About Bill:

Bill Franks is the Director of the Center for Statistics and Analytical Research at Kennesaw State University.  He is also an accomplished author in the data science space. He joins Guy to discuss the ethics of data science, the issues with analytic models, and how data informs decision-making to drive ROI.

Links

Guy’s Special Offer

Find out More about ProRelevant

Subscribe to Guy's Newsletter

Bill Franks Website  

Bill Franks LinkedIn

Purchase Winning the Room

Purchase 97 Things About Ethics Everyone in Data Science Should Know

Purchase The Analytics Revolution

Purchase Taming the Big Data Tidal Wave

Video of Interview

Hi, I'm Guy Powell and welcome to the March episode

of the backstory on marketing. If you haven't already done so

please visit pro relevant.com and sign up for all of these

episodes and podcasts. I am the author of the upcoming book, the

post COVID marketing machine, prepare your team to win. You

can find more information on this ad marketing machine dot

pro relevant comm marketing machine dot pro relevant.com.

Today we'll be speaking with Bill Franks. He's the director

of the Center for statistics and analytical research at Kennesaw

State University. Bill is also the author of the new book

winning the room. And as well as taming the big data, tidal wave

the analytics revolution and 97 things about ethics. Everyone in

data sciences should know his work has spanned clients in a

variety of industries. For companies ranging in size from

Fortune 100 to small nonprofit organizations. You can learn

more at Build dash Frank's dot com Bill dash Frank's dot com.

Welcome, Bill.

Thanks for having me, Guy.

Yeah, glad to have you here. So while you're at

KSU, let's talk a little bit about KSU. I've always been

impressed with the, with what they're doing there. And so tell

us a little bit about that before we get started.

Yeah, KSU. It's part of the University System of

Georgia, probably obviously not as well known outside the

Atlanta area, as would be University of Georgia or Georgia

Tech. But it's it's rapidly growing. We're up over 40,000

students now we've got a, you know, full range of degrees.

We're adding more PhD programs each year, it seems so I think

over the coming years, more and more people are going to hear

about about KSU

Yeah, absolutely. And, you know, I think that, you

know, certainly Georgia Tech, and University of Georgia, but

Georgia State and the whole Georgia university system has

really, really taken off and, and KSU, I think is a big piece

of that. Yeah, so anyway, let's talk about data sciences. That's

kind of what you're in my fields are with some overlap on the

marketing side. So. So data sciences seems to be growing

faster than Data Science Talent, and which leads to a lot of open

positions. So what do you see as how KSU fits in with closing

that talent gap?

I mean, well, first of all, I think that the

interesting thing to me is that the talent gap is so big, I

don't even think universities all combined are necessarily yet

cranking out enough new talent to really fill all of that I

think we have, we have work to do as a as a society to

encourage more people to get into these fields. But I think

one of the things that we're doing specifically here is we're

really focused on having some very applied programs. So even

with our Ph D program, we look to get each student at least a

couple of years working with a corporate client. That's

actually the role of the center that I oversee that events begin

the Center for statistics and research, it partners with big

corporations, and they fund projects that we do joint

research with them on. So we do projects with companies like you

know, traveler's insurance and Home Depot, Georgia Pacific. And

these are real problems. So we're really trying to get our

students versed in what's actually happening with data and

data science in a business environment. And particularly as

we get on to the master's level, and so forth, we have capstone

courses, we support a variety of internships and so forth. So we

were trying to put out students that have a knowledge of how to

apply what they're learning rather than just the theory. And

to me, that's part of the gap we've had historically, as too

many, too many programs historically, put out people who

were technically highly proficient. But and I'll put

myself in this bucket when I came out of school, I was very

technically proficient, I didn't have any real practical

experience in how to apply any of that in a real world setting.

And I had to learn that all on my own, which, of course, is

time and money for the company that does hires you initially,

to get you up to speed. So I think that the more that

universities can and have started to do like KSU and focus

on not just the academic components, so I have a degree

in STEM field, but the applied nature of that, I think it's,

it's going to be beneficial.

Yeah, absolutely. I've been working with the Emory

consulting team in it well, business consulting, but usually

I get paired up with the marketing side of things and and

I think it makes a huge difference. And you can just

tell that those students that have had that practical

background, and have seen in the real business problem and how

the true data datasets work, and what the challenges are in the

real world, that makes a huge difference. And, and if you're a

hiring manager, that's really what you want to look for is you

want to look for somebody that's, that has some kind of an

understanding of what the real challenges are going to be in

terms of being able to actually apply those technical skills

that they've learned. A lot I can tell you, I, you know,

I'm personally overseeing one of the corporate

projects with some master students, and I was early on the

the status call with that client right before I joined you today.

And you know, they wanted us to change direction on one thing

and do something slightly different. So you know, I'm

having a conversation with the students, you realize pretty

much all the code you've been building to create these

variables is still fine. You have to do it as of a different

date, right? They no longer care about when the person initially

registered, they care about when they initially bought all of

your logics, the same with the exception that you have to

identify what did the customer look like, at a different point

in time? You know, I made the point that this is a common

thing. And you know, this market, right? The as of date is

something that I mean, you do it again, and again, and again. And

it's ubiquitous in any in any analytic it and in a lot of

different application areas. But it again, it's not something

that necessarily gets taught per se, right? It's it for the

students, it was new, oh, doesn't this change everything?

Well, no, you're just going to grab the customer record from a

different point in time. And it's really easy to identify the

proper record based on when it you know, when what the dates

related to it are. But at the same time, while it's very easy

to make that change, if you don't make that change, you

jeopardize all of the analytics, because you have the totally

wrong view of what that customer look like. So these are the

little practical things that I think it's honestly, without

getting your hands dirty. With a real project, it's almost hard

to, you can't really want to list every little thing that you

ever need to know, a lot of it you're gonna learn, but the more

that you can get students to see these things, while they're in

school, the easier it'll be for them to extend those when they

get their, you know, their real job.

Yeah, absolutely. And I think, you know, one of

the biggest challenges is, we're on the marketing analytics side.

So kind of a subset of what you guys are, what your folks are

doing. But the hardest challenge is really specifying what that

business question is that you're trying to answer. And then

really understanding that as you peel the layers back to say,

Okay, well, this is the business question, but you know, is it

this field, so to speak, you know, into your into your point,

or is it some other field, and then working through each of

those things as you really kick off that project? And, and, you

know, the problem is, you know, garbage in garbage out, if you

are using the wrong data set, totally, then then what you're

doing is worthless. If you're off a little bit, then you're

not really providing that little bit of extra value to that

business question that can really make the the value

overall to the company, really shine.

Yeah, you know, what, it's little things like,

back to your point of the definition, why it's so

critical. I had a recent scenario where, you know, you

would people would register with this company's website. Now,

register sounds pretty simple. There's a registration date.

Okay, great. Any analysis about registration, we go off the

registration date. In deeper conversation with the business,

it ends up that there's different types of registration

that aren't necessarily flagged as a type. But, you know, they

work with corporate partners. So some corporate partners will

quote, automatically register everybody at a soft register,

just give the name and the basic contact so that all someone has

to do is define a password, but they'd still have that

registration date is then so you get into Okay, well, when you

care about when someone registered, what exactly do you

mean, now? Is it when they showed up in the system,

including if their company just did the soft registration? Or is

it when they actually came in personally typed in information

and confirmed that they wanted to make use of that

registration? And again, there's not a right or wrong answer here

necessarily, but depending on the question, you know, the

question that they want to answer, it could make a huge

difference in the actual analytics that we do. And so

that was one I thought was interesting, because it sounds

so easy. Oh, there's registration? A, we're good to

go? Well, no, Yo, you always have to dig in. Does it mean

what we're assuming it means? And in this case, the answer was

no, not always does not always mean, you assume.

Yeah, exactly. And that kind of gets into maybe a

little bit of the the next question here is, so the

question is, are the roles that support data science efforts

changing? And it almost seems like what there's there's two

separate roles that we just talked about? One is defining

the business question. And then there has to be somebody that

peels the layers of the onions back to make sure that we get to

the truth, root question that we're really trying to answer.

Yeah, well, it's interesting. I mean, I've got

like entire keynote on this topic and written a bunch about

it as well. What's interesting is if you go back to when I

first got into the family business, and I think yourself

as well, you know, in the old days, I mean, if there was one

person, I would do have to do everything. I had to get the

data. I had to prepare the data. I had to model the data. I had

to work with the business folks. on the front end on the backend

all of that. Now granted, a lot of what we were doing wasn't

nearly as sophisticated automated, etc. But that's how

it was done. And what's happening in recent years now,

because Analytics has become not just more ubiquitous, but it's

becoming more embedded in operational processes. And it's

getting scaled. That rolls getting split, you now have

people loosely called translators who focus almost

exclusively on these conversations we just talked

about on the front end, what what exactly do you mean? And

what do you need to solve and making sure they understand that

from the business side, and then translating that to requirements

for the technical coders, and on the back end, those same people

would help position the results to the to the client. But now

you have a you hear about data engineering as a thing, right?

Well, data has gotten so messy these days, with all the

different formats and all the different locations. Now in many

cases, you have a whole discipline around just getting

that data together so that someone can model it. On top of

that, because of the deployment, than being so much more scaled,

you now have people these ops roles, ml ops, you know, for

machine learning ops, or AI ops, there's DevOps in general and

software, where these are roles are really systems people with a

focus on analytics who are doing classic system optimization, how

do we get the processing capacity allocated properly? How

do we make sure the data is going through the network

properly, you know, very technical, deep network stuff,

but focused exclusively on the analytical type processing. So

there's this a variety of roles now, where as you look across

time in an analytical process, the steps where we're, you know,

specializing within each of those phases, and sometimes even

within a phase, so within data science itself, there's no you

know, there are still generalist data scientists, and we still

need them. I liken that to a general practitioner doctor,

they can do the triage and figure out the general

direction. But now you've got people specializing in nothing

but language processing, nothing but image recognition, nothing

but you know, classic risk analytics in a banking context.

And, and it's because they become so complicated, and so

sophisticated, each of those areas that you can't just dabble

in anywhere, if you're going to be effective, and you're going

to deliver the quality that's required, you've got to be a

specialist much like doctors have. So I think it's both

across and within these disciplines. There's just a lot

of specialization these days, which, back to the talent

crunch. So now, you know, we need everybody that we used to

need plus a bunch more in each of those roles. So yeah,

no, absolutely. And I was on a call with a friend of

mine. And what they do is, we specialize kind of in the mass

media and the online media space. And they're specializing

more or less in kind of the CRM in the sales space. Both of them

have overlapping components. But there's certainly a difference

in the, in the analog and analytics knowledge that you

need, in order to be able to really, really get that last

extra little percent out of the out of the models, and and out

of and then really be able to answer those business questions.

And so I think you're right, when you said, you know, you've

got, you know, for each kind of a role, you have sub roles that

that have to deal with different pieces of it. I

think that's absolutely right. And you hit on

something really important about the different skill levels as

well. I mean, this is the core theme of my of my book, Damocles

revolution was about this embedding and automating of

analytics and operational context where, you know, think

about a website making millions and millions of decisions a day

in milliseconds of what you know, what are the products,

it's going to try and cross sell you. And those kinds of

scenarios, you know, you can't afford to have really expensive

people spending a whole lot of time on each individual model,

you have what I call, I kind of call commodity models, you need

an automated process that will build a pretty decent model

pretty quickly, where there's some automated checks. And if it

passes them, it just goes live, right. And there's people

monitoring the overall pool of these than looking for any that

are going to miss to to yank them out, potentially. But you

can't you can't be building bespoke super fancy models for

every possible application high value ones. Absolutely. But

marketing's an example where a lot of things just have to be

automated. And you can make the argument well, you know, we

could get another another, you know, relative percenter to lift

out of this, if we really put some data scientists on it for a

few months, well, true. But you've got 200 of the models.

And to put those data scientists on that for those months on each

of those is going to end up being timely and cost

prohibitive. And it's just not practical. So embrace what

you're capturing with your somewhat automated models, as

long as they're working well, and take down you know, the 10

of the 11% theoretical max that you're actually able able to get

I think that's been a mind a mind shift, both within the

business community and the data science community. Early in my

career, I would have had a heart attack at the thought of I'm

leaving one out of 11% on the table. Now I'm like, look if we

can take down 10% In a matter of a couple of weeks and move on to

another problem to get the first 10%, that's actually a lot more

value than spending the extra month to get the extra 1% On the

first problem. And so that's back to the scale, the scale

changes the equation, when there's only five problems to

focus on, you're going to take them all down to the, to the

ultimate level of detail, but we no longer have that now we have

dozens or hundreds or 1000s, depending on the size of the

company, and you've got to get in, get a good value and move

on.

Yeah, absolutely. And that, we call that the

minimum viable product, so to speak. And, you know, and if you

can get 80% of the value out of, you know, let's say, you know, a

handful of weeks work versus a handful of months or a handful

of years, you're much better doing that. And then just like

you said, move on to the next big one. And then, you know,

capture all the big ones, and then you can reprioritize and go

to the next level. The other challenge I have is, as well, as

you know, I don't know, maybe it was five years ago, maybe

longer, really, before data analytics and data sciences kind

of took off, you'd have these spreadsheets and you know, you'd

build a spreadsheet, and then you'd add a little bit, you'd

add a little bit. And all of a sudden, that spreadsheet is a

monster, and it has so much in it that it absolutely has errors

in it that you will never find. And so if you were even to build

some very complex data analytic solution, you don't know if you

have an error in it, because you, you know, it gets so big,

that you might actually, you know, you're you're still adding

value, but you may actually be doing some things wrong in

certain areas, because it's just too big to see and have a, you

know, a good view of actually how all of the thing puts

together, it comes together to really solve them, you know, the

the business challenges that it's trying to solve? You

know, it's intro, that's a great comment. And, you

know, I find a lot of people get hung up on that, oh, my gosh,

you know, we made an error in this decision here and that

decision there. And that's the whole point of statistics and

probabilities about odds, right? It's the casino business, right?

At the end of the day, when I build a model, no matter how

good it is, what we're saying is, we think we, on average, can

can make the correct prediction as to say, Who will buy, you

know, 70% of the time, whatever it is. But what that means is

30% of the time, we're wrong. And we could be wrong, because

someone just changed their behavior unexpectedly, it could

have been a data error could have been our model wasn't

capturing something that it should have captured. But it

doesn't matter. The point is, on average, are you capturing about

70% of those who respond? And if so, you know, that's phenomenal.

But you can't let the exceptions you know, obviously, you can't

let the exceptions drive the rules. And, you know, real world

example that I used to hammer this idea home is, you know, we

should be looking at on average, does this model work better than

not having the model and as long as it's not doing anything

harmful, you know, on the mistakes making, it's a good

thing. So autonomous vehicles are out there. And we still are

in this world now, where anytime an autonomous vehicle hits or

kills a person anywhere in the world, it makes international

headlines. And everyone calls for either completely stopping

autonomous vehicle creation or regulating it further and

further. And my point is, that's the wrong way to look at it. You

got to look at for every 100,000 miles driven, what's the what's

the accident and death rate of autonomous vehicles and people,

as long as the cars are worse, obviously, then we want to be

very cautious. But we're going to get to a day where the

autonomous vehicles are, say, 1/10 1/20 of people. But that

doesn't mean they'll be error free, there will still be cases

where an autonomous vehicle will wreck where most reasonable

people would because of when usual later, whatever the case

is, you have to look at that trade off of Well, that's

unfortunate this accident happened here. And we know how

to get the autonomous code to work better. But we're saving,

you know, 10 accidents per 100,000 miles driven over people

who would have made these other accidents happen that the car

wouldn't have. The problem is you never see the examples. That

didn't happen, because someone wasn't driving the car. But it

applies very much here in analytics in the modeling as

well. You could always go find those cases where somebody got

the wrong offer, or was given the wrong diagnosis, prediction,

etc, etc. But on average, overall, are you much better

than you were before you had any models at all?

Yeah, absolutely. And it's funny, my brother just

sent me a video of a Tesla was in an accident, and the battery

exploded and they were out of the car all over the place. And

and I I responded back, yeah, but don't guess tanks explode.

And so you know, is one it's this, it's exactly the same

thing. You know, here you have something that, you know, okay,

so batteries, you know, are saving energy, you are more eco

friendly or whatever. And, and, and yet, you know that one case

makes the news, whereas all of the other cases don't make the

news. I've done. We did some consulting for loss prevention.

And one of the challenges with with loss prevention is you

can't prove how much loss you prevented, because it didn't

happen and You know, and so to your point as well, is that how

many accidents don't happen because you have all this

automation in the car or the automated driver or whatever it

is? And yes, absolutely, there's going to be one or two failures

where the the light, like you said the lighting isn't right.

So yeah, I agree with that. 100%. So

it's funny when people are uncomfortable with

this conversation we're having, I always point out that while

it's it's a dirty secret, it's how all public policy is built.

So for example, car mandates for airbags, we didn't used to have

to have airbags, because they were way too expensive

initially. And they effectively that both the government and

manufacturers of all types effectively do a, you know, life

saved, or injuries saved per dollar of cost. And at some

point, it's too high $1 per cost, and even the federal

government Yep, Nope, we're not going to mandate that because

it's too costly per life saved. But they know that that that

there would have been life saving if the airbase had been

there. So you have to have, you have to have some of that

rational trade off. And it is uncomfortable. But at some

point, I mean, if we wanted a car that would never wreck, we'd

all be driving five miles an hour in a tank, and paying you

know, and having to fill our fill our tanks with 100 gallons

of gas every 100 miles, but nobody, nobody would want that

even though objectively, it's far safer, right? And so, you,

you you accept that risk? Well, I'm going to go on the

interstate, it's convenient to go 70 miles an hour to get home.

But at the same time, if I get in a wreck at 70 miles an hour

I'm in, I'm in deep trouble. It's a it's a it's a big, you

know, there's a big risk point there. And so I think that's

what the models, and I'll do, it goes on behind the scenes in our

lives a lot more that I think that many people think about it,

and people don't don't don't realize, when you buy life

insurance, you're basically it's a bet, you're betting you're

gonna die so that you make money on it. And there are companies

betting that you won't, so they can keep your money. But it's

literally that computation, your rates are set on the

probability, they think that you'll die before that policy

expires. And when you buy it that you know, in effect, you're

you're betting on your death, because the only time the

insurance psychologically poor estate planning, it's a safety

net, but I'm saying from a money and math perspective, you win if

you die to collect more than the premiums that you put it.

Yeah, absolutely. And you know that COVID is a

prime example of, you know, how do you get to perfect so that

there's no spread of the disease with with masks or no mask

vaccine mandate or no mandate, and it's kind of a very similar

trade off. In some things, you know, you can never get a get to

100%. Perfect. And that's where then public policy comes in. You

know, oh, but speaking of public policy, then let's switch over

to Data privacy, because data privacy is now becoming part of

public policy, and certainly with the GDPR and the CCPA. And

in California, and what have you, the the value of your your

personal data is, is now coming into the public policy domain.

And so when you think about that, and especially now as we

move to the web three Oh, and Metaverse, and, and what have

you, there's a lot of ethical issues, and that marketers and

data scientists need to consider, let's talk a little

bit about that.

I mean, that's a it's a it's a big issue. And you know,

you mentioned the getting the book 97 things about ethics,

everyone data science, you know, focus on this was actually a

compilation of blog link submissions from, as the title

suggests, 97 different submitters, on various aspects

of it. And the reality is, you know, I still I've been in this

business for a long time. And I like to say, I've become

somewhat of a privacy freak, mainly because honestly, what I

do, it's like, I'm on the inside seeing what's happening. And

it's not always what's happening. But what could

happen, that bothers me more, right, I'll see the data company

has, and I'll know what the time they're not doing something I'm

uncomfortable with. But I also can identify 10 things they

could do with that data that I'd be incredibly uncomfortable

with. And the only thing stopping them from doing it is

their own sense of, Well, this would be illegal, unethical, or

otherwise, get us in trouble with the media. And then there's

companies that push those limits all the time. So I think this is

a topic that's going to be evergreen for a while. I mean,

when you when you look today, in the marketing space, and the

talk about getting rid of cookies, well, you know, people

on the one hand are cheering that this is great. Now, these

cookies, they can't track me the same way. But you know, they

postpone that once or twice because they're getting their

work arounds. And you hear about, you know, browser

fingerprinting, which is, you know, a way as I understand it,

you know, this might not be perfect, but your browser itself

reports information. When you request a web page, it'll tell

the browser version, it'll say what ad ends you have, etc, etc.

They might know things about your IP address. The point is,

there's ways to almost uniquely identify people just based on

the web request that they put in. That isn't isn't unique

because there's a cookie now it's even worse. It's your

computer and browser combination is uniquely stamping you in a

way that can be identified. I don't think most people realize

that that's possible. And the law is in the in the policies

around what's fair and unfair. With with the browser

fingerprinting, I don't think are fully developed yet either.

So everywhere we go, we're going to keep having this and you've

mentioned you know, things like this new this whole idea of

metaverse. Now, you know, it's one thing when everything that

you type or click is getting tracked. But now you get into a

Metaverse concept. And let's say you're literally having a 3d

avatar interaction with somebody. And you know, I

jokingly slap you in the 3d world. Now, is that is that

count the same as if I slapped you in the real world? Right? Am

I going to get in? Am I going to get in trouble for this? Am I

now violent? Am I now? You know, a person to suspicion because I

slapped your avatar with my avatar? I mean, I don't know.

But the point is that it that data is going to be captured and

much like people getting in trouble now for things that were

harmless 10 years ago and getting in big trouble because

it surfaces that they had done or said something that was

perfectly acceptable at the time of it now is it? Maybe today

it's a big joke, let's all go in and start slapping each other's

avatars and 10 years from now, that's considered equally bad as

an assault. No, go back, go. Look at Bill. He was slapping

and assaulting people in the metaverse in the early days left

and right. And get cancer. So you know, what is the policy on

that? What what what data should be captured? And? And how long

should it be kept? I mean, it's just it's a never ending

question. So I guess answer your question. I'm concerned about

the lat still about the lack of general focus by the average

person on this, I think most people just blow, they would

say, Oh, well, you guys are just old curmudgeons what do we care?

Who cares if they have my data, I'm not doing anything wrong.

And you know, yet, you hear about, you know, college

recruits losing a scholarship. And in Division One thing

because of one thing, they said one time on social media that

might not have even met, what it read as if it was in context.

And so I say, you know, it's all fun and games until you're the

person who gets burned, because your data has been captured and

analyzed in ways you didn't want it to. So I'd rather have the

transparent at least make sure I know how it's doing at least

make people acknowledge what's being done. And if you choose to

turn over all your data willingly, okay, that's your

choice. I just want it to be transparent, where people are

being made aware of exactly. In plain English, not these 80 page

documents in plain English, here's what we're going to

collect and do with your data.

Yeah, and that is, that's very difficult. And to

your point, as well, i The problem with public policy, or,

you know, regulation or whatever. Is that, okay? So now,

you know, the metaverse is kind of starting to take off, it's

been around a while with Second Life and a couple of other ones.

But now it seems like you know, might actually be taken off. And

the problem that public policy has to even regulate it in some

fashion is that they're always going to be five or 10 years

behind, by the time they finally figure out but there's a policy

that needs to be be in place, like no slapping on your first

date or something like that. Then it you know, the metaverse

has already moved on. And then there's some other kind of

adverse out there. So you know, it's it's very challenging, but

I did like your breakdown, you know, there's there's things

that are illegal or criminal, illegal or criminal, then

there's things that are ethical, or not ethical, and then there's

things that the media might get hold up, I kind of like that

breakdown, because that is exactly where companies and

their data and their policies and their internal actions

really have to take, you know, really have to play. And, you

know, the biggest one that I can think of is the Volkswagen

diesel gate, where they were manipulating the engines during

the, you know, the EPA tests to pass, so you'd get a good pass.

And then once that was done, they'd go back and, and then you

know, and then run the motor differently. And, and, you know,

clearly unethical, certainly bad for, for the media, I don't know

if it was illegal probably was illegal. But you know, those

things that culture within the organization is what drives that

use of the of the data. And in this case, then the use of you

know, that algorithm that's built into the computer that's

controlling the the diesel engine. Now in that hierarchy, I

always say in an ideal world, all three of those

would be lined up perfectly. But what's legal is probably the

loosest because to your point, things just haven't caught up,

right? There are things that are that that are not illegal today,

not because 90% of people wouldn't say that should be

illegal, just hasn't yet been recognized as a possibility to

be made illegal, then what's ethical, I think is tighter than

what's legal, certainly. But even what's ethical, you know,

one of the things I talk about ethics all the time is it's not

as cut and dried as you think So even something that you're

convinced is ethical. There's nothing that 100% of people are

going to agree is ethical. And so depending you do something

that seems perfectly ethical, and that the majority of people

might agree is ethical or whatever. 30 year customers

think it was totally unethical and they're the ones running to

the papers and doing the boy caught you're going to feel some

pain. So you've got to really think about all three of those

and that's where the I think the we sit today for the most part.

It's the culture and the individuals and the company.

policies that guide this, and there are companies I tend to

trust. And there are some large tech companies I don't trust as

far as I can see who I who I, you know, I believe skirt up to

and maybe over the ethical and legal lines on a regular basis

wherever they can that that's good for them and that, you

know, that's life. But I back to it. I wish I wish people would

be more aware. And frankly, I remember a shocking statistic I

saw was for all the data you get, let's I think it was

Facebook was this example for all the data you're giving them

to get the free service, the amount of ad revenue they made

off the average person use either per month or per year was

like five or $6, something like that. Probably per year, given

how big the revenues are, the point is, I would happily pay $5

a month to have a a some of the services I have that are free if

they'd be kept completely private. So it's one thing if

the if someone said, well, to get all your benefits of

Facebook is going to cost you $3,000 a year, a lot of people

might say, Well, I'm not going to pay 3000 I guess I'll just

have to give up my data. But you tell people you might pay around

5080 $100 a year and then Facebook won't have collect or

use any of that data. I think a lot of people go oh, geez, yeah.

If I'm giving up all that, and it only cost me that much. I'll

do it. And no, that's an alternative revenue model. What

is honestly, I don't see why Facebook would care if I'm going

to pay them their same ad dollars they make off me

otherwise, to keep my data private. That should be as well,

I'd love to see some companies actually go down.

Yeah. And I like your the way you're, you know,

connecting the privacy issue with what the value of that is

to the individual. And not that I think it's kind of a way to

look at a lot of these privacy rules and say, Yeah, I paid I

pay $5, or no, you know, I don't care if they take my data. And I

hate to say it. My wife hates it when I do this, but I'm a

marketer. And so I leave my cookies on and I leave a lot of

stuff on because I want to see how the marketers use it. And so

we just bought a new car. And so we were going out to the sites

Hyundai and Kia, Toyota and Ford. And it was fascinating to

see how quickly the manufacturers then took

advantage of that, and started showing us ads on our smart TV.

And it was Hyundai that was the winner. Now GM was a little bit

behind. But they were Neverland, us less their key of Ford and I

can't remember who else we looked at didn't do anything.

But Hyundai within a day, within 24 hours, we're using my

information and starting to show me ads based on the websites

that we have visited. Wow. Yeah. Yeah. So fascinating. It kind of

also kind of leads into the next question here, because one of

the challenges that you also have, and then some data sets

are siloed, because of the regulations around them. And,

and I remember going to a presentation, our this is maybe

10 years ago, and he was the chief data officer, I think for

the state of Georgia. And he said, Well, you know, what might

be available to you in this data sources not available to you

from that data source. So you'd have the Department of Labor

would say no, no, you can't have that data. And one of the other

departments would say, yeah, you can have this data. And in

particular, then you had the crossovers where you had

specially corrections officers, you didn't want that data to get

out in any one of the databases. So you had a database over here,

let's say was the Department of Labor, they had to have a flag

on that data that said, Nope, you can't allow any corrections

officers data to get, you know, sold or given out to the public

for whatever reason. And you know, just kind of fascinating.

And that then leads, though, kind of to not only data silos

for regulatory purposes, but also data silos for managerial

or power or political purposes within an organization. So maybe

talk a little bit about how you see that happening, and how, you

know, maybe data sciences can break those silos down, maybe

can't break them down, or maybe can break them down and in maybe

some kind of a white room or

whatever. You know, this is one of these things

where I think most people would agree that in an ideal world,

you just have all your data together. Right? And I think

we'd all start with that premise. But then there's always

the reasons why you have to break them, and you just raise

you, right? Oh, well, we have to have it all together. Except

here's this piece that for law reason we can't have together

now you just created a silo albeit for a good reason. Well,

you know, we've got to give this data to our partner over here,

but they're not allowed to access our actual system, nor

can they see some of the data. So we need to make an extract of

the data. That's just the data they can see and then put it

over in this other spot where they have access to and so

forth. And so I think the reality is that the reality

unfortunately, is that there's a lot of valid reasons why you end

up with silence, unfortunately, more than you like, but then

people pile on invalid reasons just because hey, you know what,

I'm annoyed that you're not getting the data back. So I'm

going to create my own database over here and pull all the data

for my own purpose, that's not necessarily a good, you know,

good reason. But it's going to be there. In fact, it's funny,

my blog last month was about, you know, you know, barrier to

scale boy analytics is the same one, it's been around for a long

time, when you have data in different places, if you need to

analyze that data jointly, you have to bring that data together

physically at the time of analysis. And if they're small

files, Excel spreadsheets, who cares, right doesn't matter. But

if you've got terabytes over here, and terabytes over there,

even today, that's still an expensive and or time consuming

process, right. So even on public cloud, you can pay for a

huge amount of capacity to suck that data over as fast as

possible. But you're going to pay a really big price for that

much capacity. Or you can pay for a more typical reasonable

amount of capacity. And it's going to take a really long time

and still run up a pretty big bill. So I think that the

challenge is to be aware of that you'll have to have some silos,

and then just be very diligent in trying to keep them to the

absolute minimum. And to the extent that there's huge large

data repositories that are going to be joined together

frequently, you're going to, you know, as much as you can get

them as close together, if not in the same, you know, overall

platform as possible.

Yeah, no, that makes a lot of sense. And I think

you're right there, data silos are going to be there. And, you

know, sometimes you can, if you get full access to it, or you

can do a full join to it. Or maybe you need anonymized

access, where maybe you get anonymized access, but you can

get a couple of fields with real detail. I think that that's one

of the and maybe that gets back to the new roles for data

scientists, one of those roles for data sciences is, is that

data governance piece to be able to make sure that the data is

properly, legally, ethically and media, I guess, being used so

that it's not being abused or misused in some fashion?

Absolutely. Yeah. That's a big lot of big companies spending a

lot of time on the on the governance and oversight of the

data itself in it, and it's useless. Yeah, yeah, absolutely.

Alright, so now I have a future question for you. Yeah, when I

was growing up, and I don't know about you, but my first

programming language was basic. And then I moved to Fortran. And

then it moved into Pascal, I think, and then it was, then,

you know, there were a handful of other ones. And then there

was C and C++ and C++ and C sharp, and then all of a sudden

programming moved into things like R now for especially for

analytics, and, and Python. And so it seems like Python is now

supplanted R. So what do you see is kind of the future of these

languages for for analysts?

Well, it's interest. So I guess, I'd be

hesitant to project because there, there could be language

10 out there that that you didn't mention that suddenly

going to take the world by storm, I mean, even, even not

too many years ago, it looked like art was gonna conquer it.

And then Python kind of came out of nowhere, at least in the

analytic space. But I'll tell you, in my mind, it doesn't

matter so much from this room. This is where I tell I tell this

to I mean, students particularly talk about this a lot. But even

professors will come go, Alright, Bill, know what

language should I know? Right? If I, if I want to get out there

and get the best job, should I know are? Should I know SQL?

Should I know? Python? What should I know? And I always say,

You know what, what you need to know is know how to program. You

need to know the logic and how to develop how to first define

analytic logic and translate it into code and do it well. And I

said, if you know one language really well, and you can show me

you can translate complicated analytical logic in that

language. I have utter confidence, because like you

just mentioned, you and I have translated in multiple languages

over the years, you can try, you can transfer that much like

speaking English. If I wanted to learn French, it's painful. But

I know exactly what I need to say. I just have to figure out

how do I say it in French, that's different than when you

were a baby. And you had to learn what language was and what

a word was and what a sentence is. And when you first learn

coding, it's a little bit like that you have to learn the

entire concept of coding incredibly, incredibly difficult

at first, but once you know how to code and you know, one coding

language to train, it's just a matter of translation. So I

always tell people, instead of trying to get a little

certificate in seven languages to claim you know that, but you

know about that D, show that you really know how to code. Well,

if you know how to code well, and I'm using a language in my

company, that's not the one that you know, I know that within a

couple months, you'll you'll pick it up in particular,

because your peers will already know that language and be able

to help tutor you along you're not going to be on your own. So

to me, it's it's really about the underlying logic in the

code. It's not even about the coding

language. Yeah, no, fair enough. And actually, you

know, good point. I don't know, when I was, in my programming

classes, I think they were trying to teach us kind of the

principles of coding. And, you know, so you kind of get that

structure. And, you know, I'm wondering if that's kind of the

same thing. Now, as you know, you need to go deep in at least

one so maybe it's our maybe a to Python, but then you also need

to kind of understand the overall theory because at some

point, Python is going to be supplanted by something else,

whatever that is. And you're going to have to be able to

translate your thinking and your methodology from Oh, you know, I

used to do it this way in Python, but now I need to do it

slightly differently, hopefully better in some new language.

Yeah. And you hit on it, too. It's I remember I've

learned basic first as well. I taught myself basic initially.

And I don't get too old. Yeah, well, and but the thing is, if

you look at it, if you go back and look at old basic code, to

be honest with you, a lot of the constructs are still present in

all these languages today. There's if and then there's, you

know, there's loops. I mean, is it a Do Loop a while loop or a

for loop? You know, I don't know, it depends on the

language. But what do they all do? They're all going to churn

through from one to 10. Right. So yeah, I think it's a to me

that I like your idea of even the foundation of coding when I

like students on the projects in the project classes, which I

tell them I say before you start coding, if you just go and start

coding, you're going to screw it all up, I promise you, before

you start coding, and there's sometimes teams where different

people on the team know different languages, right? I

said, What do I call pseudocode? map it out on the on the

chalkboard? What do you got to get to, and then you can split

up who's gonna do what piece, let them use whatever language

but you want to write down that logic and be convinced you have

the logic laid out before you code. Because once you start

coding, now you're locked into trying to fit it within what you

know in that language. And you're going to end up doing

things maybe you that weren't optimal for the problem back to

the very first point for the business problem. It's not

optimal, but it's how you know how to code in your specific

coding environment. So you do it. That's not optimal layout

what you need, and then figure out a way to do

Yeah, exactly. And so, so true. So true, that

definition of that business question at the top, and really

peeling back the layers on that. And then, and from a coding

perspective, is really outlining what your code, you know, the

big blocks of your code are going to do. And then you know,

and then going off and doing the code makes makes so much sense.

The thing I hated about coding was those those off by one

errors, I was always I always had that in there. I never could

get around them. But anyway, we're about out of time. Is

there one other thing that you'd like to talk about? Or maybe you

know, what is the future of data sciences, and then we'll close

up?

Well finish, I got to do a completely shameless

plug, guy, I'll just follow. Just this week, my new newest

book came out, you mentioned the game, Winning the Room: Creating

and Delivering an Effective Data-driven Presentation. And

what this is all about is, you know, over the years, I learned

a lot of hard lessons myself, but probably some of the most

painful meetings I've ever been in have been technical people

presenting data to typically non technical audiences, sometimes

even other technical audiences. And it goes horribly wrong,

because they they can't put it in terms people understand

they're over complicating it, it's to detail all of these

things. And so I tried to distill this down the book, it's

a little different. There's books on storytelling, there's

books on analytics, there's books on visualization, this

book is about a live presentation. Imagine you're in

front of a room, putting up a PowerPoint, what do you have to

do to make that be successful? So there's elements of

storytelling elements of visualization and such, but it's

really focused on that, how do you distill it down to a live

presentation, to a often non technical audience. And, to me,

as we continue to have all of these analytics, it's still as

marvelous as much as a problem as ever before. And you know, I

was on a call earlier this week, with a company I'm advising

wherein their client was concerned that this model they

had been delivered wasn't what they needed, they couldn't

understand it. And they weren't able to work with the data

scientist involved, to have that person help them understand, you

know, age old problem. And so the net result is maybe the

model was perfect, and maybe it was horrible. But it doesn't

matter if the clients not even understanding it, and they can't

be explained. And so I think that's where this this this,

this book was, was built. As I started teaching here at the

university, realizing, often because they hadn't had any

lessons, it's how bad the students initial presentations

were often but then how fast they improved was coaching. And

I see I got to put, I got to put some of this in a book. And it's

119 tips, just a minute or two each 140 illustrations, a lot of

the tips of illustrations to kind of show here's right,

here's wrong, very easy to digest. But I think, you know, I

think people will get it because as we move forward into the

future, it's going to be as important as ever able to

communicate these these analytics and these data driven

decisions and so forth. Yeah, effectively.

Absolutely. No, I'm looking forward to getting it.

And so tell us where can we buy it?

Everywhere where books are sold, it's I know,

it's up on Amazon. It's up on Amazon, Barnes and Noble, the

Wiley site directly. So it should should, at least online,

it should be available bookstores. I don't know, I, you

know, bookstores have their own method of choosing what books to

make it in. I'm sure you could order it online at any

bookstore, even if they didn't have it in stock in a local

store.

Alright, so why don't you show that again, it's

called Winning the Room. And what's the subtitle,

Creating and Delivering an Effective

Data-driven Presentation.

Fantastic. And I, you know, I've done data

presentations all the time, and I do them now. And you know, and

sometimes you run out of time, and you just make more mistakes.

And so we're definitely looking forward to that. So I can

mitigate, remove some mistakes that I made. So but anyway,

Bill, thank you so much been awesome. The conversation, we

could keep on going, I guess the hour but really appreciate you

participating, and really appreciate, you know, your

perspectives on data sciences and where things are going to be

going and the certainly the challenges that we've, we've got

today, definitely please go out to on Amazon or otherwise to

purchase winning the room, Bill's new book, I'm sure it'll

be great. You can also reach out to bill at Bill dash Frank's dot

com Bill dash Frank's dot com, and I'm sure you'll find you'll

be able to find more information there. Otherwise, please stay

tuned for many other videos in this series of the backstory on

marketing. And please visit marketing machine pro

relevant.com/getting started a mouthful, pro marketing machine

pro relevant.com/getting started and you will also be able to

download the first chapter of my book and other valuable

excerpts. Thank you so much, Bill.

Yeah, thanks for having me.

Absolutely. Thank you.