What’s the big deal about ma­chine learn­ing?

Accounting Today - - Technology - By Tom Shea

Ma­chine learn­ing has emerged as one of the big­gest and most promis­ing tech­nol­ogy trends to­day, poised to greatly trans­form our lives and how we do busi­ness.

We are al­ready see­ing early ap­pli­ca­tions that pro­vide smarter cus­tomer rec­om­men­da­tions for on­line ser­vices such as Net­flix or Ama­zon, the rise of vir­tual as­sis­tants like Siri, Alexa and Cor­tana, and im­prove­ments in health care such as bet­ter im­age scan­ning to de­tect can­cer. Ma­chine learn­ing is the driv­ing force be­hind au­tonomous ve­hi­cles, too.

A branch of ar­ti­fi­cial in­tel­li­gence, ma­chine learn­ing is essen­tially soft­ware that can make de­ci­sions based on ex­pe­ri­ence and with­out the need for tra­di­tional rules-based pro­gram­ming. It uses sta­tis­ti­cal al­go­rithms to learn and get smarter over time, re­train­ing it­self the more it “ex­pe­ri­ences.”

The con­ver­gence of three other tech­nol­ogy trends is push­ing ma­chine learn­ing from con­cept to com­mer­cially vi­able ap­pli­ca­tions that are au­tonomously im­prov­ing busi­ness:

We fi­nally have the raw com­put­ing power to en­able the vol­ume of trans­ac­tions or “ex­pe­ri­ences” re­quired for ma­chine learn­ing.

Mas­sive amounts of in­for­ma­tion — big data, if you will — in­clud­ing shared ex­ter­nal data from in­ter­net-of-things ap­pli­ca­tions gives ma­chine learn­ing the broad con­text it needs.

Ad­vances in com­plex neu­ral net­works, or deep learn­ing mod­els that can an­a­lyze enor­mous quan­ti­ties of data points, are

With the abil­ity to process un­struc­tured data, ma­chine learn­ing can dis­cover pat­terns and cor­re­la­tions that were pre­vi­ously un­de­tectable.

cre­at­ing ef­fi­cien­cies be­yond what lin­ear mod­els can pro­vide.

The abil­ity to go be­yond de­tect­ing pat­terns to proac­tively adopt­ing and op­ti­miz­ing a so­lu­tion is what makes ma­chine learn­ing valu­able.

Fi­nan­cial in­sti­tu­tions have used pre­dic­tive an­a­lyt­ics for more than 20 years to com­bat fraud. Where pre­dic­tive an­a­lyt­ics soft­ware helps de­tect trou­bling new trends that could be fraud­u­lent, a ma­chine learn­ing ap­pli­ca­tion can act to pre­vent po­ten­tial fraud. Plus, it con­tin­u­ously im­proves, get­ting ever smarter at preven­tion.

Ma­chine learn­ing is es­pe­cially ef­fec­tive for ap­pli­ca­tions that process vast amounts of data, par­tic­u­larly where tra­di­tional lin­ear mod­els can be lim­it­ing. With the abil­ity to process un­struc­tured data, ma­chine learn­ing can dis­cover pat­terns and cor­re­la­tions that were pre­vi­ously un­de­tectable.

Be­yond fraud preven­tion and other ob­vi­ous uses, ma­chine learn­ing is also au­tomat­ing tasks that can be te­dious, al­low­ing IT staff to fo­cus on more strate­gic projects.

Early suc­cesses have been largely in the con­sumer space, but we are now see­ing busi­ness ap­pli­ca­tions for ma­chine learn­ing, with great po­ten­tial across mul­ti­ple busi­ness func­tions:

For the sales team, ma­chine learn­ing will de­liver more in­sight from cus­tomer re­la­tion­ship man­age­ment sys­tems, as well as a deeper un­der­stand­ing of cus­tomer churn and buy­ing trends to ul­ti­mately im­prove cus­tomer ser­vice and shorten sales cy­cles. This will be a com­pet­i­tive ad­van­tage to those who adopt early and get it right.

Ma­chine learn­ing can help bridge the gap be­tween sales and mar­ket­ing by de­ter­min­ing the cor­re­la­tion be­tween mar­ket­ing pro­grams and unit sales. Fur­ther, it can help with more ef­fec­tive cus­tomer seg­men­ta­tion to sup­port mar­ket­ing cam­paigns, pro­mo­tions and ef­forts to close on sales.

Ex­pect HR de­part­ments to lever­age ma­chine learn­ing for re­cruit­ing and re­ten­tion of top tal­ent.

Op­er­a­tions will get smarter at plan­ning, re­source de­ploy­ment, sched­ul­ing and pur­chas­ing as a re­sult of ma­chine learn­ing.

The fi­nance depart­ment will de­ploy ma­chine learn­ing to help man­age cash flow, speed ac­count rec­on­cil­i­a­tions, and im­prove over­all fi­nan­cial plan­ning.

But while ma­chine learn­ing of­fers tremen­dous ben­e­fits, there are many chal­lenges in im­ple­ment­ing it as well. Any ma­chine learn­ing ap­pli­ca­tion is only as good as the data to which it has ac­cess, as well as its over­all mod­el­ing struc­ture to ac­cess, learn and act. Us­abil­ity is an­other key fac­tor to en­able broad use across the busi­ness.

The real race to pro­vide the best ma­chine learn­ing ap­pli­ca­tions per­haps lies in who gets the best data sci­en­tists. We are al­ready tak­ing great strides in ma­chine learn­ing, but this is still very much the early stage in what is largely un­charted ter­ri­tory.

Data sci­en­tists to­day are de­vel­op­ing best prac­tices that could de­fine the fu­ture of ma­chine learn­ing and other forms of ar­ti­fi­cial in­tel­li­gence.

Be­yond build­ing busi­ness func­tion­al­ity and op­ti­miz­ing al­go­rithms, how we han­dle out­lier re­sults, over­fit­ting or false pos­i­tives will de­ter­mine suc­cess.

Sim­i­lar to the early days of the cloud, the in­ter­net or even com­put­ing it­self, it is an ex­cit­ing time for ma­chine learn­ing.

Given the many ex­cit­ing ap­pli­ca­tions ma­chine learn­ing prom­ises, ex­pect to see more so­lu­tions com­ing to mar­ket soon.

AT

Tom Shea is CEO of Onestream.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.