Read out

Thomas Reutterer

Video Thomas Reutterer

Thomas Reutterer

Researcher of the Month

What’s the value of my cus­tomers, and how long will they stay?

Not all cus­tomers are equally valu­able to an or­gan­iz­a­tion. Build­ing cus­tomer re­la­tion­ships with a high long-term value is ex­tremely im­port­ant. In his re­search, Tho­mas Reut­terer from WU’s In­sti­tute for Ser­vice Mar­ket­ing and Tour­ism is look­ing at how we can meas­ure and pre­dict the ima­gin­ary long-term value of cus­tomers, a con­cept called cus­tomer life­time value. Pro­fessor Reut­terer’s work shows that the reg­u­lar­ity of cus­tomer in­ter­ac­tion is a cru­cial factor. If reg­u­lar­ity is taken into ac­count, it is possible to make much more ac­cur­ate pre­dic­tions of how much value a cus­tomer will bring to an en­ter­prise.

Cus­tomer life­time value, i.e. the total value of a cus­tomer’s busi­ness with a com­pany, does not only de­pend on how much time passes until a cus­tomer re­turns to buy an­other pro­duct or ser­vice. In more ac­cur­ate terms, cus­tomer life­time value refers to the amount of eco­nom­ic­ally rel­ev­ant in­ter­ac­tions a given cus­tomer is ex­pec­ted to have with a com­pany in the long term. How many pur­chases, ser­vice re­quests, and re­com­mend­a­tions will the cus­tomer make? How long will the re­la­tion­ship with that cus­tomer last? When is the cus­tomer likely to move on to an­other com­pany? WU Pro­fessor Tho­mas Reut­terer, head of the In­sti­tute for Ser­vice Mar­ket­ing and Tour­ism, is work­ing to answer all of these ques­tions, and to answer them as ac­cur­ately as possible. Reut­terer is work­ing with Mi­chael Platzer from Mostly AI and a team of data scient­ists to develop and test stat­ist­ical mod­els for pre­dict­ing cus­tomer life­time value. In a fur­ther step, these mod­els can also be used to identify rel­ev­ant factors that in­flu­ence cus­tomer life­time value. Pro­fessor Reut­terer’s re­search fo­cuses on non-­con­trac­tual cus­tomer­-firm re­la­tion­ships, where com­pan­ies can never be quite sure if a per­son is still an act­ive cus­tomer or has already moved away.

In cus­tomer in­ter­ac­tions, reg­u­lar­ity is key

Some cru­cial factors have already been es­tab­lished in mar­ket­ing re­search: re­cency (time passed since the last in­ter­ac­tion with the cus­tomer), fre­quency (the num­ber of pur­chases a cus­tomer has made), and mon­et­ary value (the amount of money a cus­tomer has spent on the trans­ac­tions). In ad­di­tion to these known factors, Tho­mas Reut­terer has iden­ti­fied an­other key di­men­sion: reg­u­lar­ity. “Our re­search clearly shows that if we ac­count for this small piece of ad­di­tional in­form­a­tion – the reg­u­lar­ity of in­ter­ac­tions between the cus­tomer and the com­pany – we are able to make much bet­ter pre­dic­tions of churn, i.e. the mo­ment when the com­pany loses the cus­tomer. This al­lows us to draw much more re­li­able con­clu­sions about the fu­ture value of a cus­tomer,” Pro­fessor Reut­terer ex­plains. His re­search res­ults have been used to develop an open source soft­ware tool­box for pre­dict­ing cus­tomer be­ha­vior. This soft­ware is designed to help com­pan­ies make bet­ter pre­dic­tions and im­prove their cus­tomer re­la­tion­ship man­age­ment activ­it­ies.

Step­ping up the game with ma­chine learn­ing and deep learn­ing

In ad­di­tion to data on reg­u­lar­ity, com­pan­ies today also have ac­cess to many other types of in­form­a­tion re­lated to cus­tomer in­ter­ac­tion pat­terns. Pro­fessor Reut­terer and his team are cur­rently work­ing to in­teg­rate these types of in­form­a­tion into their mod­els. “We can as­sume, for in­stance, that cus­tomers’ activ­it­ies on so­cial net­works or their pref­er­ences for cer­tain products or ser­vices can tell us something about their cus­tomer life­time value. And we can also gain valu­able in­form­a­tion by look­ing at how cus­tomers respon­ded to pre­vi­ous mar­ket­ing activ­it­ies. We are cur­rently work­ing hard to in­teg­rate fur­ther types of data into our mod­els for pre­dict­ing cus­tomer life­time value, us­ing our ex­per­i­ences so far as a basis. In our cur­rent work, we’re also us­ing ma­chine learn­ing and deep learn­ing meth­ods, which has proven to be a very prom­ising line of re­search,” Tho­mas Reut­terer ex­plains.