Accord­ing to SEO experts, com­pa­nies should use full sen­tences in their web­site texts and put head­lines in the form of ques­tions that poten­tial cus­tomers could ask and type into a search engine. This tip is not a sur­prise: intel­li­gent search engines under­stand ques­tions as well as com­plete sen­tences and deliver rel­e­vant results accord­ingly. Usu­ally, the search engine rec­og­nizes what the user wants to know as soon as the first or sec­ond word is typed. Today, com­put­ers can be pro­grammed to “pre­dict” human inten­tions, deci­sions and the asso­ci­ated struc­tures and to out­put pre­cisely tai­lored data. And what’s more, com­put­ers are capa­ble of learn­ing. They remem­ber pre­vi­ous search queries, deci­sions and pat­terns and can use them to opti­mize future results.

Bots, AI and human agents are impor­tant parts of the puz­zle

This trend becomes vis­i­ble in the chat­bots, which cur­rently seem omnipresent. On all kinds of web­sites we see small chat win­dows where avatars offer help. Behind these win­dows hide com­puter pro­grams called chat­bots. A chat­bot is a soft­ware that can have text-​based dia­logues with peo­ple. Using key­words, word com­bi­na­tions and syn­onyms, it per­forms rule-​based if-​then oper­a­tions. This means it deliv­ers answers or con­tent to spe­cific key­words or word com­bi­na­tions that have pre­vi­ously been put in a library.

If the end of a dia­logue path is reached, i.e. the chat­bot no longer knows what to do, the soft­ware for­wards the chat to a per­son. Or, the AI-​based soft­ware can per­form prob­a­bil­ity cal­cu­la­tions based on pre­de­fined clas­si­fi­ca­tions, tar­geted queries to the chat user or key­words and word com­bi­na­tions in order to iden­tify the user’s con­cern and pro­vide a suit­able answer. Here, too, the AI uses a pre­vi­ously defined data pool. And it learns: The valid­ity of its prob­a­bil­ity cal­cu­la­tions increases with the num­ber of cal­cu­la­tions.

Chat­bot, AI and human agents – these three puz­zle pieces in cus­tomer ser­vice are com­bined in the ser­vice chan­nel “Chat”. Any­one who has ever used a chat­bot knows that the tech­nol­ogy works sur­pris­ingly well. For users, it is usu­ally dif­fi­cult to tell whether their request is being han­dled by a human agent or a machine.

Chat in cus­tomer ser­vice is already main­stream

Rule-​based bots answer fre­quently asked ques­tions or ques­tions that are easy to grasp with rules. In retail, these are ques­tions like “Where can I find short blue trousers for men?”, because the cat­e­gories “short”, “blue”, “trousers” and “man” can sim­ply be read from the prod­uct cat­a­logue with the help of if-​then ques­tions. A sec­ond exam­ple from the retail indus­try is the ques­tion of the sta­tus of an order. Here, too, the nec­es­sary cat­e­gories (e.g. the cus­tomer num­ber) can be linked eas­ily with a query from the mer­chan­dise man­age­ment sys­tem and the cor­re­spond­ing result can be deliv­ered.
Insur­ers, on the other hand, can imple­ment com­pletely chatbot-​based dam­age reports, for exam­ple. All key data, includ­ing the insur­ance num­ber, the date, the amount of costs and even pic­tures, can be cap­tured just as eas­ily with the help of a text-​based dia­logue as by tele­phone.

For com­pa­nies, shift­ing sim­ple cus­tomer enquiries from expen­sive com­mu­ni­ca­tion chan­nels such as the tele­phone to a cheaper chan­nel (the bot) offers con­sid­er­able sav­ings poten­tial. If we assume, for exam­ple, in the case of insur­ance that an agent takes an aver­age of ten min­utes to record a dam­age report by tele­phone and that the price per minute for a query of this qual­ity is around 1.50 euros, the costs amount to around 15 euros per con­tact. The oppor­tu­nity costs for the pos­si­bly frus­trat­ing wait­ing time for the cus­tomer in the wait­ing loop are not yet included. The cost of work­ing time alone, which is required to record the claim in the sys­tem, can be reduced sig­nif­i­cantly with the help of bots.

Human agents and bots work closely together

Clas­si­fi­ca­tion, on the other hand, is used for con­cerns whose direc­tion can­not be esti­mated pre­cisely and in areas where the prod­uct range is chang­ing con­stantly. The iden­ti­fi­ca­tion of assets in insur­ance com­pa­nies is a typ­i­cal case. The cus­tomer must first be assigned to a spe­cific depart­ment, such as life, health or motor insur­ance. A com­pa­ra­ble appli­ca­tion case from the e-​commerce sec­tor would be the ques­tion “What are you look­ing for?” Here, the cus­tomer must first be assigned to a prod­uct group. If, for exam­ple, he is look­ing for a new vac­uum cleaner, it can be oper­ated with a cable or bat­tery, have a bag or not, be from a cer­tain man­u­fac­turer or the cus­tomer has con­crete price expec­ta­tions.

In both cases, the bot uses con­fi­dence queries to iden­tify the right request or prod­uct. The so-​called hybrid model, i.e. the com­bi­na­tion of man and machine, is often used here. If the soft­ware is not able to assign the request with­out any doubt, the chat is given to a per­son. The agent can then, depend­ing on the require­ments, freely con­tinue the chat or fall back on pre­con­fig­ured answers.
Even with the more com­plex clas­si­fi­ca­tion scheme, work­ing time is to be shifted to a more cost-​effective com­mu­ni­ca­tion chan­nel and cus­tomer com­mu­ni­ca­tion is to be made more effi­cient. The soft­ware finds short blue trousers for men some­times much faster than a human agent.

Chat­bots change the Con­tact Cen­ter

The inte­gra­tion of chat­bots offers not only eco­nomic advan­tages, but also plays into the hands of con­sumers and thus has a pos­i­tive effect on the company’s image. The goal is to stay attrac­tive for tar­get groups that are online affine and expe­ri­enced. Cus­tomers are look­ing for a sim­ple way to com­mu­ni­cate and get quick answers 24/​7. This is shown in the “2018 State of Chat­bots Report”, for which sev­eral com­pa­nies sur­veyed a rep­re­sen­ta­tive sam­ple of the US pop­u­la­tion. Accord­ing to this study, the points “round-​the-​clock ser­vice”, “get­ting an answer quickly”, “get­ting answers to sim­ple ques­tions” and “a sim­ple means of com­mu­ni­ca­tion” are at the top of the list of advan­tages of a chat­bot from the user’s point of view.

From a busi­ness per­spec­tive, cus­tomer care must become more cost-​effective and effi­cient. Cus­tomers want to use sim­pler and faster com­mu­ni­ca­tion chan­nels and be bet­ter served at the same time. We are at the begin­ning of a major devel­op­ment that will trans­form cus­tomer ser­vice and the Con­tact Cen­ter. It is pos­si­ble that Con­tact Cen­ters that route to human agents will be com­pletely replaced in the long run. Espe­cially regard­ing AI and machine learn­ing as well as the use of incom­ing data for analy­sis, all pos­si­bil­i­ties have not been fully explored yet.