Rotman Management Magazine - - CONTENTS - In­ter­view by Karen Chris­tensen

In ad­di­tion to teach­ing at the Rot­man School, you work for a suc­cess­ful start-up. What prob­lem is your com­pany, Ru­bik­loud Technology, try­ing to solve?

Ru­bik­loud is a ven­ture-backed start-up com­pany in Toronto that uses AI to en­able in­tel­li­gent de­ci­sion au­to­ma­tion for some of the world’s largest re­tail­ers. Our goal is to build AI soft­ware sys­tems that can take the hun­dreds of de­ci­sions that a re­tailer has to make ev­ery day and au­to­mate and op­ti­mize those de­ci­sions. We’ve been work­ing with some of the largest re­tail­ers around the world, from Asia to Europe to North Amer­ica. Our most re­cent ven­ture round, led by In­tel Cap­i­tal, led to a to­tal in­vest­ment of $45 mil­lion, which we hope will en­able us to change the technology land­scape for en­ter­prise re­tail­ers.

Can you de­scribe what ex­actly your prod­uct does?

There are two main func­tions. First, on the cus­tomer life­cy­cle side, it pro­vides an un­der­stand­ing of the cus­tomer on an in­di­vid­ual level for tar­geted pro­mo­tions and mar­ket­ing ac­tiv­i­ties that en­cour­age them to shop at a par­tic­u­lar store. Sec­ond, it ad­dresses the mass mar­ket and mer­chan­dis­ing side, where we help re­tail­ers fore­cast and op­ti­mize the dif­fer­ent pro­mo­tions they do in-store. When you are at the grocery store and you see that a cer­tain brand of milk is on sale for $2.99, there is ac­tu­ally a com­plex process in the back­ground to make that hap­pen. We help re­tail­ers se­lect the right prod­ucts to put on pro­mo­tion, de­ter­mine what those pro­mo­tions should be, as well as fore­cast to make sure they have the right in­ven­tory at the stores.

As the com­pany’s Chief Data Sci­en­tist, what is your role in all of this?

I’m re­spon­si­ble for all the data science func­tions, which pri­mar­ily re­volve around build­ing out our ma­chine learn­ing and ar­ti­fi­cial in­tel­li­gence ca­pa­bil­i­ties. We have two key ar­eas of fo­cus: ap­plied R&D and fun­da­men­tal R&D. The ap­plied team is re­spon­si­ble for re­search­ing and build­ing out the core of our AI sys­tem in the kind of prod­ucts I men­tioned ear­lier, which in­clude ma­chine learn­ing mod­els and the un­der­ly­ing soft­ware sys­tems that al­low us to au­to­mate, scale and store all the data and mod­els.

The fun­da­men­tal re­search side is more fo­cused on strate­gic long-term is­sues for the com­pany. The things they work on may not even ap­pear in our prod­ucts in the next five years, but we rec­og­nize that we need to be look­ing ahead and have a long-term roadmap as to what our prod­ucts will be able to do one day. We’ve also be­gun to en­gage with the aca­demic com­mu­nity and are hop­ing to pub­lish some of our re­search in the near fu­ture.

Tell us a bit about your ca­reer path, and what led you to this role.

I joined Ru­bik­loud early, in the first year of its ex­is­tence. There’s a pop­u­lar say­ing, that ‘one year at a start-up equals three years at a big­ger com­pany’. I’ve been here four years, so maybe that re­ally means 12 years? But se­ri­ously, one thing that re­ally helps to ac­cel­er­ate a ca­reer is be­ing in­volved at an early stage with a high-growth start-up. There are op­por­tu­ni­ties to grow and learn that frankly, you can’t find any­where else. I have ba­si­cally watched this com­pany grow from a half a dozen peo­ple to over 100 in a very short time span. We’ve

been dou­bling our head­count pretty much ev­ery year. In that time, I’ve done some en­gi­neer­ing work and even some client fac­ing stuff in ad­di­tion to my data science work.

The key thing about hy­per-growth start-ups is that if you’re go­ing to thrive in one, you re­ally have to be quick to adapt and learn. In terms of man­age­ment, the ap­proach that worked with 25 peo­ple will need to change when you get to 50 peo­ple, and that ap­proach will not work when you get to 100 peo­ple. You re­ally have to pay at­ten­tion, not just to the technology, but to the or­ga­ni­za­tional and man­age­ment as­pects. I’ve been very for­tu­nate along the way, be­cause I’ve had great men­tor­ship from our CTO, Waleed Ay­oub, whom I re­port to — and who gave me the op­por­tu­nity to step up and run the data science side of the busi­ness.

What does it take for an or­ga­ni­za­tion to make the most of its data? Are there some prin­ci­ples you can share?

The ele­phant in the room is that it is im­pos­si­ble to have good AI with­out good data.

It ac­tu­ally takes a lot. The ele­phant in the room — es­pe­cially with re­spect to AI — is that it is im­pos­si­ble to have good AI with­out good data, and good data is a re­ally hard thing to achieve. From col­lect­ing it, to clean­ing it, to build­ing the right technology plat­form so that you can ac­cess it in a scal­able way — it’s very com­plex. To give you some con­text, for a large en­ter­prise to mod­ern­ize its data prac­tice, it could take tens of mil­lions of dol­lars and in­volve dozens of peo­ple over a multi-year time span.

In my ex­pe­ri­ence, most of the chal­lenges com­pa­nies face around AI are re­ally around their data. The first step for lead­ers is to un­der­stand and ac­cept the com­plex­ity of get­ting good data, and start treat­ing it se­ri­ously. The fo­cus on this should be on the same level as fi­nan­cial re­port­ing and re­sults. If you are go­ing to be­come a data-driven com­pany, it takes a lot of work.

Are busi­nesses em­brac­ing AI and ma­chine learn­ing to the ex­tent that they should?

I work with en­ter­prise re­tail­ers, but what I see with them is prob­a­bly typ­i­cal of most in­dus­tries: They are ap­proach­ing this new technology care­fully and with mea­sure. The hard part is that they are get­ting squeezed from both ends. A few re­ally big play­ers are dom­i­nat­ing the in­dus­try, like Ama­zon; then, there are all of th­ese smaller start-ups tar­get­ing niches that are also gain­ing a lot of trac­tion—and both sets of chal­lengers are us­ing AI. This has re­ally opened them up to mak­ing the tran­si­tion.

In many cases, it’s dif­fi­cult for an or­ga­ni­za­tion to do th­ese things in­ter­nally. Chang­ing ex­ist­ing busi­ness pro­cesses and legacy sys­tems and hir­ing the right peo­ple is very dif­fi­cult. That’s why a lot of com­pa­nies are will­ing to part­ner with smaller com­pa­nies to help them make the tran­si­tion to AI and the next gen­er­a­tion of technology.

Tell us how you be­came in­volved at the Rot­man School.

A good friend of mine in­tro­duced me to Mi­h­nea [ Moldoveanu], the Vice Dean of In­no­va­tion and Learn­ing at Rot­man. We both have an en­gi­neer­ing back­ground, so we got to chat­ting. He was es­pe­cially in­ter­ested in the AI work that I’d been do­ing, and he iden­ti­fied a cou­ple of re­lated ini­tia­tives at Rot­man: the new Mas­ter of Man­age­ment An­a­lyt­ics (MMA) Pro­gram and the newly cre­ated Man­age­ment Data Lab. He told me they could re­ally use some­one from the in­dus­try with AI ex­per­tise. From there, I met with Dmitry Krass, aca­demic di­rec­tor of the MMA, as well as Susan Christof­fersen, Vice Dean of Un­der­grad­u­ate and Spe­cial­ized Pro­grams. We all agreed that it would be great for stu­dents to have some­one with in­dus­try ex­pe­ri­ence help­ing out with the pro­gram, as well as in the data lab.

I started work­ing with them last April. Ev­ery­one is re­ally pas­sion­ate about try­ing to make the MMA Pro­gram as good as it can be in or­der to help stu­dents bridge the gap be­tween tech­ni­cal ex­per­tise and the man­age­ment side, as well as bring­ing the School’s data science prac­tice into the next gen­er­a­tion by build­ing up the Man­age­ment Data Lab. On the MMA, side, I’ll be teach­ing a cou­ple of mo­d­ules on deep learn­ing and neu­ral net­works, which have been all the rage in the past lit­tle while. And I’m also work­ing with the Man­age­ment Data Lab to try to un­der­stand what the needs are within Rot­man, and help­ing build out the hard­ware and soft­ware ca­pa­bil­i­ties to sup­port data science teach­ing and re­search at Rot­man.

Some peo­ple ar­gue that data can’t solve prob­lems be­cause by its very na­ture, it is in­for­ma­tion about the past, not the fu­ture. How do you re­act to that view?

I’ve heard that be­fore, and I agree with it to some ex­tent. It’s kind of like what Yogi Berra said many years ago: ‘It’s tough to make pre­dic­tions, es­pe­cially about the fu­ture’. What is miss­ing from this per­spec­tive is the idea that data can be used in many dif­fer­ent ways — and for a cer­tain cat­e­gory of prob­lems, the fu­ture ac­tu­ally does re­sem­ble the past. Ob­vi­ously in such cases, data is help­ful.

How­ever, I would ar­gue that even when the fu­ture is more un­cer­tain, data can help a prob­lem solver quan­tify that un­cer­tainty. For ex­am­ple, it can’t tell you, ‘it’s def­i­nitely go­ing to rain to­mor­row’, but it can tell you ‘there’s a 70 per cent chance of rain to­mor­row’. In the realm of busi­ness, it’s not able to say ‘this cus­tomer is def­i­nitely go­ing to buy this prod­uct’ — but it can say, ‘there is a high chance that this cus­tomer will buy some­thing’. Even if you’re not sure ex­actly what’s go­ing to hap­pen, it can in­form you about some­thing that oth­er­wise would be left to in­tu­ition.

I do agree that we shouldn’t blindly use data to solve all our prob­lems. We need to un­der­stand the ways in which data is use­ful and use it ju­di­ciously. It’s not like we’re ever go­ing to throw away hu­man judg­ment, but the fact is, data can be a huge boost to many of the de­ci­sions we are mak­ing.

Some peo­ple still fear that machines will take over most of our jobs, but in my view, they needn’t worry. Take med­i­cal di­ag­no­sis as an ex­am­ple. Al­ready, ma­chine learn­ing al­go­rithms do a pretty good job at di­ag­nos­ing — and in some cases, they do even bet­ter than hu­man doc­tors. But my wife is a physi­cian, and she un­der­stands that di­ag­nos­ing a pa­tient is not just a me­chan­i­cal set of de­ci­sions. It in­volves a lot of hu­man in­ter­ac­tion and ob­ser­va­tion, which a ma­chine can never re­place.

Fur­ther­more, the technology is nowhere near be­ing able to pro­vide ‘ar­ti­fi­cial gen­eral in­tel­li­gence’ — whereby a ma­chine could po­ten­tially re­place a hu­man mind. And as in­di­cated, many or­ga­ni­za­tions are lim­ited as to what they can ac­com­plish, given the type of data they have. A ma­chine is only as good as the data used to train it, and we have a long way to go on that front. There are also some things that we just can’t col­lect data for, or that it’s too hard to col­lect data for, and as a re­sult, at the end of the day, a ma­chine is never go­ing to have the judg­ment that a hu­man has. Hav­ing said that, hu­mans have their own bi­ases and other de­fi­cien­cies, so I be­lieve the strong­est ap­proach is a union be­tween hu­man and ma­chine ca­pa­bil­i­ties.

If you could change some­thing about the cur­rent en­tre­pre­neur­ial land­scape, what would it be?

Es­pe­cially in Toronto, I would have a lot more Sil­i­con Val­ley-type ven­ture fund­ing. The cul­ture and cal­cu­lus of how Cana­dian in­vestors think is fun­da­men­tally dif­fer­ent from their Sil­i­con Val­ley coun­ter­parts. I’m gen­er­al­iz­ing a bit here, but in Canada it is much more risk-averse. In­vestors tend to be look­ing for a ‘high like­li­hood in­vest­ment’, whereas in the Val­ley, they are very much fo­cused on find­ing ‘the next big thing’ — the next Face­book or Uber— and that leads them to in­vest in more risky propo­si­tions. They truly do not care if 19 out of 20 op­por­tu­ni­ties fail, as long as one re­sults in a big win. That’s the mind­set we need to have if we’re ever go­ing to gen­er­ate the next big tech start-up in Canada.

For a cer­tain cat­e­gory of prob­lems, the fu­ture ac­tu­ally re­sem­ble the past, and in such cases, data is help­ful.

Brian Keng, PHD is an Ad­junct Pro­fes­sor of Data Science at the Rot­man School of Man­age­ment and the Chief Data Sci­en­tist at Ru­bik­loud, a ven­ture-backed AI startup in Toronto. He teaches in the Rot­man School’s new Mas­ter of Man­age­ment An­a­lyt­ics Pro­gram.

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.