How to Create and Manage a High-Performance Analytics Team
3 Video Views·Dec 12, 2024
Data science and analytics teams are unique. Large and small corporations want to build and manage analytics teams to convert their data and analytic assets into revenue and competitive advantage, but many are failing before they make their first hire. In this session, the audience will learn how to structure, hire, manage and grow an analytics team. Organizational structure, project and program portfolios, neurodiversity, developing talent, and more will be discussed.
My name is John Thompson.
I've been involved in data, advanced analytics, artificial intelligence for 37 years.
And I have made every mistake possible when building and managing analytics teams.
So that's the reason I wrote my book, Building Analytics Teams, that came out I think three
years ago.
Three years ago, nearly three years ago, I guess it is now.
I wrote it because I wanted to really help the community.
It was one of those things that I went looking for books, I tried to find them, I talked
to publishers.
I said, look, there really isn't any books on building and managing high-performance
analytics teams.
And the response from the publishers was, yeah, there are no books because no one's
interested in it.
And I thought, that can't be, that's not possible.
So I sat down and I cranked out 100,000 words in three months.
And I started shopping it around to publishers.
And I think it was the second one that I hit, so it wasn't really hard.
It wasn't like J.K. Rowling, rejected by 35 publishers or whatever it was.
And they were really cool about it.
They said, we like your book.
It's got a lot of personality.
It's very practical.
It's pragmatic.
It talks about success and failure and all that kind of stuff.
So this talk is based on that book.
So if you're interested, you can buy it.
I'm not pitching it.
I'm not making a commercial for it.
It's a best-selling book.
It was when it came out, I'm a numbers guy, of course, so I went and looked.
In the first year, it was in the top 5% of all books sold on Amazon, which is stunning.
And I started looking around, and I thought, who's buying this book?
Who are my people?
And obviously, you're here, so you're my people, obviously.
And it's folks like you who are new to the game, first analytics team they've ever managed.
Or another subsegment of the population was students.
A lot of kids in undergraduate bought the book because they heard from Tom Davenport
that data science was the new sexy thing, the sexiest job of the 21st century, and they
wanted to know, well, what am I going to do if I'm a data scientist?
So a lot of those people bought the book as well.
So all right.
Let's jump into it.
Oh, the other thing that I said to about half the room early, about 10 minutes before we
got started, was this is really a session, a discussion.
No slides.
So I can't predict what you're thinking about, and I can't read your mind.
So whatever you want to talk about, ask the question.
So I can stand up here and talk.
I think the longest I've ever stood in front of a room and talked nonstop was five hours.
So I can do it.
So there's a lot to talk about.
But if you have a question, let's get it out.
Let's have a discussion about what you're having a challenge with.
Yeah, go ahead.
I will.
As a matter of fact, yes.
No, no, no, no.
That's a great question, and it's one I get all the time, and we will talk about that.
Absolutely.
And for some reason, I'm not hitting it.
Just raise your hand and say, get to my question.
Yeah, absolutely.
Great idea.
Her question was, could you talk about increasing the velocity of execution in analytics teams?
And I will talk about that, and it's something called the Personal Project Portfolio, PPP.
And every time I've done it, and I've built five enterprise-scale analytics teams at this
point, every time I've done it, every data scientist that's been part of it has all
grumbled, complained, and just, well, I won't use the word, complained vehemently about
it that it's bad, and it's going to be hard, and it's going to make their life tough.
And after three or four months of doing it, everybody's come back and said, I love it.
It works better for us.
So we will talk about that.
So let's start out and talk about where should a data science team fit in an organizational
structure?
I've seen it report to every C-level executive possible, CFO, CIO, CTO, very rarely the COO,
almost never the CEO.
But where do you think that team should live?
Well, in my opinion, that team should report to either the COO or the CEO.
And why do I think that?
Because high-performing, high-results analytics teams are really a transformation function.
If you go in and you analyze a capability of an organization, or how customers act,
or how you acquire customers, or how a process works, or the drug interactions, or whatever
it is, you're generally going to find that however you're doing it at that time is suboptimal.
You are going to find a way to transform and change for the better.
Now, how often have you seen the CIO or the CFO go out and have the political capital
or the wherewithal or the interest to redo a manufacturing plant?
Anybody?
No, it doesn't happen.
So you want your data science teams to be feeding the information to someone at the
C-level who has the wherewithal to drive change.
That's why I think that data science teams, they don't have to report directly to the
CEO, but they should be in alignment with the CEO.
The chief analytics officer should report to the CEO, in my opinion.
How many people think that's a good idea, bad idea?
Show of hands.
A few people, all right, great.
So how have I structured analytics teams?
Now we know where they should fit in the organization, how do we structure them?
There's two main modalities, in my opinion, in how you structure an analytics team.
One, I call an artisanal approach, or artisan approach.
The other side of the spectrum I call a factory approach.
So the artisan approach is where you have data scientists that can do everything.
They acquire the data, they clean the data, they do feature engineering, they build the
models, they validate the models, they interact with the stakeholders, and they help move
the models into production.
One person.
They do it all.
Those generally are unicorns, as we know.
They're hard to find, and they cost a lot of money.
In my last job at a company called CSL Bearing, I think I had about 10 of those people.
And what I found is that we could do a significant amount of work, but it was limited.
I mean, 10 people, that's not a very big team.
So then what I came up with was the other side of the equation was a factory approach.
That's where you have a number of different people, and a group of people does obtaining
of data, they do cleaning, another group does cleaning, another group does integrating,
that kind of stuff.
So what we did is we took the factory approach, and we had them do more of the repetitive
work.
You know, all the early stage stuff.
And then we fed those objects up into the artisanal team, and then we could do more
and more work.
So what I ended up with was a hybrid team.
I had both.
All right.
I'm going to take a drink of water, excuse me for a second.
Now we get to her question.
How do you increase the velocity of execution in those teams?
And the concept that I came up with was a personal project portfolio.
And when you first hear about it, it does sound a little crazy.
So the way I did it was that each data scientist, and now I'm talking about the artisan data
scientists, not the factory data scientist staff.
The way I did it was that each data scientist had probably two or three major projects.
What's a major project?
That would be a project that was strategic in value, connected to a business function,
manufacturing, marketing, supply chain, something like that.
And it was probably going to take somewhere between nine months and two years to complete.
So it's a big project, a lot of effort, a lot of modeling, you know, a great deal of
work to be done there.
Then in addition to those two or three projects that they had there, then they had a raft
of smaller projects, and it was maybe one to four smaller projects.
And those are things that they could do in three months, four months, something like
that.
And then, of course, I think we all have had this happen.
We all have it happen all the time, is there are fire drills.
So I could almost predict with great certainty that a month before every board meeting, the
EA of the CEO is going to show up and say, for this board meeting, I got to know this.
I got to know the price elasticity of this group.
I got to know why this manufacturing plan is producing less product than this.
I got to understand why these donors aren't coming into the center.
So we all have fire drills.
We all know that is a fact of life.
So each data scientist had major projects, minor projects, or smaller projects, and then
they had fire drills.
All right.
So, you know, what, why would you give someone that much work?
And what is your job as a leader to make that work?
You have to be able to be, as a data science leader, you have to be connected very well
with a line of business executives.
You know, you have to, you have to know the, the VP of manufacturing, you have to know
the CEO, you have to know the CFO, you have to know the head of the supply chain.
So, you know, you, you as the leader have to set the expectations, you know, because
everybody who's had this experience, someone who is a non-technical executive will show
up and say, you know, I'd like you to look at the throughput of the five manufacturing
plants, all with different technologies, all with different approaches, and come up with
a prediction on, you know, which one's going to do the best over the next 10 years.
And can you have that done by this afternoon?
Yeah, we've all heard that.
But you know, if you, you're the data science leader, you're connected with those people
and you can explain to them why that's an absolutely insane request, in nice language,
you know, you have to be cool about it, you have to, you know, connect with them.
So what you can do is you can say, yes, we can get you that.
We can make that happen.
But that's not a today request.
That's a year-long request.
You know, and their face drops and they all look, you know, stricken.
But you have that conversation and you, you set the expectation, and then what you do
for your team is you give them the license to focus on hard problems.
I always told my teams that if we weren't failing, utterly failing on about 15% of our
projects, we weren't trying hard enough, all right?
We weren't pushing the envelope to get new data, integrate data, try new algorithms,
and try things that we just didn't know if would work, all right?
So you have to, you know, tell people that there's going to be failure in there as well.
So what happens when you give somebody this corpus of work?
What happens is they feel safe.
They feel comfortable.
They feel like they can take risks, all right?
Then what happens practically is that we would have meetings every week with, I would meet
all, with my entire team every week.
We have a weekly readout on projects.
So in the early days, you hear from the project, I'm getting the great data.
I'm having all this success, and the subject matter experts are showing up.
They're all engaged.
We're having really good conversations, and that goes on for like a month or two.
And then you hear nothing, you know?
They've run into a brick wall.
Either the algorithm isn't working, or they can't get the, you know, the right compute
to work for them, or something's happening.
But then if you let it roll for a while, you know, what happens is they come into the next
meeting, and they're all fired up.
It's like, I was out riding my bike, and I had this epiphany, and I inverted this, or
I did that, or I integrated this new data set, and now it works.
So we had multiple projects at CSL Bearing, the last job that I had, where we actually
transformed the business because of this approach to data science.
So you give people a bunch of work, more than they could ever think they could do, but you
give them the right timelines to do it.
You give them the license to experiment and fail, and then you let them run.
And what happens, you end up with almost magical results.
We had CSL, you know, I like to talk about them because I can't really talk about much
what I'm doing right now.
I'm the global head of AI for EY, so a lot of what I'm doing I can't discuss, so I can
go back and discuss that stuff.
At CSL, when the pandemic hit, it's not intuitively obvious, but all of CSL's work is based on
human plasma.
So if donors don't come into the centers, no raw materials.
So when the pandemic hit, you know, the donor visits were going like this, 45 degrees down.
So we as a data science team, or the advanced analytics team, or the AI team, however you
want to say it, knew that was a problem.
So we started working on it, and we came up with three insights, three changes that the
business needed to take.
And when I went and talked to the executives about it, they pretty much laughed me out
of the room, like, nah, nah.
So when it had gotten so bad that it was a crisis proportions, they came back and said,
what did you say we should do?
So we revisited those recommendations.
They implemented all of them at the same time, which is kind of crazy.
And it reversed the trend.
Now, it could have been the time, it could have been, you know, all sorts of different
things.
We can't all just say that, hey, what we did was 100% responsible for that.
But not only did it reverse the trend, we not only recovered all the donors we lost,
we added another 50% to it.
So it was a huge difference in the business.
And a lot of that was due to the creativity and the ability of the team to look at problems
from outside the box.
All right.
So, so far, we've talked about, I've talked about where you put the team.
We've talked about the structures of the team.
We've talked about how you give them work and schedule their work and give them, you
know, opportunities to fail.
So I'm going to take another drink.
And let's see if any other questions come from the audience.
Okay, sure.
Let me grab a drink.
The question is, all right, I've advocated for teams under the CEO and the COO.
Why not the CIO?
And in this, if there's any CIOs in the room, I apologize.
You know, I'm going, I'm going to offend you.
You know, the problem with being under the CIO is that the CIO at this point in time
is really someone who's trying to keep the lights on.
Now, there's, there's, there's exceptions.
There are exceptional CIOs out there.
Not everybody fits into what I'm saying here.
But most CIOs are concerned about keeping SAP running, you know, keeping the CRM on,
or outsourcing, you know, all the on-premise hardware to a cloud provider.
They're generally not thinking about strategic change.
They're not change agents.
They're not transformational.
So what happens generally with a data science team under a CIO is that you end up working
on things that are rather mundane.
You know, you start looking at, okay, all right, can you guys and gals do the analysis
to understand what the hosting cost is compared to the on-premise server maintenance cost?
I mean, come on.
Do any of us want to get up in the morning and do that?
I don't think so.
I mean, and it's not really analytics.
It's, it's arithmetic.
I mean, I have a 23-year-old daughter and a 25-year-old son.
I can guarantee you they could do this in the span of a couple hours.
This is not something for a data science team to be doing.
So my point is, and this happens to people who work for the CFO too, is that you end
up as a data science team, if you're in the CFO's organization, you end up working on
the quarterly close or the monthly close cycle.
They always end up coming up and saying, okay, you know, we got to have this analytic done
because the books are going to close next month.
What does one have to do with the other?
But that's how people think.
So my point is, you end up being subject to the bias of the executive's organization
that you're under.
Could be the CIO, could be the CFO, could be any of them.
So that's why I advocate for a data science team to be as high in the organization as
they possibly can.
Hi.
So when I've hired on my team before, you know, the SQL phone screens and the take-home
challenges and things like that that I used to do, or I guess still do, those things could
very easily be spit out by generative AI right now.
So, you know, I'm starting to think more about what are some of the other core competencies
that I should try to build into my interview process and screening process.
Do you have any thoughts on that?
Yes, I do.
Did everybody hear that?
Or do I need to repeat it?
Screening process and hiring because gen AI is changing everything.
It's basically it.
So the way we did it was that, you know, we worked with talent acquisition or the way
we have done it is we work with talent.
I still do it today, as a matter of fact.
Work with talent acquisition to get the word out on who you're looking for.
I spend a significant amount of time myself writing the job descriptions.
I don't outsource that to anybody.
You know, I sit down and write the job descriptions because I know what I want my data scientists
to know and do.
So, you know, we get the right people in the front of the funnel and then we have the talent
acquisition people talk to them.
That's the first screen.
Can you get past someone that just wants to see if you can have a conversation, basically?
And then they come in and they talk to a data scientist, usually a principal data scientist,
one of the seniors.
After that, they go through a panel interview with probably three data scientists.
And then after that, we give them a set of practicums that they have to come in and do
in front of us with no terminal.
Pen and paper.
All right.
And then after that, they get to meet me.
So that's how we hire.
Is that akin to what you do?
Yeah.
Well, yeah.
So this is another thing that's interesting is that you talk to people and they say, well,
you know, we're an R shop.
Hardly anymore.
You hear that?
Or we're a Python shop.
Or even worse, I'm a SAS shop.
How many people are forced to work with SAS?
You know, I don't think as an analytics team that really holds, you know, because in the
past, I've had people who were generally data scientists in my shop are either their
Python adherence and they do a little bit of R or their R adherence and they do a little
bit of Python.
So I don't really see that dichotomy anymore that people are either or.
You know, it used to be, oh, I work in only Statistica or I'm a SAS person or whatever.
What I see now is there's a hybrid set of skills.
So, you know, what we're looking for is, you know, is someone curious?
Is there, are they honest?
You know, can they work diligently?
Do they get along with people on the team?
I mean, that's a big deal.
I mean, isn't it for everybody?
Isn't that a big deal?
You know, if you hire someone who's a jerk, I mean, oh my goodness, how disruptive is
that?
It's, I mean, in my opinion, it's been one of the worst things, you know, hiring the
wrong people.
You know, I would rather hire someone and then find out in the first month or two that
they're not a fit and then just, you know, say, look, this isn't working.
You know, we're going to, we're going to have to move on, go somewhere else.
We're going to have to go in a different direction.
However, you have the conversation because every time I have overridden that, that analysis,
that understanding, that realization, it's always bit me in the behind.
Every time I thought, no, it's, they're going to be good.
It's going to work out.
You know, they're going to come around.
They never do.
So, you know, as much as we want to hire for the hard skills, you know, the R, the Python,
the critical thinking, the data, you know, the data acquisition, the data engineering,
it's mostly the soft skills that will sink you to tell you the truth, you know, and it's
just something that we as a group of analytics professionals really don't do a good job on.
And that's just because we're wired that way.
You know, I think it was over here is Justin.
Yeah.
Justin and I were talking before I got up here and started prattling on that, you know,
there was, there's been some, you know, a difference.
There's a change in the workforce that we all know it's called working from home.
And, you know, and they're all trying to get us to come back into the office at this point.
And what I told my leadership at CSL was that this may not be popular, but my data scientists
are never coming back to the office.
And they looked at me like, whoa, no, whoa, whoa, whoa.
We got a three day a week policy return to the office policy.
And I said, all right, over the last two and a half years of all the applications that
we have built, how many have just fallen flat?
The answer was none.
I said, my data scientists love working from home.
They all know how to use teams.
They all know how to use zoom.
They can call each other on the phone if they want.
They know how to collaborate.
Another thing that we did at CSL that you guys all should do is that you should get,
you should make sure that your data science leader is well connected with the executives.
We've already talked about that, the line of business executives.
One of the things that I demanded, well, two things I demanded was that we would never
do a project that involved Excel, number one.
Because Excel is not a database.
We're not going to work with that.
Nothing against Microsoft.
It's a great tool.
But data scientists don't use Excel.
That's all there is to it.
And the other thing that I demanded was that if we were going to do a project for you,
say you're the head of manufacturing or whatever, you were going to assign a senior manager
or a subject matter expert who was going to be at every meeting.
Now, I know, you know, people get sick, they quit, whatever.
But the commitment from the business was they were in the project, too.
You know, we did a project early on at CSL.
And the subject matter expert was kind of AWOL, just wouldn't show up at meetings,
wouldn't respond to emails.
Just, you know, it was kind of one of those things like, it's above me or it's below me,
I guess it would be a better way to say it.
And, you know, we did the analytics and we showed it to the business and they're like,
these numbers cannot be right.
How does that be?
We did everything right.
So then we finally got a subject matter expert to come and look at and they said, oh, well,
you didn't realize that this is illegal in five states.
We're like, what?
We didn't know.
We're data scientists.
No, we took all 50 states.
They're like, no, you got to pull out California and, you know, these four other states because
you can't do that there.
Like, oh, okay.
Pulled those out.
Bang, spot on.
So then when I went to talk about project charters and things we're going to do, I said,
look, you know, if you guys aren't in the meetings with us, day in and day out and available
to talk to us about how this analysis is going to work, we're not doing it.
So it's a big deal.
You know, Justin and I were talking about it and I mentioned subject matter experts
and he's like, oh, yeah, you can't get those guys to come to meetings.
It's impossible.
I mean, I'm paraphrasing, right?
Yeah, so, you know, it has to be that way because, I mean, how many of you are data
scientists and experts in biomechanics?
How many?
I got no hands because those don't exist.
It's hard enough for us to find unicorn data scientists, let alone unicorn data scientists
who understand lean manufacturing.
I mean, it just doesn't happen.
So, you know, it's a team effort.
Data science is a team sport and the business has to understand that.
They have to be in the boat with you.
They have to be committed to the project.
They have to bring their expertise.
Now, if I was at a different conference, I would say this differently.
I would say that their subject matter expertise is as important as our data science expertise,
but we know that's all not true.
We're more important than they are, right?
We all know that.
Obviously, I just.
But, okay, I'm going to take another drink.
Got a couple of questions.
All right, come on up.
Let's take yours first, then we'll take yours.
There are many different type of analyst roles.
Data science, BI, analyst, data engineer.
We're thinking about hiring analytics engineers.
So, in the new Gen AI era, do you start seeing those roles start getting blurred and blurred
and start merging, or do you see those roles going to still stand alone in their own functions?
Great question.
How do I see Gen AI affecting different roles?
First and foremost, I'm a bit of a snob as far as analytics go.
I don't do business intelligence.
So, you know, when people come to me and say, hey, we're going to craft BI onto your AI group,
I'm like, no, they're not the same animal.
We're not doing that.
So, I generally don't have that problem.
As far as, you know, data engineers and data scientists and machine learning engineers,
we have in this field, we have too many titles.
So, in my opinion, what you have is you have people who do data work,
you know, data acquisition, data engineering, that kind of stuff.
And you have data scientists.
That's, in my world, that's the only two groups that exist.
So, the next question in your question was, does Gen AI change that?
And to some extent, it does, but not as much as you would think.
I've been buried, absolutely buried in Gen AI for the last four months.
Now, I know it started two years ago and all that kind of stuff,
and I wasn't paying attention.
But in my new role, I've just been absolutely, you know,
buried inside open AI and Google and Vertex and Microsoft and,
you know, trying to figure it all out.
And what I came to, I came to a conclusion about a month ago,
that the new generative AI skill sets and things that we have to do
are the same things that we did for NLP, for the most part.
So, that's a long way of saying, no, it doesn't really change.
Because we still have to do our information management work.
The one thing that we do have to learn as a group
is we have to learn the new architecture and the processes
and the information integration points that work for Gen AI.
It's not as easy as they would lead you to believe, you know.
We've built four different systems now, and each time we got better.
Because we're like, why doesn't it do this?
You know, the demo from Gen AI did it.
It's like, well, yeah, okay, well, yeah, they did this,
and we forgot about that, or we didn't know about it,
and we just learned about it.
So, what I've been telling, you know,
people at the Fortune 500 companies that I talk to,
you know, the whole Gen AI wave is important.
It is a big deal.
Large language models change almost everything.
Is it radically different than the things that we've done in the past,
especially with unstructured data?
No.
Once you learn the vernacular, once you learn the processing,
once you learn the architecture and how the queries flow,
and how query relevancy and accuracy is achieved,
you will be as good as anybody at it.
So, you don't have to worry about that stuff,
and you don't have to get a whole bunch of people
that tell you they're experts in these areas.
You just don't need them.
Speaking of roles within the industry,
what are your opinions and thoughts on progressing a person
within the field and how to have those conversations?
Yeah, I'm a big believer.
Everybody link with me on LinkedIn.
That's the first thing.
If you follow what I say on LinkedIn,
you will already know I'm a big believer in the next generation.
I work at the high school level.
I'm on the boards of the University of Michigan,
University of Texas, Oklahoma State,
Penn State, and my alma mater, Ferris State.
Go Bulldogs.
And I'm always trying to figure out
what we can teach the next generation,
how we can help them move up through the ranks.
And that goes back to the job descriptions
that I talked about early on,
is that I've written the job description
so it goes all the way from intern
to junior data scientist to data scientist
to senior data scientist to principal data scientist.
And I've argued effectively for compensation bans
that if you want to be a data scientist,
you never, and an individual contributor,
you never have to leave the data science profession.
You can make as much money as most managers
by staying as an individual contributor.
So the structuring of the jobs is one thing.
The compensation is another.
And then the mentoring is just as important.
You know, all the people in my team
who are seniors and principals
know that they have to mentor the interns and juniors
and just data scientists to help them get to the next level.
So we're all invested in each other's success.
So I think that's very important.
If data scientists don't have the data that they need,
how much responsibility should they take
for getting the data that they need?
You know, I always like to be a little bit controversial
and I will bite my tongue and say
what comes to my first mind,
or comes first to my mind is that,
you know, usually, let me, I'll make an assumption
that the data actually exists.
Yeah, we assume that it exists.
All right.
In a lot of cases, you know, subject matter,
the line of business and the subject matter experts
will come to us and say something along the lines of,
you know, hey, I'd like to have an analysis
of the mental state of every customer we have.
Do you have neural implants?
I don't know how to do that, you know?
I'm like, that doesn't exist.
And once you chip every person that we do business with,
then I'm all in.
But I think there's some ethical and privacy issues there.
If the data exists and it can be acquired ethically,
legally, and economically,
I think the data scientists should do everything
in their power to get it.
But if it doesn't exist,
or if it's prohibitively expensive,
it's their job to immediately raise the flag and say,
I have to have this.
But these are the concerns.
And then the business can decide
if they want to pay for it or not.
So that's the way I see it.
All right.
You spoke a little bit about how a team should be,
where a team should sit,
and how projects should be organized.
Do you have any recommendations or any words of caution
on internally how the team is organized
that could debilitate the team as an overall?
That's a great question.
I don't think teams fall apart because of structure.
I think teams fall apart because of divisiveness
and people who are jerks.
I think those are the biggest problems right there.
One of the things that I did in the last four teams that I had
was that I had weekly meetings with all my teams,
and everybody reports out what they're working on.
And it's high level.
We're not generally getting down
and doing code reviews and that kind of stuff.
I'm a lapse programmer.
So I'm not current.
But what all those teams also did
was they had another weekly meeting where I wasn't in it.
And what would happen a great deal of the time
is that I'm a passionate person.
I tend to have a bias towards urgency.
I want things to get done.
I want things to move fast.
And sometimes I make mistakes.
So what happened in those meetings that I'm not in
is all the team brought forth all the perceived errors I made.
And someone was nominated almost every week
to come and straighten out John.
And I was always open to it.
I always listened.
And if I could make a change that would help the team do better,
I did it.
So as a leader, you have to have openness and humility.
So I think communication is generally the balm
that makes these things better.
So I don't think it's a structure thing.
I think it's a communication issue and a respect issue.
Yeah, I was just wondering what your view
on applying more standard DevOps practices
like Agile to the data science team,
what your experience is there and what you think.
Okay.
So if you're an Agile adherent,
you're going to be offended by these next remarks.
Agile in data science does not work.
Don't do it.
Don't let people force it on you.
Just doesn't work.
There's too many vagaries and challenges
and iterations and failures in data science
for it to be an Agile situation.
And all the things that we've talked about data scientists,
I've had many data scientists that are on the spectrum.
You know, Agile standups, daily standups,
most data scientists will chafe at that.
They see that as babysitting.
There's a lot, you know, we had one group that came in
and said, well, we're going to run a project at Agile.
And I said, let's see how this goes.
It didn't go well.
So I don't think, thank you, sir.
I don't think that works.
And we've tried it and we've seen it fail.
So next one, we'll go to you next.
Hi, John.
Hi.
You referred to not hiring jerks
or not having someone that's jerking your team.
What's your practical advice
on basically identifying someone as jerk
during that limited time that you have interviews?
And I totally agree with you.
I don't want to have someone who's jerking my team.
Fire fast.
I'm serious.
You know, in the United States,
you know, almost everyone that's hired
has a three month probationary period
that you can fire them for any reason whatsoever.
That's the fact.
So, you know, you usually find out within three months.
As my wife says,
it's really hard to keep crazy in the bottle.
So it's going to come out, you know.
All right, who is next?
I nodded to someone.
I'm sorry.
Yeah, you right there.
I'm sorry.
Thank you.
Can you elaborate a little bit on the mentoring part?
How do you structure it
so that knowledge flows efficiently
from seniors to juniors?
Okay, mentoring.
Yes.
Generally, it comes in two or three forms.
One is the soft skills.
You know, communication,
understanding how to deal with people,
how to be sensitive to their situation,
empathetic to their state of mind.
That's one thing.
And we at data scientists are bad at that.
So, you know, we offend people all the time.
You know, you see junior data scientists
in meetings with senior vice presidents
just telling them straight up,
you're an idiot.
I'm like, whoa, maybe,
but now does not say that.
And then it comes in the hard skills,
you know, increasing difficulty of problems,
simple classification problems,
clustering problems,
and going more into forecasting
and things like that.
And then the other area that we work on
is being able to see out.
One of the things that we have done,
my teams have been able to do successfully
is that we have been able to build
analytical applications
that had multiple lives.
You know, at CSL, sorry, CSL again,
we built a forecasting application
that we've started out at the daily level
for all the 350 donation centers
for 12 months.
It worked pretty well.
Then we stretched it to 18 months,
then we stretched it to two years,
and then we stretched it down to the hour.
And the business was like,
we don't need forecast by the hour.
But what we did is we used that forecasting engine
to do labor scheduling
because we knew exactly
when the donors would be there.
And then we looked at all the employees
and all the cross-training
and we forecasted the optimal mix of labor
in every one of those 350 centers
by the hour of day.
And when we put it up
and we showed it to the senior management team,
they're like, we don't need to do this.
And then the last slide was $35 million.
And they said, what's that?
So that's the labor savings if we do this.
They're like, we're doing that tomorrow.
All right, there was a question.
I think we're actually out of time.
Oh, sorry, guys.
We're out of time.
First, you got to clear the room for the next session,
but thanks for a great Q&A.
Literally, connect with me on LinkedIn.
We'll continue the dialogue.

