted questions (for example: “What category of clothing are you looking for?” “What size?” “What color?”). “This results in quite static analytics, and by definition unable to monitor chatbots with NLP-based intelligence (designed to capture the intricacies of an issue and continually enrich their knowledge base),” Thomas insists. Sabatier. “They will not be able to understand how the speakers express themselves, in a positive or negative way, to capture the recurring themes according to the periods, the answers most often provided by making the link with their satisfaction score, etc.”
A set of information that could help optimize the intelligence of the bot, its way of expressing and responding, including consistent with the “spirit” of a brand or service. “Beyond AI, these data are also central to streamline processes, such as being aware that the bot is used in the morning for shopping, and the afternoon for support requests. to adjust the resources at the level of the management of the orders and the center of contacts “, points Thomas Sabatier. And the expert adds: “To trace all these elements, bot analytics should be equipped with engines of linguistic comprehension and decryption of conversational UX, which they do not have.”
“The solutions of Facebook and Google have one thing in common: the absence of business indicators”