0:03

What does it take to hire

0:05

at scale? When scale means

0:07

hundreds of thousands of roles

0:09

globally every year across fulfilment

0:11

centres, corporate offices, tech hubs,

0:13

headquarters and warehouse facilities? And

0:15

how do you do it

0:18

efficiently, effectively, fairly and powered

0:20

by data from start to

0:22

finish? I'm David Green and

0:24

today on the Digital H.R.

0:26

Leaders podcast. I'm joined by

0:28

Ashish Paruka. Director of Data

0:30

Science and Global Head of

0:32

Talent Acquisition Analytics at Amazon to

0:35

explore exactly that. With a career

0:37

steeped in analytics and a front

0:39

row seat to one of the

0:41

most complex and sophisticated hiring machines

0:43

in the world, Ashish brings a

0:45

unique perspective on how to make

0:47

people data work at both volume

0:49

and depth. In our conversation we

0:51

dive into how Amazon approaches the

0:53

twin challenges of high volume and

0:55

high stakes hiring. how analytics is

0:57

used to optimize cost, quality and

0:59

fairness, and what it really takes

1:01

to assess top talent without bias.

1:03

We also look beyond Amazon to

1:05

the broader people analytics profession.

1:07

What skills matter most? Where

1:09

do the biggest opportunities lie? And

1:11

how do organisations just getting

1:14

started avoid falling into the

1:16

trap of data theatre and

1:18

instead focus on driving real

1:20

impact? So if you're leading

1:22

talent strategy, building or scaling

1:24

your analytics capability, Or just

1:26

wondering how to do more

1:28

with less and do it

1:30

smarter, this episode is one

1:33

for you. So without further

1:35

ado, let's get the conversation

1:37

started. Ashish, thanks for joining

1:39

me today. To start the

1:42

conversation off, could you share

1:44

a little bit about your

1:47

career journey that has led

1:49

you to where you are

1:51

today and your role at

1:54

Amazon? Thank you David first of all for

1:56

having me on your podcast and all the great

1:58

work you do for the field. I'm a big

2:00

fan and really glad to be

2:02

on your show. So a little

2:05

bit about me. So when I

2:07

did electrical engineering and data science,

2:09

I did not anticipate being in

2:12

the people analytics space. But,

2:14

you know, early part of

2:16

my career was exploration of

2:18

different roles like data science,

2:21

marketing, product. And then it

2:23

was really some key mentoring

2:25

conversations that drove me. to where

2:27

I am. And I'll talk to

2:29

you about the one about people

2:31

analytics. So I had been working

2:33

in product for about 10 years,

2:35

really enjoying the talking to the

2:38

customers, understanding their needs,

2:40

bringing tangible things that help

2:42

people and make a positive

2:44

impact, and also bringing in

2:46

technology and analytics to do

2:48

the good work in that

2:50

space. And I had a few options

2:52

for different roles and I was having a

2:55

conversation with my mentor on like, which one

2:57

should I go with? And she said, you

2:59

know, all these roles are great, but she's

3:01

like, when I see sparkle in your

3:03

eyes is when you're talking about analytics

3:06

and when you're talking about people.

3:08

And by the way, there is a

3:10

role for that called people analytics. And

3:12

I was like, oh, wait, HR. And

3:14

you know, she caught me by surprise.

3:16

And you know, it's funny how sometimes

3:18

others know you better than you do.

3:20

And she said, you know, you're trying

3:22

to change your company's hiding strategies and

3:24

performance management and start even your day

3:27

job. And you bring so much energy

3:29

to those work streams that you should

3:31

really look into it. And you know,

3:33

I respect her a lot. So I

3:35

looked into it and more I learned,

3:37

more excited I got. Well, you've hit

3:39

a very, very prescient advice that you

3:41

got there. So you were clearly a

3:44

natural fit for people analytics.

3:46

So when you found out more

3:48

about people analytics, what really attracted

3:50

you to the field, and maybe

3:53

an add-on question there, is

3:55

what advantages do you

3:57

feel that you had from the working

3:59

outside HR, people and if you prefer

4:01

some of the skill sets that you could

4:03

bring into the field. No, that's a

4:06

great, great question. I think about this

4:08

as in three factors. One is purpose,

4:10

then the ability to make impact and

4:12

the scale of impact you can have,

4:14

and then do you have the skill

4:17

sets to actually make that impact, right?

4:19

So if you think about purpose, you

4:21

know, things like fairness, go deep with

4:23

people, right? And everybody I think

4:25

has a story of well, they

4:27

felt being wrong, then things were

4:29

not really and so on. And

4:31

I have that story too. So

4:33

when you're working in a space

4:35

that can help people get

4:37

opportunities that are based on

4:40

merit and things that more fair,

4:42

I think there's a sense of

4:44

purpose to it and sense a mission

4:46

to it that you may not find

4:49

in a lot of other spaces. And

4:51

for me, that is the

4:53

number one factor that got

4:56

me interested in people analytics.

4:58

the work that you're doing at Amazon.

5:00

Now obviously everyone listening to this

5:02

show will know that Amazon is

5:04

one of the largest organizations

5:06

in the world and I can

5:09

only imagine what the scope and

5:11

volume of hiring will be and

5:14

obviously in your role as global

5:16

head of talent acquisition analytics you're

5:18

looking all the data that supports

5:21

that process. How does a scale

5:23

and scope of talent acquisition at

5:25

Amazon differ from your previous

5:27

roles? We are hiring hundreds of

5:29

thousands of hourly employees, tens of

5:32

thousands of corporate employees each year,

5:34

right? And that might be the

5:36

size of entire companies in many

5:39

places. But I do think the

5:41

biggest difference is having skin in

5:43

the game. What I mean by

5:46

that is as my team is

5:48

building models and putting them into

5:50

production that drive marketing, that drive

5:53

the funnel that are hiring, so

5:55

for example in hourly hiring. If

5:57

our models are wrong or

5:59

they are down because of system

6:02

issue, that directly impacts if we

6:04

hire the exact number of people,

6:06

we need to hire at a

6:08

location, which impacts if packages get

6:10

delivered to people on time, right,

6:12

which impacts the bottom line of

6:14

Amazon. So that direct accountability

6:17

for something yet you do that

6:19

impacts the company's bottom line, it's

6:21

very different, right? In many other

6:23

places, you might have people are

6:26

not interested in doing reporting or

6:28

making interesting insights, then the operations

6:30

team have to take those insights

6:33

and make it a reality. Here

6:35

teams are structured for single thread

6:37

ownership and what you build you

6:39

put into production and it affects

6:41

things in market, right? Even on

6:43

the corporate side, you know, we

6:45

ran six million assessments last

6:47

year, right? And this system has

6:49

to be on globally 24-7 and

6:52

if it's down for an hour,

6:54

it can affect. tens of thousands,

6:56

lots of different businesses like prime,

6:58

AWS, you know, different devices

7:00

and you might lose out on

7:03

a really important candidate who could

7:05

build the next big thing.

7:07

I think that accountability is

7:10

a big difference, but with that

7:12

accountability also comes lots of

7:14

benefits that make it a

7:16

pleasure to work here, right?

7:18

With scale, you can actually

7:21

do advance and then it

7:23

takes. and science and show

7:25

what matters versus what doesn't

7:27

matter. What are the myths

7:29

versus what are the realities with

7:31

the underlying tech that

7:33

AWS offers us. You can have sensors

7:36

everywhere and make sure no

7:38

data is lost in the

7:40

ether, right? It's you have

7:42

reliable, always available data, which

7:44

is massive when if you're

7:46

trying to do science and

7:49

and also the company culture

7:51

being that of data back decision making

7:53

is really critical. That was one of

7:55

the main questions I had when I

7:57

was switching from the business side over

7:59

to. people and under text is like,

8:01

okay, even if you come up with

8:03

great insights, our people, our executives,

8:05

our companies really gonna make a change,

8:08

but are they gonna say like, yeah,

8:10

in theory, data back decision making, all

8:12

of us should do it, but in

8:14

my unique situation, it doesn't apply,

8:17

right? My judgment is, you know,

8:19

more important and there is tremendous

8:21

room for judgment always with

8:23

people, right? But there are things where

8:25

that numbers can get you more right

8:27

and wrong and we should leverage. How

8:30

does work really get done

8:32

in your organisation? Worklittix helps

8:35

companies measure collaboration, productivity and

8:37

AI adoption, using real data,

8:40

not guesswork. With insights into

8:42

meetings, tool usage and work

8:45

patterns, Worklittix helps enterprises optimize

8:47

team performance to get more

8:50

done faster. They're currently offering

8:52

podcast listeners a complementary AI

8:54

adoption assessment. Gain insight into

8:57

your organization's AI usage and...

8:59

and unlock its

9:02

full potential.

9:04

Limited spots

9:07

are available.

9:10

Learn more

9:12

at worklitics.co.org/AI.

9:15

That's W-O-R-K-L-Y-T-I-C-S-I-C-S-C-O-4-S-A-I.

9:28

What are some of the

9:31

innovations or tools that you've

9:33

implemented to to optimize Amazon's

9:35

high volume hiring processes?

9:38

I'm happy to talk about that, you

9:40

know, you know, things like AWS Connect,

9:42

help us build our own ATS system

9:45

that can scale and have

9:47

the fungibility and quick, quick

9:49

iteration that we need, as

9:51

well as, you know, things like Sage

9:53

Maker. help us build models and

9:55

put them in production. But let

9:57

me digress a little bit because

10:00

you know, it's it's most important

10:02

to solve the right problem,

10:04

right? And then the technology

10:06

should be at the end of the

10:08

end of the journey, right? So let

10:10

me talk about it. Okay, what is

10:12

the problem? Right. So we

10:14

have 3,000 plus sites around the

10:17

world, right? And every week, a site

10:19

may need a site might tell me,

10:21

okay, I need five hires. The week

10:23

after that, I might need two 33

10:25

hires. And the week after that.

10:27

that site made me 37 hours.

10:29

And the site next to it

10:31

might say 150 hires, no, 45

10:34

hires and 55 hires, right?

10:36

So you have 3,000 distinct

10:38

strings of demand coming through

10:41

and then we have three weeks

10:43

of heads up to know like

10:45

how much demand will be.

10:47

So you have short heads of

10:49

time, we have unique demand,

10:51

and then you have the scale

10:54

of Amazon, right? So that's

10:56

that's the. problem on one

10:59

side. On the other side, I

11:01

have 16 different dials or levers

11:03

I can pull to get exactly

11:05

the number of hires and I

11:08

need to get 95% accuracy

11:10

in this operation. So now, if

11:12

all these moving pieces and your

11:14

head can start spinning. So

11:17

that's the problem. Now let's

11:19

think about the context, right?

11:21

So six years ago, Amazon

11:24

used to outsource this hiring.

11:26

And we didn't have an internal

11:29

team hiring in the hourly space.

11:31

Six years ago, we decided, like,

11:33

okay, let's bring it in-house. And

11:35

in the first year, we needed to

11:37

double the hiring, right? And imagine doing

11:39

it for the first time and you

11:42

have to double. And then in the

11:44

next two years, we had to go 6X,

11:46

right? And a lot of that had to

11:48

be built based on judgment and

11:50

subject matter experts. So, you know,

11:52

about three years ago when I

11:54

joined, we had this massive operation.

11:56

For example, in marketing we had

11:59

300 export. sitting in different media

12:01

markets, and telling us, like, hey,

12:03

when to start, when to stop

12:05

marketing, what channels, how much to

12:08

spend. And we don't have all

12:10

this data captured. So we know

12:12

we are getting the highest we

12:14

need, but we don't know the

12:16

intelligence behind, like, what is the

12:19

team? Like, how are we really

12:21

making decisions? Where are we making

12:23

good decisions versus not? So the

12:25

first step was actually capturing all

12:27

the data, right, like a lot

12:29

of the times people. really underestimate

12:32

investment in data and that get

12:34

distracted by shiny AI topics. If

12:36

you don't invest in data, you're

12:38

really capping the upside of what

12:40

you can achieve. So the first

12:43

year was really about capturing data

12:45

on what are the inputs people

12:47

are using to make decisions, how

12:49

are they making decisions, when do

12:51

the outcomes actually pan out versus

12:54

not. And the next step was

12:56

just to put some simple rules

12:58

around it. Like, hey, what is

13:00

the common theme across our whole

13:02

network when certain decisions work versus

13:04

others don't? So some of the

13:07

decisions started getting automated through rules,

13:09

there was still a lot of

13:11

judgment involved, right? And that alone

13:13

saved or reduced our cost by

13:15

50%, right? And then the year

13:18

later, we were able to reduce

13:20

the cost by 90% by taking

13:22

the rules and replacing them with

13:24

machinery. Right. So there is this

13:26

journey, right, of capture the data,

13:29

simple things first, put it into

13:31

market, even if it's not perfect,

13:33

learn from it, and then go

13:35

to more advanced machine learning models.

13:37

And even with machine learning models,

13:39

it's not fully automated. We still

13:42

have judgment in the right spots.

13:44

Right. So we've, we've done that

13:46

transition over time. And when you

13:48

come to that final stage, and

13:50

that's where I can talk about

13:53

our. and sage maker and how

13:55

it can make help us you

13:57

know run these models at globally

13:59

for these 3,000 slides. and the

14:01

thousands of hires that we are

14:04

making at that scale and be

14:06

accurate, right? But it comes at

14:08

the end of the journey, right,

14:10

of prototype and iterate over time.

14:12

Let's switch a little bit to

14:14

corporate hiring. Now you'll obviously do

14:17

a lot on the corporate hiring

14:19

side at Amazon, but obviously not

14:21

anywhere near the volume of that

14:23

you're doing on the just on

14:25

the hourly paid workers. What are

14:28

you currently doing to ensure that

14:30

you're acquiring? the top talent in

14:32

the space, but also assessing them

14:34

in a fair and unbiased manner.

14:36

Let me start personally by saying,

14:39

like my mission is to minimize

14:41

guesswork and bias when it comes

14:43

to people's decisions. Like that's what

14:45

drives me, right? And corporate hiding

14:47

space is where that is probably

14:49

most applicable. So if you think

14:52

about our hiring experience today for

14:54

most companies, from candidate perspective, They

14:56

feel that they're tossing their resume

14:58

on a pile of thousands of

15:00

resumes They don't know if anybody's

15:03

even gonna get take a look

15:05

at it and They are probably

15:07

their baseline expectation is not to

15:09

hear back from anyone when you

15:11

apply on that right and they're

15:13

told okay, you need to network

15:16

with the hiring manager or the

15:18

recruiter da da da da da

15:20

and then there's a lot of

15:22

luck involved and if you even

15:24

get a fair shot at evaluating

15:27

what you can do and how

15:29

does that relate to a job

15:31

Right. And then on the flip

15:33

side, if you're thinking about the

15:35

recruiter, you've got stacks of thousands

15:38

of resumes, right? It's not practical

15:40

for you to really evaluate all

15:42

thousand resumes, right? So you put

15:44

some shelters and narrow things down

15:46

and you're able to maybe look

15:48

at a few hundred and, you

15:51

know, 15 to 30 seconds at

15:53

a top, right? And in that

15:55

time, it's a really tough job

15:57

to pick out of that pile,

15:59

really the top. candidates, right? And

16:02

that's where, again, you know, human

16:04

judgment and machines together. can do

16:06

a better job. So we have

16:08

automated resume reviews, we have online

16:10

assessments that we have scaled for

16:13

like hundreds of job families. So

16:15

we have 455 job families for

16:17

which we have online assessments. We

16:19

ran 6 million plus assessments last

16:21

year. And the thing I love

16:23

about assessments is it's an opportunity

16:26

for candidate to show what they

16:28

can do. Right. Here the skills

16:30

that are important for the job.

16:32

Show me what you can do.

16:34

Right. And so we had, we

16:37

gave people six million plus metered

16:39

base shots to get a job

16:41

at Amazon, right? And for some

16:43

candidates that can truly be life

16:45

changing and it could be life

16:48

changing, you know, for the company

16:50

and our customers because, you know,

16:52

if you hire the top talent

16:54

who was best suited for that

16:56

job, then we can make a

16:58

bigger difference for the customers. And

17:01

we also have machines. helping candidates

17:03

navigate, right? So each company has

17:05

different titles and different ways of

17:07

organized function. And so when candidates

17:09

come in, they can upload a

17:12

resume and we can help them

17:14

direct like, hey, based on your

17:16

skills, here are actually the jobs

17:18

you might be best suited for

17:20

Hilton levels in families. So there's

17:23

a lot of technology that can

17:25

help improve both the candidate experience

17:27

as well as a reporter experience

17:29

to create the best match in

17:31

the end. between the job and

17:33

the person. And then that's really

17:36

what we are going for in

17:38

the corporate space. Very good. Very

17:40

good. That's 6 million assessments. Wow.

17:42

Yeah. Yeah. That's a lot of

17:44

data as well, obviously. Yes, that's

17:47

the time of data, but it

17:49

also requires some incredible subject matter

17:51

expertise and things like AIO psychology,

17:53

as well as engineering to make

17:55

sure all you can deploy these

17:58

assessments and keep it up. 24-7

18:00

all through the year. So a

18:02

lot of amazing work that the

18:04

team is doing. Yeah, that really

18:06

good. So on the topic of

18:08

corporate hiring, let's turn the lens

18:11

on people analytics itself, which in

18:13

some respects is still an emerging

18:15

field, still growing. What would you

18:17

say to someone considering a career

18:19

in this area, and that by

18:22

extension, what would you say are

18:24

the top skills that you look

18:26

for when hiring people and the

18:28

six professionals? Yeah. So my pitch

18:30

for anybody who's, you know, maybe

18:32

on the tech side, business side,

18:35

to come over to people and

18:37

analytics is the same as we

18:39

talked up to. So, hey, this

18:41

is a really mission driven field.

18:43

You can make 10x, 100x difference

18:46

in this field. And a lot

18:48

of the skills that you learn

18:50

on the business side and the

18:52

tech side can translate very nicely

18:54

in this view, right? So it

18:57

can be a really fulfilling endeavor

18:59

and you can be in the

19:01

frontier of lots of problems that

19:03

haven't been solved yet. So that

19:05

would be my pitch for somebody

19:07

considering. And especially at Amazon, you

19:10

have the scale of data to

19:12

actually do the similar types of

19:14

analytics and science that you could

19:16

do on the business side. In

19:18

terms of what I would advise

19:21

people in general, if they're thinking

19:23

about career in this space, it

19:25

may not be too different from

19:27

career in other spaces, is I

19:29

feel going forward big wins are

19:32

going to be at intersection of

19:34

functions. So let's take data engineering,

19:36

right? I feel especially not just

19:38

in people on analytics, but with

19:40

the advances in AI, a key

19:42

skill that everybody's gonna need is

19:45

data engineering, right? And it's quite

19:47

scarce, actually. We find that finding

19:49

great data engineers is harder than

19:51

finding great scientists, just based on

19:53

a number of professionals in the

19:56

space. But if you are a

19:58

greater data engineer with advances in

20:00

tooling from like ABUUS and what

20:02

AI can do these days, going

20:04

deep in data engineering alone may

20:07

not be enough. And we are

20:09

gonna need some people who are

20:11

like incredibly deep in data engineering,

20:13

but for most people, if you

20:15

can marry your skills of data

20:17

engineering by also understanding the process,

20:20

understanding the business. then you can

20:22

make the right decisions on where

20:24

to put the sensors, what is

20:26

it that you need to capture,

20:28

how do you prioritize work, right?

20:31

And that's going to be most

20:33

valuable to your end customer. That's

20:35

what's going to be helping you

20:37

make the most impact, right? So

20:39

intersection of those two things. Let's

20:42

take about, talk about science, right?

20:44

I talked about like in selection

20:46

space, you need deep expertise in

20:48

iO psychology. But if you marry

20:50

that with ML. you can scale

20:52

so much faster, right? Like one

20:55

of the biggest challenges I've heard

20:57

in the time I've been in

20:59

the people analytics space is like,

21:01

it takes too long for us

21:03

to build selection mechanisms, right? Like,

21:06

it takes 18 months, right? Oh

21:08

my God, like world changes in

21:10

18 months, right? But how can

21:12

we marry multiple? So like, you

21:14

know, we had automated resume review

21:17

for 20 job families. last year

21:19

and now we have it for

21:21

220, right? Similar scale upgrowth in

21:23

other areas like online assessments. So

21:25

how can we marry different sciences

21:27

together to scale? In product, I

21:30

think that's where probably three intersections

21:32

come in. I think product as

21:34

a discipline is really hard to

21:36

find great product managers, but if

21:38

you can do product and software

21:41

and science, That's a killer combination,

21:43

right? Is it really, and if

21:45

you think about where the world

21:47

is going, right? You are going

21:49

to need all these, these skills

21:51

to really leverage harness the goodness

21:54

that AI can bring you. It's

21:56

going to be sciences. going to

21:58

be software and it's going to

22:00

be product. So in my, in

22:02

general, I would say intersection of

22:05

things is where big wins are.

22:07

So if you are only in

22:09

your career, explore versus maximize, right?

22:11

If you just stay in one

22:13

function and maximize, you may be

22:16

limiting the upside long term and

22:18

might be better off the new

22:20

career going broader and collecting a

22:22

broader set of skills so that

22:24

you are building a good foundation

22:26

for a. a greater future in

22:29

the long term. I want to

22:31

take a short break from this

22:33

episode to introduce the Insight 222

22:35

People Analytics Program, designed for senior

22:37

leaders to connect, grow and lead

22:40

in the evolving world of people

22:42

analytics. The program brings together top

22:44

HR professionals with extensive experience from

22:46

global companies, offering a unique platform

22:48

to expand your influence, gaining your

22:51

influence, tackle real-world business challenges. As

22:53

a member you'll gain access to

22:55

over 40 in-person and virtual events

22:57

a year, advisory sessions with season

22:59

practitioners, as well as insights, ideas

23:01

and learning to stay up to

23:04

date with best practices and new

23:06

thinking. Every connection made brings new

23:08

possibilities to elevate your impact and

23:10

drive meaningful change. To learn more

23:12

head over to Insight22. board slash

23:15

program and join our group of

23:17

global leaders. Interestingly, maybe it's just

23:19

two parts of this question as

23:21

well, you know, what roles are

23:23

you currently hiring for on your

23:26

team at Amazon? And then the

23:28

second part of that, you know,

23:30

how the skills and capabilities evolving

23:32

over time, but the second part

23:34

is also, you know, how's... Where

23:36

do you envisage maybe in the

23:39

next 12 months, 18 months that

23:41

you might be hiring additional skills,

23:43

maybe new skills into into into

23:45

your too? Yeah, so we have

23:47

25 rolls open right now. Given

23:50

the size of the team, we

23:52

typically hire about 50 people each

23:54

year. We've got rows in data

23:56

engineering, as I said, like, you

23:58

know. We have a large presence

24:01

in data engineering globally. We've got

24:03

software engineers for deploying things and

24:05

automated resume reviews. We also have

24:07

IOCology roles as well as data

24:09

science and machine learning roles. And

24:11

as always, I always would look

24:14

for product people who can do

24:16

all three, like product science and

24:18

software that is always a need.

24:20

for that. So you would have

24:22

a range of roles available and

24:25

I would invite both people who

24:27

are in people under takes for

24:29

a long time as well as

24:31

who are on the business side

24:33

to come take a look because

24:35

you know one of the hesitations

24:38

people may have from moving from

24:40

the business side is the scale

24:42

of data and access to technology

24:44

that you may have on the

24:46

business side. Do you have that

24:49

on HR side as well and

24:51

like at a company like Amazon

24:53

you do. So it might be,

24:55

if you're considering a change, this

24:57

might be a good place to

25:00

start, right? And then, then you

25:02

can go from there. And if

25:04

we can talk a little bit

25:06

about the product management piece, because

25:08

again, it's an area that we've

25:10

noticed growing in many people analytics

25:13

functions around the world, you know,

25:15

and we've identified a gap in

25:17

the research that we've done around

25:19

democratizing data and the adoption of

25:21

those products as well. and product

25:24

management plays a really important role

25:26

in that, doesn't it? Not just

25:28

of our user experience, but it's

25:30

around making the products easier. Yeah,

25:32

products. I love to talk a

25:35

little bit to the product management

25:37

piece. That'd be great. So on

25:39

one side, if you think about

25:41

the users, they are on this

25:43

journey about data literacy and then

25:45

come science literacy. So what they

25:48

deeply know is the problems they

25:50

haven't had today, right? But if

25:52

you if you ask them for

25:54

solutions, they might just tell you,

25:56

hey. help me do what I

25:59

do faster or easier, right? But

26:01

you may not get the transformative

26:03

idea from your users, right? So

26:05

as a product professional, you have

26:07

to understand the need, but then

26:10

also understand the technology to see

26:12

what is possible, right, to come

26:14

up with a function, come up

26:16

with a solution, and then you

26:18

have to sell that solution to

26:20

the customer, right? That's also really

26:23

challenging because they might not believe

26:25

you. Right. And then in the

26:27

end, as I said, they are

26:29

on the hook to deliver, right?

26:31

So you paint a picture which

26:34

may look like a crystal ball

26:36

and they might not sign up

26:38

for that solution. So like how

26:40

do you bridge? You deeply need

26:42

to understand the technology so that

26:45

you're not promising a crystal ball.

26:47

You're promising something real. And at

26:49

the same time, you have to

26:51

sell it to the customer so

26:53

that they believe it. And then

26:55

once you have that alignment, you

26:58

have to go build it, right?

27:00

And in building it, there is

27:02

such a dramatic curve of improvement

27:04

in technology that is so, you

27:06

know, what used to be technical

27:09

debt over years, you can accumulate

27:11

technical debt within months now, right?

27:13

The things you're building your infrastructure

27:15

on can be obsolete within months.

27:17

So how do you also stay?

27:20

on the frontier of technology where

27:22

things are going to make the

27:24

right investment decision and build things

27:26

right, give your... functionality to be

27:28

on the most modern technology. That's

27:30

the other side of the story.

27:33

So really incredible opportunity for people

27:35

who can do that. They can

27:37

really unlock the potential of all

27:39

the individual functional experts we have,

27:41

if you have great productivity. So

27:44

as someone who's been deeply involved

27:46

and embedded in talent acquisition analytics,

27:48

and I know you were doing

27:50

similar before you came to Amazon

27:52

as well, what do you see

27:54

as the biggest opportunities for people

27:57

analytics to drive even further impact

27:59

with regards to tenant acquisition? Yeah,

28:01

I think in talent acquisition space,

28:03

the biggest problem is quality of

28:05

art, right? If you can nail

28:08

that, I think businesses would be

28:10

willing to invest a lot more

28:12

in this space, right? And it's

28:14

about measuring the quality of hire

28:16

and then showing you can actually

28:19

move the needle, right? In general,

28:21

even if you are just doing,

28:23

let's say, skill-based hiring and using

28:25

best practices, you're probably in the

28:27

top half of the companies anyway,

28:29

right, that a lot of companies

28:32

who are still not doing. skill-based

28:34

hiring. They're not doing structured interviews,

28:36

right? These are like proven practices

28:38

for, you know, decades now. So

28:40

if you're not doing that, do

28:43

that, right? I think that would

28:45

be the first step. But even

28:47

if it's skill-based hiring, and you

28:49

know, given the limited time we

28:51

have to evaluate talent, you know,

28:54

a few hours, yeah, getting skill-based

28:56

signals is probably a good, good

28:58

way right now. But that still

29:00

leaves the unknown of can this

29:02

candidate put these, put these skills

29:04

together to deliver the tasks that

29:07

they need to complete in. environment

29:09

of this company in the culture

29:11

of this company, right? So that

29:13

remains unknown. A lot of the

29:15

misses we end up having are

29:18

because people may have the underlying

29:20

skills but cannot put it together

29:22

or maybe cannot put it together

29:24

in this environment. It's a great

29:26

experience for the candidate to see

29:29

how it's really like. to work

29:31

in a particular role in a

29:33

particular company, what they would be

29:35

asked to do day to day,

29:37

and can they pull all the

29:39

skills they have together, both the

29:42

soft skills and the functional skills,

29:44

to deliver the solutions that the

29:46

company needs, right? And that's where

29:48

I believe the world is going.

29:50

And if you are able to

29:53

do that, then you will be

29:55

able to show measure more accurately

29:57

a person's ability to do high

29:59

quality work. in that particular job

30:01

and it would be better for

30:04

the candidates as well because it's

30:06

a life-changing decision to you know

30:08

change a job move potentially to

30:10

another place and then you have

30:12

all these unknowns that you don't

30:14

know how we're going to cut

30:17

out right so more closer we

30:19

can get to real-life job preview

30:21

and I believe with AI we

30:23

can I think that would be

30:25

the next frontier. If we look

30:28

more generally at the field of

30:30

people analytics, you know, and this

30:32

leads quite nicely at you, one

30:34

of the key topics for everyone,

30:36

business leaders talk about its productivity,

30:39

we've discussed it in the past.

30:41

You posted some excellent articles on

30:43

it a few years ago as

30:45

well. But for the benefit of

30:47

listeners, what role can people analytics

30:49

play in your opinion in terms

30:52

of unlocking productivity at the individual,

30:54

the team, and maybe the organizational

30:56

level as well? Great question. And

30:58

you know these days it's pretty

31:00

easy to jump to AI as

31:03

the answer for productivity. and you

31:05

know I always say it's most

31:07

important for us to solve the

31:09

right problem first right and get

31:11

the structure right and the technology

31:13

comes at the end like it

31:16

don't start with technology right and

31:18

speaking more broadly of people analytics

31:20

a big part of people analytics

31:22

is consulting with the business right

31:24

so if you're thinking about organizational

31:27

level productivity I still think it

31:29

is more about How do we

31:31

make sure we're solving the right

31:33

problems? How do we make sure

31:35

we have the right people to

31:38

solve those problems? How do we

31:40

make sure those people have the

31:42

resources to solve the problem and

31:44

so on so forth, right? So

31:46

making sure we have that structure,

31:48

right? And a lot of those

31:51

things might be more about having

31:53

clear decision-making mechanisms, right? Having a

31:55

great way to do workforce planning

31:57

to make sure your top talent.

31:59

with the right skills and are

32:02

in the right roles. And there

32:04

are areas where AI plays a

32:06

role in that, but it's not

32:08

just about AI, right? Like, you

32:10

know, when I talked about, you

32:13

know, we cut costs and now

32:15

we're adding by 50% and, you

32:17

know, improved accuracy in the high

32:19

90s, that it had nothing to

32:21

do with AI, right. So it's

32:23

really important not to be jumping

32:26

to that. and state do the

32:28

right problem solving techniques that break

32:30

down the big complex productivity problem

32:32

into components. Man it, make sure

32:34

that you can solve each of

32:37

the components, make sure you understand

32:39

the ecosystem and you can do

32:41

system thinking to figure out what

32:43

you should solve first before we

32:45

solve next and how, how, what

32:48

is a knock-on effect of solving

32:50

one thing on the another. And

32:52

then be aware of what AI

32:54

can do. in those steps. So

32:56

be aware of AI and apply.

32:58

that too, when is the right

33:01

tool for the job, right, versus

33:03

using AI, you know, looking, hammer

33:05

looking for an ale, that's probably

33:07

not the right way to apply

33:09

approach productivity. That's really, it's really

33:12

interesting, Chief, because I mean, when

33:14

we were But Jonathan and I

33:16

were writing Excellence in People Analytics

33:18

and prior to that and we

33:20

weren't the only people saying it

33:23

as well, not going to claim

33:25

ownership on this. You know, when

33:27

People Analytics was maybe in the

33:29

mid-2010, you know, lots of people

33:31

wanting to jump to the fanciest

33:33

type of analytics. I want to

33:36

do organizational network analytics and, you

33:38

know, we'd always say, well, what's

33:40

the problem you're trying to solve?

33:42

You said, but we'll properly define

33:44

what the problem is. Maybe have

33:47

some... questions that you want to

33:49

answer, some hypotheses that you want

33:51

to test. And only then, once

33:53

you've gathered the data together, do

33:55

you start thinking about, oh, what's

33:58

the solution for this, what, you

34:00

know, and that's when AI comes

34:02

in there, and not, as you

34:04

said, not, not a hammer looking

34:06

for a nail, I like that,

34:08

and an analogy. I'll give you

34:11

an analytics for our software organization,

34:13

right. And the team was quite

34:15

burnt out. and the customers were

34:17

not very happy about the outcomes.

34:19

And the team had received 400

34:22

different requests for projects through the

34:24

year, the previous year before I

34:26

joined. And they had all great

34:28

science and they had applied the

34:30

best available tools at their disposal,

34:32

but nobody was happy, right? So

34:35

I was like, okay, in that

34:37

situation, try to go as high

34:39

up as possible in terms of

34:41

customer. So run to the CIO

34:43

and ask him, okay, you're funding

34:46

this team, right? Tell me, because

34:48

of the work we have done,

34:50

is there something you are doing

34:52

differently this year compared to Vastir,

34:54

right? And he could. Okay, nothing.

34:57

There was nothing that the tech

34:59

organization was doing differently because of

35:01

the 400 things we did last

35:03

year and the team is burnt

35:05

out, customers, nobody's happy. Right. Now,

35:07

as I look, all right, so

35:10

what if we did two things,

35:12

right? And really made a difference

35:14

in two things and what would

35:16

be your two things? At that

35:18

time, it was about hiding and

35:21

repension. All right, so we're gonna

35:23

staff 60% of the team on

35:25

these two problems, right? And then,

35:27

you know, you know, there's always

35:29

this executive concer executive concierge executive

35:32

concierge. you know, questions from up

35:34

top that come and you want

35:36

to have like 10% for that

35:38

and then rest is Katie at

35:40

all right. So if that is

35:42

what we did, we might get

35:45

two, that's better than zero. Now,

35:47

this is a risky endeavor, right?

35:49

You might have this conversation and

35:51

you might be like, okay, I

35:53

actually don't need this team because

35:56

I haven't done anything different. But

35:58

so it's a delicate conversation. But

36:00

yeah, at the end of it,

36:02

when we actually focused on those

36:04

two problems and now we were

36:07

able to. increase the throughput of

36:09

software engineering by 50% but just

36:11

focusing on that piece, solving that

36:13

problem deeply. Then there were conversations

36:15

about, okay, how can we fund

36:17

this team for a third priority?

36:20

Right. So really like if we

36:22

had stayed in this like, you

36:24

know, solve 400 problems with top

36:26

technology, we may not have actually

36:28

made any impact. So that focus,

36:31

solving the problem, the right problem,

36:33

the right way. And maybe AI

36:35

is the answer to some of

36:37

those problems, maybe it's not, maybe

36:39

sometimes four-week average can solve your

36:42

problem. And you don't need AI.

36:44

It's really important, I mean, so

36:46

many times we hear about analytics

36:48

that are burning out because they're

36:50

just doing too much, but they're

36:52

not having much impact. And it's

36:55

because they're doing too much and

36:57

not prioritizing the right things. And

36:59

that's such a great example. of

37:01

doing that, you know, and that's

37:03

again how you build trust with

37:06

your internal customers, isn't it? What

37:08

are their key challenges that... trying

37:10

to solve for, help them solve

37:12

for them, they'll work with you

37:14

more and then, and actually it's

37:17

the business that will help you

37:19

get more investment in the team,

37:21

whether it's people, technology, other resources.

37:23

So yeah, really good example, yeah,

37:25

there as well. So I'm going

37:27

to go to the question of

37:30

the series as Shish now and

37:32

then I'm going to come back

37:34

and just ask for you maybe

37:36

to give some sort of key

37:38

things for people to take away

37:41

for them. So the question of

37:43

the series for those first time

37:45

listeners, this is something we ask

37:47

everyone in a series of the

37:49

Digital H.R. leaders podcast. And it

37:51

really is. We're going to talk

37:54

about AI a little bit here.

37:56

How can H.R. use AI to

37:58

improve employee experience and well-being? And

38:00

if you want to extend that

38:02

into candidate experience, please feel free

38:05

to do so. Yeah. You know,

38:07

the one way we are thinking

38:09

about this. is where is human

38:11

touch truly needed and truly beneficial,

38:13

right? And what is taking away

38:16

from humans being able to spend

38:18

that time in that human touch?

38:20

So think about recruiting, right? On

38:22

one side, there's a candidate who

38:24

could be making a life-changing decision,

38:26

right? Could be uprooting their family

38:29

and moving somewhere or being at

38:31

a job, they're doing really well,

38:33

but they're interested in something different.

38:35

These are big decisions. And it's

38:37

really important to have somebody you

38:40

can trust on the other side,

38:42

right? And then that is really

38:44

where humans can differentiate, which is,

38:46

you know, at least personally for

38:48

me talking to a machine may

38:51

not feel the same way. But

38:53

recruiters today have do so much

38:55

admin work, right? There are like

38:57

30, 40% of that late can

38:59

be admin work and that takes

39:01

them away from being connected to

39:04

the candidate, right? And how can

39:06

AI take off that stuff from

39:08

that plate to make sure, not

39:10

just in, let's say, operations roles,

39:12

even in data engineering? Right. I

39:15

remember back in the day, working

39:17

on name frame and you know,

39:19

other technologies, it was a bear

39:21

to understand where your data is,

39:23

where it is coming from. Today,

39:26

you're the data of US. It's

39:28

like having a conversation with your

39:30

data, right? Understand the metadata behind

39:32

like where does this field? What

39:34

does it mean? What is the

39:36

range? What is the source? What

39:39

is the latency? I can have

39:41

a conversation with my data. That

39:43

is just phenomenal experience for a

39:45

data engineer. And now they are

39:47

freed up to do the engineering

39:50

work they need to do versus

39:52

the admin work of just understanding

39:54

the basics, right? We've got some

39:56

last question, Shish, really enjoyed this

39:58

conversation and, you know, and I

40:01

think it hopefully... listeners will really

40:03

be able to take a lot

40:05

from it. But for those people

40:07

listening that are working in organizations

40:09

that are maybe just beginning their

40:11

journey to embed analytics into recruitment

40:14

or not an acquisition or maybe

40:16

want to take it to the

40:18

next level, what advice would you

40:20

give? Where should they start? And

40:22

what are the key success factors?

40:25

And I appreciate that not the

40:27

same success factors don't necessary apply

40:29

to every organization. So I think

40:31

there are a few different patterns,

40:33

right? So one pretty well-known. pattern

40:35

is types of analytics. So you

40:38

could have descriptive, like what is

40:40

happening, right, diagnostic, like why is

40:42

it happening, predictive, like what will

40:44

happen, and then prescriptive, what should

40:46

I do about it, right? So

40:49

people analytics teams evolve in that

40:51

direction as they have, you know,

40:53

more time under their belt, more

40:55

data, more size of the team,

40:57

they can move. Usually everybody starts

41:00

with this script, right. Now, that

41:02

is one model. Now, the thing

41:04

I would add to that is

41:06

rather than going descriptive on everything,

41:08

a lot of people on analytics,

41:10

I see, do tons of reporting

41:13

for everything under the sun. So

41:15

they are taking the descriptive piece

41:17

and going really broad. I would

41:19

actually highly recommend people to go

41:21

deep in few problems. So don't

41:24

do reporting for everything. Pick two

41:26

problems or one problem that business

41:28

really cares about solving this year

41:30

and go deeper in that problem

41:32

and make a difference in that

41:35

problem. Because let's say, you know,

41:37

this year, it might be about

41:39

workforce planning. If you can really

41:41

make a difference there, then the

41:43

team will, your team will get

41:45

funded to solve another problem and

41:48

another problem. But it's from the

41:50

business perspective, it's really about, are

41:52

you making an impact at the

41:54

bottom line? Before we wrap up,

41:56

where can listeners find out more

41:59

about you and learn about the

42:01

work you're doing and you and

42:03

your team are doing at Amazon?

42:05

And maybe, is there some way

42:07

they can go? If someone's thinking,

42:10

okay, I'd like to work with

42:12

Ashish and his team as well.

42:14

Yeah, so best way to find

42:16

me is on LinkedIn. And best

42:18

way to work with us is

42:20

go to the Amazon website and

42:23

search ITA, intelligent talent acquisition. So

42:25

if you search ITA in quotes,

42:27

you'll get the direct match for

42:29

25 roles that we are having

42:31

for today. I'm sure there will

42:34

be more coming up very soon.

42:36

And we'll put that in the

42:38

show now, but basically go to

42:40

the Amazon site, search ITA, intelligent.

42:42

talent acquisition and you'll see the

42:45

roles that Ashish was talking about

42:47

earlier. Asish, thank you very much.

42:49

I look forward to bumping into

42:51

you hopefully at a conference in

42:53

the next few months as well.

42:55

And thank you so much for

42:58

having me. Conversations like this remind

43:00

us that scale isn't just a

43:02

numbers game, it's a strategy game.

43:04

And when done right, people analytics

43:06

doesn't just support hiring, it transforms

43:09

it. Thank you to Ashish for

43:11

joining me and offering such a

43:13

clear-eyed view on how Amazon is

43:15

navigating hiring complexity with clarity and

43:17

using data to drive real im-

43:20

And thank you to you

43:22

all of you,

43:24

as always, for

43:26

the sling and

43:28

tuning week. Here at

43:30

week. 222, we are on a

43:33

mission to we are

43:35

on a mission

43:37

to help as

43:39

many HR and

43:41

people analytics professionals

43:44

and leaders skills,

43:46

the skills, strategies

43:48

and confidence needed

43:50

to drive real

43:52

business value. if you enjoy

43:54

if you enjoyed

43:57

today's episode, please

43:59

do leave a leave

44:01

a review with

44:03

share it with

44:05

your we can we

44:08

can help people

44:10

drive meaningful business

44:12

transformation. And as always, always,

44:14

like if you'd

44:16

like to dive

44:19

deeper and learn

44:21

more about us

44:23

here at Insight222,

44:25

follow us on on

44:27

explore our resources

44:29

resources at .com, or

44:32

subscribe to our -weekly

44:34

newsletter at my .com.

44:36

all That's all for now. Thank you

44:39

for tuning in we'll we'll be back

44:41

next week with another episode of of

44:43

HR Leaders H.R. Leaders Until then, take

44:45

care take stay well. well.