Un­bounce’s AI com­piles mar­ket­ing in­sights

The Georgia Straight - - High Tech -


n 2004, Block­buster em­ployed 84,000 peo­ple and op­er­ated in 14 coun­tries, rent­ing out videos and games to mil­lions. Six years later, af­ter the ad­vent of video-on­de­mand and stream­ing com­pa­nies, it was forced into ad­min­is­tra­tion. The story is not too dis­sim­i­lar from that of cam­era gi­ant Ko­dak. Af­ter an em­ployee in­vented and patented the dig­i­tal cam­era in 1975, the com­pany shelved the de­vice be­cause ex­ecs be­lieved that cus­tomers would never want to look at pho­tos on a tele­vi­sion. Too slow to switch to dig­i­tal, the business filed for bank­ruptcy in 2012.

Both are cau­tion­ary tales of what hap­pens when busi­nesses fail to em­brace tech­no­log­i­cal in­no­va­tion. It’s a mis­take that Van­cou­ver com­pany Un­bounce is not will­ing to make.

“This is straight out of one of my favourite books, The In­no­va­tor’s Dilemma,” Carl Sch­midt, the com­pany’s CTO, tells the Ge­or­gia Straight at Un­bounce’s an­nual Call to Ac­tion con­fer­ence. “If you don’t keep pace—if you don’t adapt your business model to chang­ing tech­nol­ogy—you will die. It’s that sim­ple. Gen­er­ally your cus­tomers will be solv­ing to­day’s prob­lems, and gen­er­ally you need to be in­vest­ing in the prob­lems that they’re go­ing to face a cou­ple of years down the line.”

For the past few years, Un­bounce has cho­sen to fo­cus on how ar­ti­fi­cial in­tel­li­gence can be used to fig­ure out which fac­tors make up the most at­trac­tive pages on the Net. A business that al­lows com­pa­nies to eas­ily de­sign web­pages and A/B– test which vari­ants per­form best in front of cus­tomers, Un­bounce has cre­ated a ma­chine-learn­ing sys­tem that asks com­put­ers to rate land­ing pages—typ­i­cally, a site’s home­page—on how likely users are to en­gage with them.

Shock­ing Sch­midt, the ma­chine was able to guess which web­page would per­form best at a rate of 79.7 per­cent. Hu­mans, mean­while, were only right half of the time.

“The sur­pris­ing thing is just how im­pact­ful words ended up be­ing,” he says. “We were able to pre­dict a lot of things based just on the words. That told us that words played a po­ten­tially out­sized role in the per­for­mance of these pages. We’ve said this a lot, that the other el­e­ments are clearly im­por­tant, but per­haps—and we still have to do more work here—there’s a strong sig­nal here that copy and writ­ing is dom­i­nant.”

Us­ing its ma­chine-learn­ing al­go­rithm, the com­pany was able to com­pile a num­ber of in­sights that mar­keters could use to tar­get con­sumers. Real-es­tate pages, for ex­am­ple, per­form best when writ­ten at a sev­enth-grade read­ing level, whereas col­lege ad­mis­sions web­sites are more likely to en­cour­age user en­gage­ment with uni­ver­sitylevel copy. Lan­guage that builds a feel­ing of an­tic­i­pa­tion or pres­sure in the home-im­prove­ment in­dus­try turns vis­i­tors off, while an ex­cess of pos­i­tive words on a web­site for le­gal ser­vices makes it seem less cred­i­ble to readers.

Af­ter gath­er­ing a glut of re­search data, the com­pany felt suf­fi­ciently con­fi­dent about its re­sults to re­lease its AI into the wilds of the In­ter­net. Over 30 Un­bounce clients of­fered to be guinea pigs for new tech­nol­ogy, and con­nect their cam­paigns to the com­pany’s en­gine. Af­ter a few weeks, those busi­nesses saw a lift of around 20 per­cent in con­ver­sions, with one tal­ly­ing a 117-per­cent im­prove­ment. Sch­midt at­tributes that suc­cess to the AI’S abil­ity to de­liver the right page to the right per­son.

“There’s no one mes­sage which is go­ing to work best for your en­tire au­di­ence,” he says. “We know that words af­fect us dif­fer­ently based on our own bi­ases and per­cep­tions and ex­pe­ri­ences. There’s no way that one page can be the best for all of your vis­i­tors. Right now this is patented tech, so we’re go­ing to hold it a lit­tle close un­til we fig­ure out how we’re go­ing to prod­uct it. Then we’ll be talk­ing a bit more about how it does what it does.”

If the com­pany is able to repli­cate those re­sults on a wider scale, Un­bounce’s AI tech­nol­ogy could spell big changes for con­sumers across the In­ter­net. Al­though the com­pany has not di­vulged ex­actly how its cre­ation will op­er­ate, ma­chine­learn­ing tech­nol­ogy in the fu­ture could be able to tai­lor each page for an in­di­vid­ual, cre­at­ing a more per­son­al­ized ex­pe­ri­ence, and en­cour­ag­ing more sales for com­pa­nies. In Schimdt’s view, AI is a win for both browsers and busi­nesses.

“The rea­son AI is im­por­tant is be­cause the world is be­com­ing pretty over­whelm­ing,” he says. “We’ve got flooded in­boxes; we can’t keep up on our so­cial feeds. It’s hard to just get the in­for­ma­tion that is re­ally rel­e­vant. I think that AI can help us as con­sumers ul­ti­mately deal with the in­for­ma­tion fire­hose bet­ter, and be a fil­ter for us. In the end, that will give us back more time.

“AI is go­ing to be one of those dif­fer­en­tia­tor tech­nolo­gies,” he con­tin­ues. “Same as when the In­ter­net came along and changed a whole bunch of business mod­els, and chal­lenged print, chal­lenged mail-or­der—all of these things. It’s that class of tech­nol­ogy.”

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.