The Three Cor­ner­stones of a Smart Busi­ness

Rotman Management Magazine - - QUESTIONS FOR -

Adapt­able Prod­ucts.

Al­go­rithms can­not it­er­ate with­out the prod­ucts— the on­line con­sumer in­ter­face that de­liv­ers cus­tomer ex­pe­ri­ence di­rectly while gath­er­ing con­sumer feed­back to ad­just al­go­rithm mod­els. Google’s search bar is a clas­sic ex­am­ple of prod­uct de­sign. Cus­tomers en­ter key­words into a sim­ple bar and then im­me­di­ately see the prod­uct—the search re­sult page. Google puts tremen­dous re­sources into de­sign­ing this prod­uct so that cus­tomers can find what they want in less time and with less ef­fort.

Datafi­ca­tion and Live Data.

The start­ing point for the smart sys­tem that un­der­lies My­banks mi­croloans is cap­tur­ing busi­ness prob­lems in dig­i­tal form. How should off­line busi­ness ac­tiv­i­ties be ren­dered into data? Be­yond sim­ple op­er­a­tions and sales met­rics, which present num­bers that are eas­ily dig­i­tized, other valu­able ques­tions can be an­swered through the use of more in­di­rect data points. For ex­am­ple, to gauge how com­mit­ted and ac­tive a seller is, Alibaba’s ear­li­est mi­croloan prod­uct looked at how many sec­onds a seller takes to re­spond to cus­tomer queries and com­ments on prod­ucts and ser­vices. Dat­i­fi­ca­tion is not easy or cheap, but it is the most im­por­tant step in data in­tel­li­gence.

Ma­chine Learn­ing: Iter­at­ing Al­go­rithms.

Data can only cre­ate value when it is pro­cessed through a ma­chine­learn­ing en­gine. For a busi­ness, al­go­rithms must make ex­plicit the un­der­ly­ing prod­uct logic or mar­ket dy­nam­ics that the busi­ness is try­ing to op­ti­mize. Be­sides con­stantly re­fin­ing them­selves, al­go­rith­mic en­gines can also test var­i­ous sce­nar­ios to im­prove busi­ness per­for­mance. For ex­am­ple, My­bank’s data sci­en­tists embed an ex­per­i­ment into the lend­ing in­ter­face, giv­ing dif­fer­ent groups of sell­ers dif­fer­ent in­ter­est rates and mea­sur­ing real-time re­sponse rates. Each time the model gives a seller a par­tic­u­lar rate, the plat­form records the seller’s re­ac­tions in real time. This data in turn is used as in­put for the lend­ing model to ad­just its pa­ram­e­ters in a con­stant stream of cal­i­bra­tion.

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.