00:00:00
Hello and welcome. You're listening to Marketing Matters on the Wharton Podcast
00:00:05
Network, our weekly podcast where we analyze the latest in advertising, marketing, customer behavior, new
00:00:11
product launches, retailing, and pharmaceuticals. Uh, I'm Barbara Khan, the Patty and JH Baker Professor of
00:00:17
Marketing, and I'm joined by my co-host, Americus Reed, the Whitney M. Young Jr.
00:00:22
Professor of Marketing and the brand identity theorist. Hello, Americus. >> Hi, Barbara. So, uh, this is the first
00:00:29
week of classes. It's an incredible sort of energy on the campus. I'm loving it.
00:00:33
Uh, one of the things we talked about today is level setting. >> I just mentioned it's your first week of
00:00:38
classes. I've been teaching seven weeks. >> Seven weeks in. So, you're like, listen,
00:00:44
I may be I'm into it now. America, you're brighteyed and bushy tail, but
00:00:48
hey, it's going to catch up with you. >> We're all on midterms. I don't know
00:00:51
where you are. >> I'm introducing the syllabus. And so, yeah. No, it's it's it's fascinating
00:00:58
that you mentioned that because you know all of the things that are going on campus. One of the things I talked about
00:01:02
today, Barbara, is we're talking about marketing and people think marketing is
00:01:05
just advertising, but I was trying to explain to my class marketing is about strategy. And in fact, marketing
00:01:11
actually starts with science and ends with art. And so I'm very fascinated by
00:01:17
this interface of data science, artificial intelligence as a new technology that is making its way into
00:01:24
marketing strategy, various ways that it's happening and all of these different interfaces. And so have you
00:01:30
got anything for me today that can help illuminate this and further allow me to make the case to my students that
00:01:37
there's a lot going on under the hood with respect to marketing strategy? Oh,
00:01:41
well actually we have a very interesting guest today who can talk about a lot of
00:01:45
those things because he's have a long and varied career. I'd like to introduce
00:01:50
Shawn Buick who's currently the senior vice president of artificial intelligence and data at Amgen which is
00:01:58
a pharmaceutical or biotech or anyway health kind of related company but he has a long history starting out at
00:02:06
Stanford as a psychology and biological science major all the way through Google
00:02:12
and Facebook and Nike. So he's kind of been at the right place at the right time. Anyway, welcome Sean to our show.
00:02:22
>> Thank you, Americas. Thank you, Barbara. It's so great to be here. >> So, let's build on on Americas's um
00:02:29
quest or introduction in some sense or question and just quickly go through your background because you do have a
00:02:35
knack of being in the right place at the right time starting well Stanford wasn't
00:02:39
the right place. You should have been a Wharton, but okay. >> Had to get that plug in. Very, very well
00:02:44
done, Barbara. But then uh I think you went first to Google then to Facebook then to Nike and now to Amgen just as
00:02:51
each one of those areas was doing some pretty interesting area connecting marketing with data science and now what
00:02:59
I really like to talk about is what you're doing at Amgen. But let's start
00:03:02
with your history. >> Sure. Yeah. I got my start uh up at Stanford and um I was working in a
00:03:09
neuroiming lab. So this was right around the time you know this these new brain scanners were be being able to gaze into
00:03:17
the working human mind functional magnetic resonance imaging. Um and and there were kind of two things that
00:03:23
happened there. I think the first was you know this new technology allowed us to for the first time answer some
00:03:30
questions that for 70 years in cognitive science weren't really answerable. And
00:03:35
so I became really convinced up until that point I'd been you know studying the science but I became convinced that
00:03:41
technology was going to be really important in understanding the world. And then the second thing was I had to
00:03:48
learn how to code because this these data sets were so much bigger than the reaction time and wet lab and um you
00:03:56
know survey data that sort of everybody was using at the time. And so that's
00:03:59
what really got me started um in in my career and I I got a job at Google right at the beginnings of what you would now
00:04:07
call consumer data science. >> Um had had just a a you know wonderful set of mentors there as we were um sort
00:04:15
of exploring what was possible with this field. I went to Facebook and helped them launch the the ads business um
00:04:22
where I led the the measurement science team, the data science team on the on the ads business. And then I spent the
00:04:28
last 10 years at Nike where um I helped them with their direct to consumer digital transformation leading their
00:04:35
data science and analytics and customer insights teams there. But as you said, I
00:04:39
made a big uh switch about a year ago. So now I met I'm at Amgen which um if if
00:04:46
folks don't know is one of the world's leading biotechnology companies. It
00:04:49
makes medicines um for things like uh cancer and inflammatory disease and cardiovascular disease and rare disease.
00:04:58
Um it's been around for for over 40 years. It was sort of one of the pioneering companies in the biotech
00:05:04
revolution of the 1980s. Um, and my job here is to try to figure out how we can use AI and data uh to make better
00:05:11
medicines for for patients. >> So, let me just build a little on this. Now, you've been there a year and um I
00:05:19
mean I know you were what that health what did you say bio health or biological science undergrad major but
00:05:25
that's not like major health credentials. So, but obviously they saw some advantage in what you did do for
00:05:34
the last 10, 20 years. Uh, but how have you like learned about what Amgen does and h how did how did you get up to
00:05:42
speed? How did you do that? >> Yeah. And and it's been a long time since I took my last biology class at at
00:05:51
Stan Smith. >> Yeah. And I mean, the science here is is really complex. you know, it's genetics,
00:05:58
it's it's biology, it's pharmarmacology. Um, and so it has been definitely a a a
00:06:05
big ramp to to learn all about what we do at Amgen. And and you know, it is pretty different than online ads or
00:06:13
shoes in a few in a few ways. I mean, I think the first is just the fact we're,
00:06:18
you know, making medicines for for patients, which creates this incredibly high standard around reliability and
00:06:25
safety and compliance and scientific accuracy. Um, but it is a scientific discovery, especially in in this space.
00:06:35
Um, you know, where you're trying to take this molecule and turn it into a a
00:06:39
medicine. It it uses really large data sets. It's actually surprisingly quantitative for me. I mean, there's
00:06:47
definitely beers and labs and experiments, but it's also very quantitative. Um, and and it's
00:06:54
increasingly relying on technology because the data sets are huge and these, you know, new artificial
00:07:02
intelligence and machine learning algorithms we think um has has the potential to really change um sort of
00:07:08
how we bring these these medicines to patients. So, so some of it is definitely new and and hard and and a
00:07:16
ramp, but some of it feels, you know, much more comfortable, the the core first principles of data science or or
00:07:22
the um, you know, underlying technology stack and the partners and things like that.
00:07:27
>> So, if everybody is using AI, then is how do you differentiate? I mean, it
00:07:33
seems like if you have AI, everybody's doing the same thing. Then how how does
00:07:38
that work? No, you you bring up a great point, Barbara, and and I do think there's a lot of convergence now on how
00:07:46
well these algorithms perform, you know, so in the old days, um you know, when I
00:07:51
was getting my start as a as a data scientist, you know, your your magic sauce was very much like how good your
00:07:58
your algorithm was, how good your model was. Um nowadays I think I think companies should be thinking about two
00:08:06
things and and certainly we are here at at Amgen but I think if you look across uh you know my work work previously
00:08:13
there was it was the same concepts you know the first is you the underlying data that you have as a company is is a
00:08:21
pretty interesting differentiator and and I'm super excited about some of the
00:08:26
data we have here at Amjen you know genetics data a long history of designing proteins um and molecules and
00:08:34
and all those scientific experiments along the way that there's that you know
00:08:38
that that set of data uh is is a really interesting differentiator >> and then I think increasingly what
00:08:46
you're seeing is that there is an art and a science because you mentioned kind
00:08:50
of the art and the science um of how do you implement these algorithms in a way that you know is impactful in a way that
00:08:57
gets to scale quickly >> um and and realizes whatever impact you're trying to drive at at your
00:09:04
company. And I think in both spots, there's going to be differences in how successful companies are. And so, um, I
00:09:11
think I think that's kind of one of the most important and underappreciated parts of of how to win here.
00:09:17
>> I love that point, Sean. One of the things that I talked to my class Sean is
00:09:20
the idea of adoption of AI technologies and the statement that you made which is
00:09:25
how do you make the company strategy infused deeply with the DNA of AI that seems not necessarily easy because you
00:09:34
have different folks maybe even if it's a large company and you know aside from
00:09:38
just having a bunch of young people who are sort of born into it in your company
00:09:42
how do you go about creating that the right organizational culture and atmosphere
00:09:48
where the adoption of these technologies and understanding of how to use and deploy them appropriately and well
00:09:53
differentiated like Barbara said, how do you make that happen? What are the what
00:09:57
are the things that you do to create the cohesiveness inside the building to be able to adopt that kind of culture?
00:10:04
>> Yeah. I mean, this is a podcast, right? So, we have to have some controversy.
00:10:07
Can I bring up the MIT AI paper because it Yes. Definitely a controversial topic
00:10:14
and I think one that's worth digging into a little bit because I think it it
00:10:17
strikes right at the heart of of kind of your question. Okay. So for if >> if people are not tracking, there was a
00:10:24
a paper by MIT that came out uh probably four weeks ago that you know the headlines have all said 95% of AI pilots
00:10:32
in companies don't generate value. Um now that's the headline. The real story
00:10:39
I think is is is twofold. So the first story is that if you're in a big company
00:10:44
and you do a pilot, it never generates scale business value. That's the whole
00:10:49
point of a pilot, right? Is it's little. Um now there's a muscle around how do
00:10:56
you choose the right pilots? because pilots are only valuable if they teach you something you didn't know or derisk
00:11:04
an investment by by you know trying it out in a specific zone or or sometimes I think what you as you mentioned you know
00:11:11
shines a light on a success story that can absolutely inspire and teach your your colleagues and your workforce a
00:11:18
path forward but lots of pilots lots of sort of poorly thought through poorly targeted random pilots is not a measure
00:11:28
of success. And there's a lot of companies, I think, doing way too many pilots that are kind of not ever going
00:11:33
to generate real value because then you have to have the practice and the and the skill set and the technology stack
00:11:40
and the monitoring capabilities to take the winners and get them scaled and adopted and and that I think is a is is
00:11:49
a real lesson um from this paper. Now, now why did these 95% of these programs in this paper fail?
00:11:58
This isn't always in the headline. It wasn't the AI technology. It wasn't like
00:12:02
the algorithms were crappy. It wasn't that the that the AI failed. It was everything else about it. There wasn't
00:12:10
alignment on the budget. Business processes weren't evolved. There wasn't
00:12:16
receptivity on the part of the people using it. They didn't get adoption. And
00:12:21
and so that speaks exactly to your point, which is even if you have an amazing capability, if if you don't get
00:12:28
everybody else lined up around it in a big company, you're never going to get
00:12:32
the the business value out of it. And I think that's another piece that companies are struggling with is is how
00:12:37
do you line up everything such that you can really pull these through to value um and and make an impact on the on on
00:12:44
whatever mission you're trying to accomplish. >> Well, so how do you do it?
00:12:50
No, it I I think it's this there's there's three pieces. I think you have
00:12:54
to have the discipline around where are you going to do the pilots and what are they designed to teach you? And you
00:12:59
know, here at Amgen, we have a we have a very rigorous process. It it's not anybody can run any pilot anywhere
00:13:05
around AI. We're we're directing those pilots at specific zones that are going
00:13:10
to teach us something. You have to have this muscle around how do you take a great pilot and and get it
00:13:17
to scale. So and and and you know that's a technical question around you know do
00:13:23
you have the right data infrastructure do you have the right um technology portfolio do the right monitoring uh
00:13:29
programs to get these algorithms implemented and then there's all of the change management stuff that you know is
00:13:37
not all that different than every other technology transformation that you guys have studied you know that we've looked
00:13:43
at digital marketing you know 15 years ago direct to consumer you know 10 years ago So um but if but if you don't have
00:13:50
all three of those I think it's really hard to make a big impact and and you
00:13:55
know one or two of them isn't enough. >> So let me let me just summarize that. So
00:14:00
like the first one is kind of how to ask the right questions right where to focus
00:14:04
is the first thing and then and you can ask a lot of wrong questions but understanding where you're focusing is
00:14:10
basically what you're saying and I could totally see how that'll differ from firm
00:14:14
to firm to firm. So it wouldn't result in the same thing. And then once you get
00:14:18
viable ideas, implementing and and bringing to and operationalizing and then scaling is a whole other issue. And
00:14:25
then the third one is getting the organization around this new is that it >> Yeah, I think that's a great a great
00:14:34
summary. It's it's you know the discipline around pilots, where you play, how you test ideas, how you turn
00:14:40
the Yeah. turn turn the big ones the the best into impactful scale programs and then how do you line up all of the
00:14:48
pieces that are necessary to to make to make the impact through those through those uh scale pilots for sure. So it
00:14:56
seems like how to do it right would suggest that there's like some measurement system or some objectives
00:15:02
you and some like metrics that you have along the way in order to see if you're
00:15:08
making prog is that kind of the way it goes or how do you know you're on the
00:15:12
right track? Yeah, I look I think there are um you know a whole host of technical things we do to make sure
00:15:20
these models are performing well and you know having thoughtful evaluation frameworks and um you know looking at
00:15:27
their performance and um all of that stuff is is sort of the job of the data science team or the AI team. But I think
00:15:35
you know the the bigger question goes back to that conversation we were just having around is it actually making an
00:15:41
impact on your strategy. You know is it is it driving whatever outcomes you as a
00:15:47
business are trying to are you trying to trying to enact. And that I think is the
00:15:52
place where sometimes you know the plot gets a little bit lost. you focus too much on the technical implementations or
00:15:59
the technology roadmap or the or you know the path that the that all the um the data and the algorithms and the
00:16:07
technology stack need to be um all right and and sometimes you lose track of the
00:16:12
the organizational and business partnership. I mean we understand what you're saying Sean which is like the the
00:16:18
specific role of the technology needs to be managed in a way that is built around
00:16:22
the synchronization of its adoption within the organization and how to in your case how to figure out the bi the
00:16:30
specific biology that produces new opportunities for wellness products or health related uh protocols and things
00:16:39
of that nature. So we definitely understand that it's just challenging because and whenever I hear a talk Sean
00:16:45
on AI and you're a guru of this I sit back and I have to like suspend a lot of
00:16:49
my judgment because I just don't know what you're talking about in terms of
00:16:52
the technical pieces of it. You know all of these s sort of things. I understand
00:16:56
the logic of it but I want to know much more to Barbara's point about you know
00:17:01
how does this work? What is a what's an example of a specific pilot? if you could share that, you know, things that
00:17:06
we've that you guys have done. Maybe that's the secret sauce and you can't
00:17:09
talk about it, but like taking us through like here's something we did that's, you know, more public
00:17:14
information now and here's an example of how we really brought in AI to really
00:17:19
help us drive a specific understanding of a biological element that then turned into something that we could quickly
00:17:24
scale and create something that adds value to the marketplace. Is there an example you could share with us?
00:17:29
>> Sure. Yeah. I mean I think what is exciting and you know part of why I wanted to come to Amgen is I really do
00:17:34
think biotechnology is is a spot where um some of the biggest advances in AI and data science are going to happen
00:17:44
over the course of the next few years. I mean I I am so excited about what's
00:17:48
possible and there's been some big breakthroughs and you know just last October
00:17:53
um the winner of the Nobel Prize in chemistry was shared between the Google deep mic guys Dennis and a professor at
00:18:01
the University of Washington David Baker. And you know what that that breakthrough um was was an algorithm
00:18:07
called alphafold which was solving this like incredibly complicated uh problem in science which is how do
00:18:14
you predict the way po proteins fold if you know the >> the the two-dimensional sort of
00:18:20
structure of the protein can you predict the three-dimension structure and that >> interesting
00:18:24
>> you know fortunately that actually happened after I joined Amgen because it
00:18:28
attracted a whole bunch of news to the fact that AI and data science had had a potential potentially, you know, a big
00:18:35
impact on on this field. But since I had joined a few months earlier, it looked like I was super smart and my timing was
00:18:41
perfect as opposed to uh joining joining after the the news coverage. But that's
00:18:45
an example, I think, of a big breakthrough. And, you know, there are relatively few of those along the way. M
00:18:52
>> um you know I think the the more common instantiation of like an incredible
00:18:58
impact of a of a data science uh project is actually just much more simple. So let me give you one.
00:19:05
>> Okay. >> I did not know any of this a year and a half ago but but it turns out to be a
00:19:11
good medicine the the molecules have to be a certain very narrow range of viscosity. M
00:19:19
>> so like imagine >> you have a molecule it works pretty well but it's like peanut butter like too
00:19:25
thick >> you can't inject that into the human body it's not going to work and so one
00:19:29
of the most important criteria our scientists have to consider when they're designing these molecules is like how
00:19:35
viscous is the solution of these of these proteins after after we uh engineer them and you know even that
00:19:43
developing a a machine learning algorithm >> takes into account all the historical
00:19:48
experiments >> got it >> and makes a prediction as to whether these molecules are going to be in the
00:19:54
right zone to be a medicine >> can dramatically improve the efficiency of our scientists. So you know they
00:20:02
don't have to make a hundred of these molecules and then see how viscous they
00:20:07
are. they can use this algorithm and it dramatically cuts may not it may not be right all the time but it at least
00:20:13
improves the efficiency of those scientists lets them direct their energy their creativity their time against a
00:20:20
higher um fidelity of candidate right a small smaller pool more likely to be successful so you know those are the
00:20:26
type of like blocking and tackling yeah incremental work >> that you know before we had access to
00:20:33
all of that data being digitized and before our algorithms were powerful enough to make strong predictions, you
00:20:39
couldn't couldn't reduce that workload. You couldn't shape shape where they're
00:20:42
spending their time and attention. >> Interesting. >> That's the type of stuff just across the
00:20:46
board we're looking at of how do you you know add a little bit of data science
00:20:50
into the day of a protein engineer or a biologist or a geneticist um or a process engineer um you know a supply
00:20:58
chain expert, right? And and and just incrementally move the needle on how efficient and effective they can be. So
00:21:05
Sean, one of the things I'm interested in as you're telling us the critical
00:21:08
role that AI does play and how it can really make things better. One of the things people are really afraid of is AI
00:21:15
replacing other, you know, people working at a company. But the way you're describing it, it doesn't sound like
00:21:22
that to me, and I wonder if you agree with it. It actually sounds like there's
00:21:26
more room for people who know how to use AI and are trained to use AI and it might even provide more jobs than less.
00:21:35
I almost got that impression from what you're saying, but I don't know if you
00:21:38
want to go so far as to say something like that. >> Yeah, I'm no economist, so I won't weigh
00:21:43
in on on maybe the um the economics of it, but I'm I'm thinking about it in in
00:21:48
two ways. I think as we're implementing AI systems today, that human AI interface is actually critically
00:21:57
important. And so, you know, we call it human in the loop. that's kind of the
00:22:01
the term folks folks are using, but particularly when it pertains to like scientific inquiry and ensuring quality
00:22:10
and compliance and adhering to like, you know, the incredibly high standards of of scientific accuracy that we have to
00:22:17
adhere to. Having a human inside that system engaging with the technology and the AI algorithms themselves and the
00:22:26
output is is critically important. And I I don't see that changing anytime soon.
00:22:31
>> The other thing is as we as we build these systems, >> there's a whole set of new roles that
00:22:37
are being created >> that exist on on this edge. I mean, look, just just to zoom in on my career,
00:22:43
I have never had a job >> in which there had been a prior incumbent in that job over the last 25
00:22:52
years. about you, but >> yeah, >> I don't want to overextend from an N of
00:22:57
one, Barbara, but I I think everyone would agree that's dangerous in in science and marketing.
00:23:04
>> But I do think it speaks to this is an exciting time. I mean when I look at my
00:23:11
whole career and and know I've been around data science since the really early days this is the single most
00:23:18
interesting exciting moment that I've seen in that time. I think the capabilities you know the combination of
00:23:27
computing power and data that's that's been digitized and available um and the
00:23:33
performance of these of these new models is is really transformational. I think,
00:23:38
you know, tools like chat GPT are are just incredible. I'm sure many of your
00:23:42
listeners are are big users of it. >> Yeah, I'm sure they use it all the time.
00:23:47
>> And then, you know, I really do think one of the biggest challenges in data
00:23:50
science over the last 20 25 years has been just it's been really restricted to
00:23:55
people who are really highskilled like you know they know we know Python, know how to use the technology, work at a
00:24:02
company that has big data sets. But the other thing that's really fun here is is
00:24:06
this has really democratized the power of data and algorithms in a way that almost everybody out there listening can
00:24:12
probably find, you know, a handful of spots even just in their own lives, whether it's making shopping lists or
00:24:18
planning vacations or automating workflows, you know, getting work done faster. That's going to have a pretty
00:24:23
big impact. So, I'm I'm really excited about, you know, the positive impacts
00:24:27
that that these new technologies are going to have on on us at work and us outside of work. Yeah, Sean, this is uh
00:24:33
super intriguing to me. I I want to, you know, ask this question because I think
00:24:37
it's important when we have the gurus available to us to put them into the time machine and give us some insight
00:24:43
into their crystal ball about what is the state of the world in your domain, artificial intelligence, uh data
00:24:50
science, all of these things. Um what is that going to look like in your view five to 10 years from now?
00:24:57
Well, I am too good of a data scientist to get busted by making predictions too far into the future. So, I'm gonna I'm
00:25:04
gonna tell you what I think now and then maybe gaze just a little bit more approximately into the future. But um
00:25:09
you know, look, I think the biggest mistake people make in a moment like this around a big technological uh
00:25:17
revolution, evolution of of data science and algorithms and data is that they over focus on the here and now. the
00:25:25
failures of of technologies, the hype cycles that turn out to be, you know, overheated and they underestimate in the
00:25:34
medium term how pervasive and impactful some of these technologies can be. And I
00:25:39
I think that's the big disconnect I see right now between kind of the here and
00:25:44
now. you know, we've all had a moment where the algorithm didn't do what we
00:25:47
wanted it to do or there's not enough data to make the prediction or, you know, chatbt gives you a lousy answer.
00:25:53
And and those things are true. Um, but I I do think in the medium term, folks are
00:25:59
underestimating how big of an impact uh this this new set of capabilities is going to have. And that's a big part of
00:26:07
why I came to Amjen is I'm I'm really bullish on what I think is is possible
00:26:11
here. Now, you got to do all the stuff we talked about to to actually make the impact at at scale. Um, but I I think uh
00:26:19
this is a really exciting space to watch and folks should folks should lean in and watch it.
00:26:24
>> Well, Sean, thank you so much for joining us today. It's always a pleasure
00:26:28
to have you. And um where can our listeners go to to keep up with what you're doing? Do you have a website or
00:26:34
or public facing side of Amgen so people can figure out, you know, how amaning into AI? Yeah, definitely. If you're
00:26:42
interested in either the science or the AI of what we've talked about today,
00:26:47
check out amjen.com. There's a bunch of great stories and teams there keep everybody up to date on what we're
00:26:52
working on. >> Okay. Well, that's great. Thank you very much. And that's all we have time for
00:26:57
today. We'd like to thank our producers, Aaron TR and Marissa Rena. And thank you
00:27:03
all for listening. We'll be back next week. Till then, this has been Marketing
00:27:06
Matters on the Wharton Podcast Network. I'm Barbara Khan here with America's
00:27:12
Reed. [Music]