Gartner-首席數(shù)據和分析官如何領導數(shù)據科學和機器學習領域的技能提升計劃（英）

GartnerResearchHow

CDAO

s

Can

LeadUpskilling

Initiativesin

Data

Science

andMachine

LearningPeterKrensky15December2022How

CDAOs

Can

Lead

Upskilling

Initiatives

in

DataScienceand

Machine

LearningPublished15December

2022-ID

G00780993-

12min

readBy

Analyst(s):Peter

KrenskyInitiatives:ChiefDataandAnalyticsOf?cerLeadershipAs

thehiring

boom

for

data

science

talent

continues,

initiatives

toupskill

quantitativeprofessionalsremain

just

as

prominent.

CDAOsshould

use

this

high-level

guidanceto

help

develop

in-house

talentand

improve

data

science

and

machine

learning

literacy.OverviewKey

Findings■Machine

learningliteracy

remains

lowin

many

organizations;

concerted

educationand

culture

change

are

necessary

but

dif?cult

due

to

the

gap

between

datascientists’technical

expertise

and

businessusers’domain

knowledge.■Abundant

educational

opportunitiescombined

with

talent

acquisition

and

retentionchallenges

motivate

organizations

to

upskill

their

data

professionals

at

alllevelsofsophistication,especially

with

an

aim

to

grow

their

citizen

data

scientistpopulations.■An

overwhelming

number

of

toolsand

approachesare

available

to

expert

and

citizendata

scientists.CDAOsmust

navigate

a

vast

landscape

to

match

diverse

users

toappropriate

solutions

and

corresponding

educational

paths.RecommendationsCDAOsresponsible

for

analytics,

BI

and

data

science

solutions

should:■Raise

overall

data

science

and

machine

learningawareness,adoption

and

literacyby

providingcentralized

educational

resources

and

showcasingexisting

use

casesand

success

stories,

both

internal

and

external.Gartn

er,

Inc.

|

G00780993Page

1

of

12■Identify

citizen

data

science

(CDS)

candidates

in

their

organization

by

creating

aninventory

of

in-house

skills

and

ambitions.

Match

upskillingpaths

to

the

variousbackgrounds

and

aspirations

of

CDS

candidates.Lookto

build

interconnectedcommunitiesof

data

scientists,citizen

data

scientistsand

other

ML

pipelinestakeholders.■Buildrepeatable

and

sustainable

education

programsby

designingdifferentupskillingroadmaps

for

average

consumersof

analytics,

CDS

candidates

and

expertdata

scientists.Strategic

Planning

AssumptionsBy

2024,

75%

of

organizations

willhave

established

a

centralized

data

and

analytics(D&A)

centerof

excellence

to

supportfederated

D&A

initiatives

and

prevententerprisefailure.By

2025,

50%

of

data

scientistactivitieswillbe

automatedby

arti?cial

intelligence,easingthe

acute

talent

shortage.IntroductionThe

talent

gap

in

data

science

may

never

be

fully

closed

—

but

it

can

be

narrowed.ManyGartner

clientsstill

reportdif?culty

?nding

and

attracting

talent.Retainingproductive

datascientistsfor

longtenures

is

alsoa

major

challenge.CDAOsneed

to

learn

howto

builddevelopmentpaths

for

experts

and

supportbuddingcitizen

data

scientistswith

the

righttools,trainingand

structure

(see

Note

1

for

CDAO

role

de?nition

andNote

2

for

a

de?nitionof

citizen

data

scientist).Even

organizations

that

build

highvolumes

of

complex

andaccurate

modelshave

to

diligently

foster

data

literacy

and

proper

adoption

of

solutions.For

example,CDAOswho

invest

much

more

in

resources

and

talent

are

1.8x

moreeffective

and

successful

with

their

data

literacy

programs.

1

Upskillingshouldbepromoted

throughout

the

organization,

with

targeted

trainingfor

a

selectgroup

ofindividuals

—

both

experts

and

new

CDS

candidates

—

as

well

as

general

education

for

allconsumersof

analytics.Thegreatest

opportunity

for

most

organizations

to

grow

theirtalent

poolfor

data

scienceand

machinelearning

is

through

theupskilling

of

current

staff.Gartn

er,

Inc.

|

G00780993Page

2of

12Mostsuccessful

upskillingincorporatessome

formaleducation

and

training,but

often

themostimpactfullearninghappens

“on

the

job”

duringthe

completion

of

a

new

projectorthe

assumption

of

new

analyticalduties.Many

self-identi?ed

data

scientistshave

fewerthan

?ve

yearsof

experience

working

with

ML.

Leadersshouldexpectsome

growingpains

and

steep

learningcurves.Timelines

for

developmentshouldbe

?exible

wherepossible,and

early

projectsshouldbe

limited

in

scope

and

risk.AnalysisRaiseMachine

Learning

Literacy

Among

Consumers

and

PromoteCollaboration

With

Data

ScientistsRaisingthe

level

of

discourse

arounddata

science

and

machine

learningis

the

?rststeptoward

upskillingyour

workforce.Begin

by

ensuring

that

allline-of-business(LOB)

leadersand

decision

makershave

a

clear

understanding

of

howdata

scientistscreate

value.

Thiscan

be

done

through

simple

workshopsor

exerciseswhere

data

scientistsand/or

otherparticipants

begin

to

question

some

existing

KPIs,data

or

metrics.Aspiring

modelbuildersand

heavy

consumersshouldhave

a

foundationalunderstanding

of

the

machinelearninglife

cycle

—

particularly

data

preparation,feature

engineering,testing

and

training,and

deployment

(seeFigure

1).

Emphasize

to

consumersthat

analytics

consumersare

thekey

to

generating

value

from

the

work

of

data

science

teamsand

givingfeedbackfor

thefuture

projects

and

model

iterationsthat

willgothrough

this

cycle.Gartn

er,

Inc.

|

G00780993Page

3of

12Figure

1.

The

Machine

LearningLife

CycleHelp

enthusiastic

individuals

become

familiar

with

the

basics

of

several

machine

learningtechniques,such

as

regression,clustering

and

classi?cation.

Encourage

or

require

datascientiststo

regularly

holdopen

sessionsto

discussa

current

project

(in

layperson’sterms)

or

introduce

anaspectof

data

science

they

are

passionate

about.

Considergami?cation

of

developmentthat

encouragesupskillingindividuals

to

attend

regulartraining,engage

new

subjectsor

enterintohealthy

competition

with

peers.Create

a

Skills

Inventory

to

Identify

In-House

CDS

Candidates

and

FosterInterconnected

Data

Science

CommunitiesTalentalignment,careerdevelopmentand

retaining

talent

are

the

major

leadershipdemandsneeded

for

sustainingsuccessful

upskillinginitiatives.

The

role

of

the

citizendata

scientistpresents

unique

demands

for

CDAOsin

termsof

talent

recognition

anddevelopment.CDAOsneed

to

understandwhat

the

CDS

persona

is

(seeFigure

2)

andrecognize

the

skills

and

pro?lesthat

make

for

goodCDS

candidates.Gartn

er,

Inc.

|

G00780993Page

4of

12Figure

2.

BreakingDown

the

CDS

PersonaTalkto

potential

candidates

to

gauge

their

interest

and

aspirations

aroundcareersin

datascience.Getcandidates

to

complete

self-evaluations

of

their

backgrounds

so

that

adetailed

inventory

of

skills

can

be

established.This

inventory

willbe

invaluable

fordesigningtrainingprogramsand

makingtechnology

investments.The

mostpromisingcandidates

for

upskillingoftenhave

educational

and

professional

backgrounds

inphysics,chemistry,biology,actuarialscience,

computer

science,

engineering,?nance,economicsand

mathematics.Use

that

data

from

the

inventory

to

identify

communitieswithin

IT

and

businessunits

andacrossbusinessfunctions.

Support

and

grow

these

communitiesvia

formalmentorship,structured

collaboration

and

strategic

software

investments.Lookfor

leadersandpractitionersthatwillkeepthesecommunitiesactiveandtalkingtoeachother.

Formoreonspreadingupskillinginitiativesthroughcommunities,seeUseSocialIn?uencerPrinciplestoBuildCommunitiesandIncreaseDataandAnalyticsAdoption.Gartn

er,

Inc.

|

G00780993Page

5

of

12Design

Upskilling

Roadmaps

for

Expert

CDS

Candidates

and

AnalyticsConsumers

and

Support

the

Continuous

Developmentof

ExpertsConductfoundationalupskillinginitiatives

arounddata

science

and

machine

learningwithin

twogroupssimultaneously:

CDS

candidates

and

analytics

consumers(seeFigure3).Figure

3.

Simultaneous

Upskilling

RoadmapThe

Path

for

CDS

CandidatesThe

CDS

candidate

group

willembarkon

a

structured

upskillingprogram

acrossthreestages.Stage

1:

ApproachSelection

and

Formal

TrainingRegardlessof

the

selected

approach,all

CDS

candidates

shouldset

out

with

a

clear

visionof

the

skills

they

willneed

to

acquire

and

hone.CDAOsshouldtasktheir

teamsto

workwith

HR

to

design

career

paths

for

citizen

data

scientists,includingdifferentspecializations.Gartn

er,

Inc.

|

G00780993Page

6of

12Formal

trainingcan

be

a

course,

bootcampor

tool

trainingprogram.CDAOsand

CDScandidates

interested

in

pursuingupskillingthrough

free

study

shouldexplore

themassive

open

online

courses(MOOCs)

offered

by

organizations

such

as

Coursera,DataCamp,

edX,Udacity

and

Udemy.CDAOsshouldensure

that

the

numeroustutorialsavailable

on

YouTube,Kaggle

Learnand

elsewhere

are

investigated.Leadersand

CDScandidates

shouldinvestigate

local

in-person

options

as

well.When

overseeing

a

freestudy

program,develop

clear

incentivesand

milestonesfor

CDS

candidates

in

order

toaccurately

measure

their

progress.Classroom

learningwillserve

as

a

foundation

forextensive

project-based

learning.Stage

2:

ExperimentationandPrototypingOnce

CDS

candidates

complete

their

formaltraining,they

shouldbegin

experimentingwith

their

new

skills

and

designingprototypes.Leadersshouldprovide

sandboxenvironmentsfor

new

ML

practitionerswhere

userscan

work

with

desensitized

data

fortrial

and

error.

Whenever

possible,place

citizen

data

scientistsunder

the

mentorship

ofdata

scientistswho

can

review

their

work

and

provide

feedback.LeadersshouldensureCDS

candidates

also:■■Begin

honingthe

communication

skillsnecessary

for

successful

data

science

andevangelize

the

methodology

and

potential

of

their

work

with

machine

learning.Establish

regular

contactand

collaboration

with

analytics

consumersand

expertdata

scientists.■■■Examine

analytics

consumers’suggestionsfor

new

models.Share

their

domain

expertise

with

their

data

science

teammates.Initiate

a

reverseknowledge

transfer

to

?ll

gaps

among

even

the

mosttenured

datascientists.Stage

3:

DeliveryandOperationalizationThe

third

stage

of

CDS

upskillingis

the

delivery

and

operationalization

of

new

models.The

emerging

rolesof

chief

data

scientistand

ML

engineer

shouldtake

anactive

role

inshepherding

citizen

data

science

projectsfrom

experimentation

to

production.Citizendata

scientistsshouldalsopromote

utilization

among

analytics

consumersand

work

toestablish

continuous

feedbackto

improve

existing

modelsand

exchange

ideas

for

newones.Gartn

er,

Inc.

|

G00780993Page

7of

12Gartner

has

publishedresearch

that

further

explores

best

practices

for

managingcitizendata

scientists

andcommon

pitfalls

toavoid(see

Lessons

FromData

Scientists

on

TheirEducation

andCareer

Development).The

Path

for

Analytics

ConsumersThe

second

group

requiring

simultaneous

structured

upskillingis

consumersof

analyticsthroughout

the

organization

—

againacrossthree

stages.Stage

1:

Introduction

to

Data

Science

andMachine

LearningBegin

the

upskillingroadmapfor

the

average

consumer

with

introductory-level

educationon

data

science

and

machine

learning.

This

shouldcome

in

the

form

of

anexternal

guestspeaker

or

dedicated

time

with

aninternal

data

science

professional.Showconsumersinternal

case

studieswith

measurable

businessvalue

and

recruitthem

as

enthusiasticstakeholdersin

your

entire

data

science

initiative.

CDAOs

shouldview

this

phase

as

a

keyopportunity

to

establish

new

working

relationships

between

the

data

science

team

anddifferent

LOBs.Stage

2:

BrainstormIdeas

onMetrics

andAnalyticsFollowupthis

introductory

trainingwith

brainstorming

sessionson

metrics

that

analyticsconsumerswouldlike

to

betterpredictand

optimize.This

in

turn

shouldleadto

proposalsfor

investmentsin

packagedapplications

with

the

highest

ease

of

implementation

anduse.At

this

stage,put

analytics

consumersin

contactwith

CDS

candidates

to

share

their

ideasand

hear

what

may

be

possible

upon

the

completion

of

the

upskillinginitiative.

Analyticsconsumerscan

help

brainstorm

ideas,and

the

twogroupscan

then

collaborate

onfeasibility

and

prioritization.Stage

3:

Test

the

Functionalityof

the

ModelsFinally,

analytics

consumersshouldreceive

and

utilize

the

?rstroundof

modelscreatedby

citizen

data

scientists.CDAOs

shouldtasktheir

team

members

to

monitor

theconsumption

of

new

modelsto

ensure

active

participation

at

the

LOB

level.Analyticsconsumersshouldoffer

feedbackto

data

science

teamson

the

strengthsandshortcomingsof

new

modelsand

participate

in

brainstorming

for

the

nextroundof

modeldevelopment.Gartn

er,

Inc.

|

G00780993Page

8

of

12Drive

Upskilling

of

Expert

Data

ScientistsSuccessful

data

science

requiresconstant

learningat

alllevels—

especially

amongexperts.Encourage

and

supporttheir

continued

knowledge

developmentin

allareasof

thefast-movingAIspace.Create

space

indata

science

professionals’schedulesforindependentlearning.Common

areasof

study

and

developmentfor

expertdata

scientistsinclude:■■■■■■■■■■■■DomainknowledgeCommunication

skillsStrategic

and

executive

skillsDeep

learningReinforcementlearningAugmented

data

science

and

machine

learningInternetof

Things

and

edge

computingComputer

visionNatural

language

processingMachine

learningoperations(MLOps)AItrust,fairness

and

explainabilityDigitalethicsIn

addition

to

hardskills

and

hot

topicsin

AI,

expertdata

scientistsespecially

bene?tfromleadership

and

humanskills

development.Data

scientists,particularly

junior

datascientists,tendtohavelessstrengthandexperienceinareassuchaspresentationsskills,mentorship,collaborationandprojectprioritization.Formoreonthistopic,seeAnExecutiveLeader’s

GuidetoStaf?ngEffectiveDataScienceTeams.Gartn

er,

Inc.

|

G00780993Page

9

of

12CDAOs

who

have

tenured

data

scientistsshouldcollaborate

with

these

individuals

tounderstandtheir

ambitions,then

design

appropriate

careerpaths.

Do

your

data

scientistswant

to

be

pure

practitionersor

develop

intomore

hybridroles?

Senior

or

chief

datascientistscan

step

intoa

role

primarily

focused

on

management

and

are

no

longer

day-to-day

practitioners,which

may

not

appealto

many

expertdata

scientists.Reserve

such

aleadershiprole

for

mature

data

science

operations

that

have

multiple

junior

datascientists

andan

establishedCDS

initiative.

For

more

on

this

topic,see

Roles

andSkills

toSupport

AdvancedAnalytics

andAI

Initiatives.EvidenceGartner

Chief

DataOfficer

Agenda

Surveyfor

2022:

This

study

was

conducted

to

exploreand

track

the

businessimpact

of

the

CDOrole

and/or

the

of?ce

of

the

CDOand

the

bestpracticesto

create

a

data-driven

organization.

The

research

was

conducted

online

fromSeptember

through

November

2021among496respondentsfrom

acrossthe

world.Respondentswere

required

to

be

the

highest

level

data

and

analytics

leader

in

theorganization:chief

data

of?cer,

chief

analytics

of?cer,

the

mostsenior

leader

in

IT

withdata

and

analytics

responsibilities,or

a

businessexecutive

such

as

chief

digitalof?cer

orother

businessexecutive

with

data

and

analytics

responsibilities.The

survey

sample

wasgleanedfrom

a

variety

of

sources(includingLinkedIn),with

the

greatestnumber

comingfrom

a

Gartner-curated

list

of

more

than4,519CDOs

and

other

high-level

data

andanalytics

leaders.

The

study

was

developedcollaboratively

by

Gartner

D&A

analysts

andthe

Primary

Research

Team(See

CDAOAgenda

2022:Focus

on

Value,Talent

andCulturetoPullAhead).Disclaimer:

Resultsof

this

study

do

not

representglobal

?ndings

or

the

marketas

a

wholebut

re?ect

sentimentof

the

respondentsand

companiessurveyed.Note

1:

CDAO

Role

DefinitionChief

data

and

analytics

of?cer

(CDAO)

refers

to

the

businessleadership

role

that

has

theprimary

enterprise

accountability

for

value

creation

by

means

of

the

organization’sdataand

analytics

assets,

and

the

data

and

analytics

ecosystem.Equivalent

titlesfor

this

roleare

chief

data

of?cer,

chief

analytics

of?cer

(if

the

CDAO

role

or

equivalent

is

not

in

theenterprise),chief/headof

data

and

analytics

and

other

variations.Gartn

er,

Inc.

|

G00780993Page

10

of

12Note

2:

Citizen

Data

ScientistA

citizen

data

scientistis

a

person

who

createsor

generatesmodelsthat

use

predictive

orprescriptive

analytics,

but

whose

primary

jobfunction

is

outside

the

?eld

of

statisticsandanalytics.

The

person

is

not

typically

a

member

of

an

analytics

team

(for

example,ananalytics

centerof

excellence)

and

doesnot

necessarily

have

a

jobdescription

that

listsanalytics

as

his

or

her

primary

role.

This

person

is

typically

in

a

line

of

businessoutside

ITand

outside

a

businessintelligence

(BI)

team.However,

an

IT

or

BI

professional

may

be

acitizen

data

scientistif

their

workon

analytics

is

only

a

secondary

role.Citizen

datascientistsare

“power

users”

who

are

able

to

use

simple

and

moderately

sophisticatedanalytics

applications

that

wouldpreviously

have

required

more

expertise.Recommended

by

theAuthorSome

documents

may

not

be

available

as

part

of

your

current

Gartner

subscription.Lessons

FromData

Scientists

on

Their

Education

andCareer

DevelopmentHype

Cycle

for

Data

Science

andMachine

Learning,20223Steps

toBuildandOptimize

a

Portfolioof

Analytics,Data

Science

andMachineLearningToolsTopTrends

in

Data

andAnalytics,2022Tool:

Data

Literacy

Personas

toDrive

a

Data-Driven

CultureRoles

andSkills

toSupport

AdvancedAnalytics

andAI

InitiativesGartn

er,

Inc.

|

G00780993Page

11

of

12?

2023

Gartner,Inc.and/oritsaf?liates.Allrightsreserved.Gartnerisa

registeredtrademarkofGartner,Inc.anditsaf?liates.ThispublicationmaynotbereproducedordistributedinanyformwithoutGartner's

priorwrittenpermission.ItconsistsoftheopinionsofGartner's

researchorganization,whichshoul