Intelligent bots as a smart solution?

A bou­quet of pos­si­bil­i­ties

“Always stay in touch!” – Nowa­days, this motto no longer only applies to the count­less stars, star­lets, influ­encers or local politi­cians who inform us almost con­tin­u­ously about news from their lives on var­i­ous chan­nels. It also applies to all service-​oriented com­pa­nies that place value on cus­tomer prox­im­ity and want to per­fect their cus­tomer jour­ney. The old motto “Click, buy and for­get” has long since become a thing of the past. On the one hand, because com­pa­nies always want to keep their cus­tomers up to date on new prod­ucts, cur­rent offers and pro­mo­tions. On the other hand, because exist­ing and poten­tial new cus­tomers also demand this com­mu­ni­ca­tion.

Over the past three decades, the range of com­mu­ni­ca­tion options avail­able for this pur­pose has expanded dra­mat­i­cally. Tra­di­tional options such as per­sonal con­tact, let­ters, tele­phone calls and faxes have been sup­ple­mented by e-​mails, chats or social media posts. Some­times the chan­nels are even tan­gled up in a wild, unpre­dictable mix.

But how do com­pa­nies react to the new require­ments in the area of cus­tomer com­mu­ni­ca­tion? Apart from the sim­ple offer of con­tact­ing via a spe­cial chan­nel, are there other tech­nolo­gies and meth­ods to make con­tact­ing and com­mu­ni­ca­tion as prof­itable as pos­si­ble for both the cus­tomer and the com­pany? For­tu­nately, this ques­tion can be answered with a resound­ing “Yes!” How­ever, it should be noted that each of the numer­ous com­mu­ni­ca­tion chan­nels must be con­sid­ered indi­vid­u­ally in order to be able to develop tailor-​made solu­tions for each chan­nel. Com­mu­ni­ca­tion chan­nels can basi­cally be divided into two classes: syn­chro­nous and asyn­chro­nous. Cus­tomers who make con­tact via syn­chro­nous chan­nels usu­ally expect an imme­di­ate response: calls should be answered imme­di­ately. For chat requests – via webchat or mes­sen­ger app on the smart­phone – a fast, at best imme­di­ate response is required. And, of course, cus­tomers also expect to be served quickly and sat­is­fac­to­rily in per­sonal con­tact, for exam­ple when vis­it­ing a branch office.

The asyn­chro­nous chan­nels include let­ters, faxes, but also e-​mails or social media posts. Of course, reac­tions are expected here as well, but they do not have to be imme­di­ate. Instead, a cer­tain delay seems not only to be tol­er­ated but usu­ally even expected. In the case of the clas­sic let­ter request, this wait­ing time is prob­a­bly the high­est (in some cases weeks may pass until the response is received). For e-​mails the accepted dura­tion is already much lower (rather days) and on some social media chan­nels an even faster response is expected (even hours).

Bots as helpers in syn­chro­nous cus­tomer dia­log

The tran­si­tion between syn­chro­nous and asyn­chro­nous chan­nels is par­tic­u­larly fluid on the thresh­old between chat and social media. In the fol­low­ing, the focus is on syn­chro­nous chan­nels. Here, com­pa­nies face the chal­lenge of meet­ing high cus­tomer and ser­vice demands. They must have suf­fi­cient com­pe­tent per­son­nel on hand to be able to respond to cus­tomer enquiries not only appro­pri­ately but also as quickly as pos­si­ble.

Many com­pa­nies meet this chal­lenge by pro­vid­ing their cus­tomers with an arti­fi­cially intel­li­gent con­tact per­son in cer­tain sit­u­a­tions. So-​called bots are able to make tele­phone calls as well as enter into a cus­tomer dia­log via chat.

Basi­cally, dia­log on both chan­nels – speech and text – is based on an almost iden­ti­cal prin­ci­ple. The only major dif­fer­ences are in the pre­pro­cess­ing of the input and in the for­mat­ting of the out­put: While the voice­bot deals with mum­bling callers or dialects in the input, the chat­bot has to be able to deal with spelling mis­takes and sym­bolic inputs (emo­jis). The voice­bot must again make sure that the out­put is very fast, because even the short­est pauses in a phone call are per­ceived as unpleas­ant. The chat­bot, on the other hand, should be able to sym­bol­i­cally sig­nal and trans­mit mes­sages, if this is desired.

Con­se­quently, each of these two vari­ants of human-​machine com­mu­ni­ca­tion has its own spe­cial fea­tures. This arti­cle focuses on the chat­bot and the oppor­tu­ni­ties and risks asso­ci­ated with its inte­gra­tion into the cus­tomer dia­log.

The art of con­duct­ing a con­ver­sa­tion

The first ques­tion is how com­mu­ni­ca­tion via chat works in gen­eral. Clas­si­cally, the con­ver­sa­tion here is ini­ti­ated by the cus­tomer, who sends a cor­re­spond­ing request to the com­pany. The con­tact is linked to the expec­ta­tion of receiv­ing a quick answer. Con­cep­tu­ally, at this point it does not mat­ter whether the respond­ing con­tact per­son is a human being or an automat. A free con­tact per­son will accept the chat request, greet the cus­tomer in the inter­est of the com­pany and inquire about their deliv­ery with a spe­cific ques­tion. Based on the customer’s answer, a dia­log will then be ini­ti­ated in which his request will be defined more pre­cisely and all gen­eral data for the ful­fil­ment of this request will be clar­i­fied.

The dialogs always fol­low the same steps. While the chat­bot reveals infor­ma­tion and asks fur­ther ques­tions, the cus­tomer pro­vides answers and – if nec­es­sary – makes selec­tions. In addi­tion, the chat­bot draws sup­ple­men­tary infor­ma­tion from back­end sys­tems and can make deci­sions based on this infor­ma­tion, which are impor­tant, some­times even ele­men­tary, for the dia­log. For exam­ple, if a cus­tomer inquires about the where­abouts of a long-​awaited order, the bot asks for the cor­re­spond­ing order num­ber and uses it to check the deliv­ery sta­tus. If a track­ing num­ber is also avail­able, it can also be trans­mit­ted, so that the cus­tomer can keep him­self informed about the fur­ther progress of his order.

How does the bot tick?

In order to find out which tech­nol­ogy enables chat­bots to con­duct intel­li­gent dia­log, it is essen­tial to look behind the facade. Here, too, there are basi­cally two vari­ants. Either the course of the entire dia­log is man­u­ally designed so that the chat­bot ulti­mately only has to process a dia­log script and fol­low a strict if-​then logic. Or the bot relies on spe­cial machine learn­ing algo­rithms to pro­vide appro­pri­ate answers based on the broad­est pos­si­ble infor­ma­tion base. In this case, the term Arti­fi­cial Intel­li­gence (AI) is often used, even if the key­word cer­tainly applies to both vari­ants of auto­mated dia­log guid­ance.

Of course, both types have their respec­tive advan­tages and dis­ad­van­tages. In the case of the scripted vari­ant, the oper­a­tors of the bot face the chal­lenge of antic­i­pat­ing all even­tu­al­i­ties in the dia­log processes and plan­ning ade­quate responses. The choice of words and the log­i­cal sequence of events should not be based on inter­nal com­pany prac­tices, but should be con­ceived from the per­spec­tive of a cus­tomer who lacks exten­sive knowl­edge of inter­nal processes and con­texts.

On the other hand, those who rely on self-​learning bots face a com­pletely dif­fer­ent chal­lenge. In this case, a suf­fi­ciently large amount of data must be pre­pared and processed from which the chat­bot can feed its knowl­edge. For exam­ple, exist­ing chat dialogs con­ducted by employ­ees could be col­lected and analysed. In the post-​processing phase, it would then be nec­es­sary to man­u­ally note whether the indi­vid­ual dia­log steps are to be clas­si­fied as cor­rect or incor­rect. In the next step, the algo­rithm can use the data gen­er­ated in this way to learn how it should behave in com­pa­ra­ble sit­u­a­tions in the future.

If such a chat­bot with auto­mat­i­cally learned behav­ior is then used in prac­tice, there is a chance that it will pull itself out of the swamp by its own hair – sim­i­lar to the famous Münch­hausen baron of lies. This means that the dialogs he con­ducts are saved and indi­vid­ual dia­log steps are eval­u­ated by employ­ees in order to increase the data­base and improve the reli­a­bil­ity of the bot. How­ever, once the chat­bot has insisted on always han­dling cer­tain sit­u­a­tions incor­rectly, the learned mis­be­hav­iour can­not be pre­vented so eas­ily – in con­trast to the scripted ver­sion. Instead, the data­base must be enriched with exam­ples of the desired behav­ior pat­tern until the bot changes its learned under­stand­ing of “right” and “wrong”.

Opti­miza­tion of cus­tomer com­mu­ni­ca­tion

But how can chat offers be used sen­si­bly in cus­tomer com­mu­ni­ca­tion? And how should their ser­vice be dosed? Regard­less of the cho­sen vari­ant, the ques­tion arises how to deal with mis­un­der­stand­ings and where the lim­its of the bot lie in com­mu­ni­ca­tion. So when is the point reached where the chat­bot can no longer solve a prob­lem on its own, but has to hand it over to a human col­league?

While bots don’t know when to stop work­ing and are gen­er­ally avail­able 24/​7, human ser­vice staff are often not avail­able around the clock. There­fore, it has to be decided whether the chat­bot is only used when the han­dover to a human col­league is pos­si­ble or whether the dia­log is aborted out­side ser­vice hours if nec­es­sary. In the lat­ter case, the bot could refer to the ser­vice hours or ini­ti­ate the arrange­ment of a per­sonal appoint­ment with an employee. The deci­sive fac­tor for this deci­sion is the extent to which the chat­bot can cover self-​services. If a rea­son­ably com­plete port­fo­lio of fre­quently used ser­vices can be achieved, round-​the-​clock oper­a­tion seems to make sense.

In order to avoid mis­un­der­stand­ings, ambi­gu­i­ties and ambi­gu­i­ties in the dia­log with the chat­bot, there is the option to refrain from enter­ing free text and instead pro­vide the cus­tomer with a lim­ited selec­tion of answers. In this way, lin­ear dia­log can also be used to limit the response options of the bot – and thus the effort required by the oper­a­tor. The selec­tion options can be dis­played in dif­fer­ent for­mats. In addi­tion, they can be enriched with graph­i­cal infor­ma­tion – for exam­ple, prod­uct images – to visu­al­ize the var­i­ous options.

In sum­mary, the automa­tion of the cus­tomer dia­log is an extremely com­plex field of appli­ca­tion in which it is nec­es­sary to analyse and weigh up in detail for what kind of requests a cer­tain approach can be prof­itably imple­mented and used. Bots are one of sev­eral ways in which com­pa­nies can effi­ciently han­dle one of the most dif­fi­cult tasks in the busi­ness envi­ron­ment, namely com­mu­ni­cat­ing with cus­tomers. After care­ful analy­sis and tak­ing into account the specifics described above, bots can become a deci­sive fac­tor in opti­miz­ing cus­tomer com­mu­ni­ca­tion. Their inte­gra­tion into the cus­tomer dia­log leads to sig­nif­i­cant sav­ings of two valu­able resources: time and money. While pre­vi­ously stressed employ­ees are relieved of rou­tine tasks and gain new capac­i­ties, cus­tomers are happy to receive assis­tance that is as fast as it is uncom­pli­cated.