Governments are racing to deploy AI. Most have the foundations, but struggle to translate them into faster, smarter execution without losing public trust. BCG's Miguel Carrasco and Daniel Selikowitz break down what separates the countries getting this right from those still stuck — and what every leader can take from their example.

You’ll Learn:

Why AI governance frameworks are quietly becoming the biggest barrier to AI deployment

How to tell the difference between necessary caution and unnecessary bureaucracy

How and why some countries are already getting AI governance right

Learn More:

Trust Imperative 5.0: https://on.bcg.com/3QO3fOm

BCG’s Latest Thinking in Public Sector: https://on.bcg.com/4f69kis

Meet the Experts

Miguel Carrasco, BCG Managing Director & Senior Partner: https://on.bcg.com/4gGH8Uz

Daniel Selikowitz, Managing Director & Partner: https://on.bcg.com/4oMRKTY

Watch The So What from BCG on YouTube: https://www.youtube.com/playlist?list=PLMJgyXjV5gMI9JV-GcF_D1Y6zyf1Eab_0

Chapters

(0:00) AI Is Already in Government. Here's Where You'll Find It.

(3:12) Public vs. Private: Why the Stakes Are Different

(4:40) What Leaders Are Saying Across Industries

(6:15) What Role Does Fear Play in AI Adoption?

(8:14) Inside Governments Deploying AI

(9:55) How Are Governments Governing AI?

(11:56) How Do Countries Compare on AI Adoption?

(13:37) Useful Caution Vs. Unnecessary Bureaucracy in AI Implementation

(15:42) Who Is Keeping Pace with AI Innovation?

(16:46) What Private Sector Leaders Can Learn from Governments

(19:34) What Should Leaders Do Now?

This podcast uses the following third-party services for analysis:

Podtrac - https://analytics.podtrac.com/privacy-policy-gdrp

Speaker: 00:00:00

- What we wanted to achieve

Speaker: 00:00:01

through this report was really look at

Speaker: 00:00:05

how are governments applying risk

Speaker: 00:00:08

and assurance frameworks in practice.

Speaker: 00:00:11

The last edition that we did looked at AI

Speaker: 00:00:14

and the extent to which

AI could help accelerate

Speaker: 00:00:18

and continue to sort of

build trust in government.

Speaker: 00:00:21

And this time around, what

we wanted to focus on was

Speaker: 00:00:24

the voice of the practitioners

Speaker: 00:00:27

and public servants within government

Speaker: 00:00:28

who have been trying

to build and deploy

Speaker: 00:00:32

these sorts of use cases and

applications in practice.

Speaker: 00:00:36

- If I were to describe

the archetypal mood

Speaker: 00:00:38

of these discussions

that we're having,

Speaker: 00:00:40

it's half extreme optimism and excitement

Speaker: 00:00:44

and half powerful frustration.

Speaker: 00:00:47

And the conversation often

begins with, "Here are all

Speaker: 00:00:50

of the things that I can

see that are possible,"

Speaker: 00:00:52

and then quickly we'll segue into,

Speaker: 00:00:54

"but we've had a pilot that's

been stuck in production

Speaker: 00:00:58

for 12 months," or "We have

a package of programs,

Speaker: 00:01:02

but they're caught up

in a risk process,

Speaker: 00:01:04

and we just can't

get those approved."

Speaker: 00:01:06

- Welcome to "The So What from BCG,"

Speaker: 00:01:08

the podcast exploring the big ideas

Speaker: 00:01:09

shaping business, the

economy, and society.

Speaker: 00:01:12

I'm Georgie Frost.

Speaker: 00:01:14

Governments around the world

are investing heavily in AI

Speaker: 00:01:17

with the promise of better

services, faster decisions,

Speaker: 00:01:20

and higher productivity.

Speaker: 00:01:22

But the challenge of turning AI ambition

Speaker: 00:01:24

into real-world delivery

isn't unique to governments.

Speaker: 00:01:28

Organizations everywhere

are wrestling with

Speaker: 00:01:30

how to scale these systems responsibly,

Speaker: 00:01:32

consistently, and at speed.

Speaker: 00:01:35

So what can leaders in

every sector learn from

Speaker: 00:01:37

where countries are succeeding

and where they're struggling?

Speaker: 00:01:41

Well, joining me are

Miguel Carrasco,

Speaker: 00:01:43

member of BCG's

Responsible AI Council,

Speaker: 00:01:46

and Daniel Selikowitz,

Speaker: 00:01:47

who leads BCG's government

finance segment globally.

Speaker: 00:01:51

Miguel, Daniel, welcome.

Speaker: 00:01:54

Before we talk about your report,

Speaker: 00:01:56

just explain, if you would,

Speaker: 00:01:57

how AI is actually showing

up now in the public sector.

Speaker: 00:02:01

Where might we encounter it?

Speaker: 00:02:04

- So I think AI is already there

Speaker: 00:02:07

in the public service around the world.

Speaker: 00:02:10

Sometimes it's there in ways

that are clearly visible

Speaker: 00:02:13

to citizens or to businesspeople.

Speaker: 00:02:15

That might look like

a chatbot on a website

Speaker: 00:02:18

or an agentic smart search that helps you

Speaker: 00:02:21

to navigate your tax

compliance obligations

Speaker: 00:02:24

or healthcare payer benefits.

Speaker: 00:02:27

In other cases, AI shows up

in ways that are invisible

Speaker: 00:02:31

to citizens but no less important

Speaker: 00:02:33

and valuable to those

in the public service.

Speaker: 00:02:36

That might look like summarization

Speaker: 00:02:38

or synthesis of a complex policy document.

Speaker: 00:02:41

It might look like call

transcription for an agent

Speaker: 00:02:44

working in a government contact center

Speaker: 00:02:46

interacting with citizens.

Speaker: 00:02:48

So there's many different ways,

Speaker: 00:02:49

just as there are in the private sector,

Speaker: 00:02:51

that AI is showing up in government.

Speaker: 00:02:53

- Yeah, I think what we're

hoping to see though is more

Speaker: 00:02:58

of the citizen-facing

and direct engagement

Speaker: 00:03:02

where, you know, it

could really help improve

Speaker: 00:03:07

and make it easier for citizens

Speaker: 00:03:09

to navigate sometimes the

complexity of government.

Speaker: 00:03:12

- Daniel, you mentioned there

about the private sector.

Speaker: 00:03:15

Where does it show up in a way

that's, that's quite similar,

Speaker: 00:03:18

and where does it differ do you think?

Speaker: 00:03:20

- It's a great question.

Speaker: 00:03:21

I mean, I think in terms

of the actual use cases

Speaker: 00:03:24

and the value that can

be unlocked through AI,

Speaker: 00:03:27

there are far more

similarities than differences

Speaker: 00:03:29

between the public and private sector.

Speaker: 00:03:31

So citizen-facing

applications like the ones

Speaker: 00:03:35

that Miguel mentioned or

that I discussed--

Speaker: 00:03:37

chatbots, smart search,

Speaker: 00:03:39

ways of streamlining

the client experience--

Speaker: 00:03:43

those are equally applicable

between government

Speaker: 00:03:46

and, say, a large financial

institution or a telco.

Speaker: 00:03:49

I think where there is the

most obvious difference is

Speaker: 00:03:52

what's at stake when things go wrong.

Speaker: 00:03:55

For those who are working

in private corporations,

Speaker: 00:03:58

there is more room, I think, to test,

Speaker: 00:04:01

to try new things with AI,

Speaker: 00:04:03

to learn from failures,

and to pivot.

Speaker: 00:04:05

And there may, of course,

be consequences

Speaker: 00:04:07

for customers and for shareholders,

Speaker: 00:04:10

but those are generally

somewhat contained.

Speaker: 00:04:13

In a government context,

Speaker: 00:04:14

there's a lot more at

stake if things go wrong,

Speaker: 00:04:16

and there can be,

of course, widespread

Speaker: 00:04:18

and quite deleterious

outcomes if AI makes mistakes

Speaker: 00:04:23

with eligibility

for benefits programs

Speaker: 00:04:25

or with the quantum of

a benefit that's paid.

Speaker: 00:04:27

So I think, understandably, citizens

Speaker: 00:04:30

and governments want to hold

a higher bar when it comes

Speaker: 00:04:33

to ensuring that we're

being thoughtful

Speaker: 00:04:35

about where, how, why

AI is being used.

Speaker: 00:04:40

- I want to dig into

that in more detail,

Speaker: 00:04:41

but before I do, you speak to

leaders across the board,

Speaker: 00:04:45

private, public sector, both

of you do, across the world.

Speaker: 00:04:49

What are they saying to you?

Speaker: 00:04:51

What is top of mind, biggest concerns,

Speaker: 00:04:53

greatest opportunities, Miguel?

Speaker: 00:04:56

- I think what we're seeing

at the moment is, you know,

Speaker: 00:04:59

leaders are excited about the opportunity

Speaker: 00:05:04

that AI could have,

both in terms of

Speaker: 00:05:08

improved services.

Speaker: 00:05:12

It could help in terms of

even designing policies

Speaker: 00:05:16

and programs in government.

Speaker: 00:05:17

The thing they're trying

to navigate is

Speaker: 00:05:19

how to move forward responsibly

Speaker: 00:05:24

without being overly cautious.

Speaker: 00:05:26

And the challenge they're

having is navigating some

Speaker: 00:05:30

of the frameworks and processes

Speaker: 00:05:32

and tools that have been put in place,

Speaker: 00:05:34

which are sometimes overlapping

Speaker: 00:05:37

or inconsistent or unclear.

Speaker: 00:05:40

- Daniel, what are leaders saying to you?

Speaker: 00:05:42

- I think that's right.

Speaker: 00:05:43

If I were to describe the archetypal mood

Speaker: 00:05:46

of these discussions

that we're having,

Speaker: 00:05:48

it's half extreme optimism and excitement

Speaker: 00:05:51

and half powerful frustration.

Speaker: 00:05:54

And the conversation often

begins with, "Here are all

Speaker: 00:05:58

of the things that I can

see that are possible,"

Speaker: 00:06:00

and then quickly we'll segue into,

Speaker: 00:06:02

"but we've had a pilot

that's been stuck

Speaker: 00:06:04

in production for 12 months"

Speaker: 00:06:07

or "We have a package of programs

Speaker: 00:06:09

that we'd really love to see implemented,

Speaker: 00:06:11

but they're caught up

in a risk process,

Speaker: 00:06:13

and we just can't

get those approved."

Speaker: 00:06:15

- It's interesting, Daniel,

when you were saying that,

Speaker: 00:06:17

you said, "50% excitement,"

Speaker: 00:06:18

and in my head and you went "50%,"

Speaker: 00:06:20

and I said "fear" in

my head, "trepidation,"

Speaker: 00:06:23

and you went,

"no, frustration."

Speaker: 00:06:24

So there's not an element of,

you know, trepidation, fear;

Speaker: 00:06:28

it's just frustration or excitement?

Speaker: 00:06:30

- I think certainly

there is trepidation,

Speaker: 00:06:33

and maybe that's implicit

in both of those things.

Speaker: 00:06:36

So, you know, certainly I think in,

Speaker: 00:06:38

in government in particular,

Speaker: 00:06:39

people do keenly feel the anxiety

Speaker: 00:06:43

around what could go wrong

Speaker: 00:06:44

and, of course, the

responsibility to citizens

Speaker: 00:06:47

and to taxpayers of doing

things in the right way.

Speaker: 00:06:50

But I think that the view is generally,

Speaker: 00:06:53

and we'll get into it

soon I'm sure, that a lot

Speaker: 00:06:56

of those fears have been

very well documented

Speaker: 00:06:59

and inculcated in frameworks

and governance mechanisms,

Speaker: 00:07:03

so I don't think there's too

many people that we speak with

Speaker: 00:07:05

who feel that those fears are

not being adequately looked at

Speaker: 00:07:08

and addressed.

Speaker: 00:07:09

The, the concern I think

is more whether there's,

Speaker: 00:07:12

we're erring too much on the side of fear.

Speaker: 00:07:14

- In some of our research

that we've done as part

Speaker: 00:07:19

of BCG's global digital government survey,

Speaker: 00:07:21

we've asked people questions

Speaker: 00:07:24

about their usage of AI

Speaker: 00:07:26

and also whether they see the

benefits outweighing the risk.

Speaker: 00:07:31

And I think what, what we can see

Speaker: 00:07:35

through some of that data is

Speaker: 00:07:36

that the more that

people use and adopt AI,

Speaker: 00:07:40

the less fearful they become

Speaker: 00:07:43

and the more they sort of

understand the potential

Speaker: 00:07:46

and the capabilities,

Speaker: 00:07:47

and some of the, you

know, the trepidation

Speaker: 00:07:51

or the fear that they

might feel of the unknown

Speaker: 00:07:53

and uncertainty dissipates

as maturity increases

Speaker: 00:07:57

and people sort of adopt and use it.

Speaker: 00:08:00

And I think that's also

what we're finding, too,

Speaker: 00:08:03

in government,

in the public sector.

Speaker: 00:08:04

The more that public servants

embrace the technology

Speaker: 00:08:08

and use the technology

as part of their work,

Speaker: 00:08:11

some of the fear factor disappears.

Speaker: 00:08:14

- Well, let's talk a bit more

Speaker: 00:08:16

about the latest BCG

"Trust Imperative 5.0" report.

Speaker: 00:08:21

You looked across

a range of countries

Speaker: 00:08:23

at how governments are building

Speaker: 00:08:24

and applying AI governance.

Speaker: 00:08:26

Tell me more about it.

Speaker: 00:08:28

What were you looking for,

Miguel? What did you find?

Speaker: 00:08:32

- What we wanted to achieve

Speaker: 00:08:33

through this report was really look at

Speaker: 00:08:37

how are governments applying

Speaker: 00:08:40

risk and assurance frameworks in practice.

Speaker: 00:08:43

So in the Trust Imperative

series, together

Speaker: 00:08:48

with Salesforce, we've been

looking at the question

Speaker: 00:08:51

of the relationship between

citizens and government

Speaker: 00:08:55

and, in particular, some of

the things that help build

Speaker: 00:08:58

and erode trust in government.

Speaker: 00:09:00

In the previous editions

of the series,

Speaker: 00:09:03

we've looked at the importance

Speaker: 00:09:05

of good customer service experience,

Speaker: 00:09:08

the extent to which personalization

Speaker: 00:09:11

and other things sort of

drive trust in government.

Speaker: 00:09:15

And the last edition

that we did looked

Speaker: 00:09:18

at AI and the extent

Speaker: 00:09:19

to which AI could help

accelerate some of that.

Speaker: 00:09:23

This time around, what

we wanted to focus on was

Speaker: 00:09:27

not so much the voice of the citizen

Speaker: 00:09:29

but the voice of the practitioners

Speaker: 00:09:32

and public servants within government

Speaker: 00:09:34

who have been trying to build and deploy

Speaker: 00:09:37

these sorts of use cases and

applications in practice.

Speaker: 00:09:41

And the sort of questions

that we were looking at was,

Speaker: 00:09:44

you know, these risk assurance frameworks

Speaker: 00:09:46

that have been established,

are they working in practice?

Speaker: 00:09:51

What's working well?

What's not working well?

Speaker: 00:09:53

How could they be improved?

Speaker: 00:09:55

- What's the common pattern

in how governments are trying

Speaker: 00:09:57

to govern AI?

Speaker: 00:09:58

- So I think the headline message

Speaker: 00:10:00

that we found in the report was

Speaker: 00:10:03

that many governments

have already put in place

Speaker: 00:10:06

the foundational elements,

so principles

Speaker: 00:10:11

for ethics and transparency,

frameworks,

Speaker: 00:10:16

risk assessment models,

Speaker: 00:10:18

accountable official roles,

and things like that.

Speaker: 00:10:23

The challenge, I think, has been more

Speaker: 00:10:24

at the operational level.

Speaker: 00:10:25

So all of the practitioners

Speaker: 00:10:28

and people who are trying

to apply these frameworks

Speaker: 00:10:30

and tools told us that they are

Speaker: 00:10:34

sometimes facing challenges

with the lack of clarity

Speaker: 00:10:38

or inconsistency

Speaker: 00:10:40

or ambiguity in definitions

Speaker: 00:10:43

where roles have been established

Speaker: 00:10:45

but the accountabilities

have been unclear,

Speaker: 00:10:49

the processes are not very well defined,

Speaker: 00:10:51

and sometimes they have

to navigate quite a lot

Speaker: 00:10:54

of different requirements.

Speaker: 00:10:56

One of the practitioners

told us that they had

Speaker: 00:10:58

in that process, you know,

Speaker: 00:11:00

to navigate 71 different

points in the process

Speaker: 00:11:03

where people were asking for the same

Speaker: 00:11:05

or similar information.

Speaker: 00:11:07

The Japanese government, for example, has

Speaker: 00:11:11

a multistage process for AI,

Speaker: 00:11:15

which has four different stages,

Speaker: 00:11:17

initially with a sandbox or pilot,

Speaker: 00:11:20

then moving to a control deployment,

Speaker: 00:11:24

then moving to broader scale

with stronger oversight,

Speaker: 00:11:27

and then finally ongoing monitoring

Speaker: 00:11:29

with very clear triggers for reassessment.

Speaker: 00:11:32

- Can I ask what countries

you were looking at?

Speaker: 00:11:36

- So for this study, we

conducted interviews

Speaker: 00:11:39

with people in 10 different

countries, some in Europe,

Speaker: 00:11:44

in the US and Asia Pacific.

Speaker: 00:11:47

And that was also supplemented by some

Speaker: 00:11:49

of the research from BCG's

global citizen survey,

Speaker: 00:11:51

which actually covers

40 different countries

Speaker: 00:11:54

around the world.

Speaker: 00:11:55

- So a really broad spectrum

Speaker: 00:11:58

of countries, I suppose, across the world.

Speaker: 00:12:01

Where were there similarities

in the challenges?

Speaker: 00:12:03

Where were there things

that were perhaps unique

Speaker: 00:12:06

to regions, Daniel?

Speaker: 00:12:08

- If I were to characterize

some of the common points,

Speaker: 00:12:11

firstly, I think admirably most

Speaker: 00:12:14

of these governments had developed

Speaker: 00:12:15

AI assurance frameworks

quite some time ago.

Speaker: 00:12:19

And in that sense, they were

proactive in getting ahead

Speaker: 00:12:22

of this technology and thinking

Speaker: 00:12:23

through some of the risks and the benefits

Speaker: 00:12:25

and how to manage that.

Speaker: 00:12:26

But therefore, in most

cases, with few exceptions,

Speaker: 00:12:30

these frameworks had not

been meaningfully updated

Speaker: 00:12:33

to reflect just how far

Speaker: 00:12:34

and how quickly the

technology has developed.

Speaker: 00:12:37

We all know from our own

experience as consumers

Speaker: 00:12:40

or in corporations

that we work in,

Speaker: 00:12:42

the technology is developing

every week if not every day,

Speaker: 00:12:46

and that's not how

Speaker: 00:12:47

these governance frameworks have evolved.

Speaker: 00:12:49

There are some,

for instance Singapore,

Speaker: 00:12:51

that have grappled directly,

for instance, with agentic AI,

Speaker: 00:12:55

but for the most part, these

frameworks predate a lot

Speaker: 00:12:58

of the frontier models

Speaker: 00:12:59

and other technologies

that exist currently.

Speaker: 00:13:02

I think the other main commonality,

Speaker: 00:13:03

as Miguel referenced, is

Speaker: 00:13:05

that they tend to be

fairly broad brush

Speaker: 00:13:08

across jurisdictions in terms

of how they think about risk.

Speaker: 00:13:12

So the sorts of risk tiering

that you would expect

Speaker: 00:13:15

to see here that would

enable much faster progress

Speaker: 00:13:18

and movement on relatively

straightforward cases

Speaker: 00:13:22

don't exist so much.

Speaker: 00:13:24

One senior official

quoted in the report,

Speaker: 00:13:27

who I think could have

been speaking for many

Speaker: 00:13:29

of the countries that

we talked about, said,

Speaker: 00:13:31

"We are exquisitely governing

Speaker: 00:13:33

very basic, low-risk AI use cases

Speaker: 00:13:35

to within an inch of their lives."

Speaker: 00:13:37

- How do you tell the difference between

Speaker: 00:13:40

what is genuinely useful caution

Speaker: 00:13:43

and what is just unnecessary

bureaucracy, overgovernance

Speaker: 00:13:47

getting in the way of yourself?

Speaker: 00:13:49

- What looks like very onerous

Speaker: 00:13:51

and unnecessary governance can later prove

Speaker: 00:13:53

to have been essential

and vice versa.

Speaker: 00:13:56

I think, in general, from

what we have observed,

Speaker: 00:13:59

what you want to see in a

good governance framework is,

Speaker: 00:14:02

first of all, that there

is some delineation

Speaker: 00:14:04

based on the type of use case,

Speaker: 00:14:07

the specific example or application of AI,

Speaker: 00:14:10

and how much intrinsic risk

and complexity there is.

Speaker: 00:14:13

So is AI actually being used to make

Speaker: 00:14:15

or meaningfully inform decisions

Speaker: 00:14:17

that have significant import,

Speaker: 00:14:19

or is it synthesizing documents

Speaker: 00:14:21

for internal discussions

or transcribing calls

Speaker: 00:14:24

as just another reference point to inform

Speaker: 00:14:26

internal decision-making

and consideration?

Speaker: 00:14:29

Of course, it's always important

Speaker: 00:14:31

and it always needs some kind

of governance and oversight,

Speaker: 00:14:34

but you would want to see

different levels of process,

Speaker: 00:14:38

different levels of onerousness

Speaker: 00:14:41

in terms of approval

depending on those factors.

Speaker: 00:14:44

- How do you get the

balance right there, Miguel?

Speaker: 00:14:45

What does, I suppose, the

gold standard here look like

Speaker: 00:14:49

in an ideal world?

Speaker: 00:14:51

- The challenge, I think,

has been the definition

Speaker: 00:14:55

or sort of lack of

clarity about how

Speaker: 00:14:59

to assess whether something is

indeed low risk, medium risk,

Speaker: 00:15:03

or high risk and the

judgment that's required

Speaker: 00:15:07

from public servants and officials.

Speaker: 00:15:10

And that guidance hasn't always

Speaker: 00:15:12

necessarily been very clear.

Speaker: 00:15:15

We heard of a very good example

Speaker: 00:15:16

in the New South Wales government,

Speaker: 00:15:19

which recently took its risk triage tool,

Speaker: 00:15:24

which required sort of

very specialized expertise

Speaker: 00:15:27

and a lot of data and evidence

Speaker: 00:15:30

and took sort of more than 40

hours on average to complete,

Speaker: 00:15:35

and I've simplified it now down

Speaker: 00:15:36

into a process that people can do

Speaker: 00:15:39

and can complete it in like 15 minutes.

Speaker: 00:15:42

- Daniel, can you tell me a bit more

Speaker: 00:15:43

about the Singapore example?

Speaker: 00:15:45

- Sure. So Singapore published in 2019

Speaker: 00:15:50

its Model AI Governance Framework,

Speaker: 00:15:52

but since then they've made

a bunch of changes and updates

Speaker: 00:15:55

that I think encapsulate

the point we're making

Speaker: 00:15:58

around reflecting shifts

in the technology.

Speaker: 00:16:00

So a year later in 2020,

they updated that framework

Speaker: 00:16:03

to translate the

high-level principles

Speaker: 00:16:05

into practical guidance

Speaker: 00:16:07

for folks working

in the public service.

Speaker: 00:16:09

In 2022, they added AI Verify

as a testing framework

Speaker: 00:16:13

and toolkit to make it

even more practical.

Speaker: 00:16:16

In 2024, they published a

new governance framework

Speaker: 00:16:19

that specifically covered

generative AI.

Speaker: 00:16:21

And now this year,

they've launched

Speaker: 00:16:22

a Model AI Governance

Framework for Agentic AI.

Speaker: 00:16:26

So while there's no perfect approach,

Speaker: 00:16:28

I think the fact that Singapore

has continually evolved

Speaker: 00:16:31

its governance mechanism,

both to make it more practical

Speaker: 00:16:35

and specific for practitioners

Speaker: 00:16:38

and to reflect changes in

this fast-moving technology,

Speaker: 00:16:41

is admirable and something

Speaker: 00:16:43

that other jurisdictions

can certainly learn from.

Speaker: 00:16:46

- So what does this mean for

large organizations outside

Speaker: 00:16:49

of government trying to

scale AI beyond pilots?

Speaker: 00:16:51

What can they learn from governments?

Speaker: 00:16:55

- Well, I think a lot of

the same lessons apply.

Speaker: 00:16:57

There needs to be the

right focus on the risk

Speaker: 00:16:59

of action and inaction.

Speaker: 00:17:01

There needs to be the right

tiering of different risks

Speaker: 00:17:04

with respect to AI.

Speaker: 00:17:06

And most importantly, it needs

Speaker: 00:17:07

to be really grounded

in the practicalities

Speaker: 00:17:09

of individual use cases.

Speaker: 00:17:12

What are we actually talking about?

Speaker: 00:17:14

For instance, in a contact center,

Speaker: 00:17:17

there is potentially a big

difference between an AI IVR

Speaker: 00:17:21

that's actually answering calls

Speaker: 00:17:23

and interacting with customers as opposed

Speaker: 00:17:26

to AI-enabled transcription

that is keeping a record

Speaker: 00:17:30

of what was discussed for later reference.

Speaker: 00:17:32

You know, those could be

lumped together very easily

Speaker: 00:17:34

as AI in the contact center,

Speaker: 00:17:36

but they're really very

different in practice.

Speaker: 00:17:39

- Miguel?

Speaker: 00:17:40

- Well, we've covered

a few of them,

Speaker: 00:17:42

but I wanted to maybe just

leave you with three others.

Speaker: 00:17:46

So one is the,

Speaker: 00:17:50

the clarity around, sort

of, accountabilities

Speaker: 00:17:53

within the tech stack

Speaker: 00:17:57

and making sure that the, you know,

Speaker: 00:18:02

the questions are directed

to the right party overall.

Speaker: 00:18:07

And what I mean by that is,

Speaker: 00:18:09

you know, there's, there's things

Speaker: 00:18:10

that the business owners

are responsible for,

Speaker: 00:18:12

things that the, the LLM provider

Speaker: 00:18:16

or the large language

model, foundation model

Speaker: 00:18:20

should be accountable for.

Speaker: 00:18:21

The second thing that I

think is also relevant

Speaker: 00:18:23

for private sector is

investing in, in capability,

Speaker: 00:18:29

in literacy, in maturity.

Speaker: 00:18:31

So one of the things we've seen

Speaker: 00:18:32

leading organizations do is

Speaker: 00:18:34

put in place like a

certification framework

Speaker: 00:18:36

with different levels.

Speaker: 00:18:37

As people sort of become more

familiar with the technology,

Speaker: 00:18:41

they can, they sort of can do

Speaker: 00:18:43

the next level of certification,

Speaker: 00:18:46

and that maturity and awareness,

Speaker: 00:18:49

understanding,

literacy, et cetera,

Speaker: 00:18:51

helps in the adoption of technology

Speaker: 00:18:55

and diffusion throughout the organization.

Speaker: 00:18:57

And then lastly, measurement--

Speaker: 00:19:00

so how we track and measure

Speaker: 00:19:03

what we're implementing.

Speaker: 00:19:06

In terms of risk and assurance,

Speaker: 00:19:08

it's not just about sort

of measuring the activity,

Speaker: 00:19:11

but actually measuring

is it effective.

Speaker: 00:19:13

So are we actually reducing risk?

Speaker: 00:19:16

Are we reducing the number of escalations?

Speaker: 00:19:19

How many things are being

approved first time 'round?

Speaker: 00:19:23

And then on the benefits side,

Speaker: 00:19:25

are we actually

adopting the solutions?

Speaker: 00:19:29

You know, are people using them?

Speaker: 00:19:30

Are we getting the benefits?

Speaker: 00:19:31

Are we saving money?

Are we saving time?

Speaker: 00:19:34

- Finally, what is the now what--

Speaker: 00:19:36

the next immediate steps

Speaker: 00:19:38

that leaders can take, Daniel?

Speaker: 00:19:40

- So I think, very practically,

Speaker: 00:19:42

one thing is understanding properly

Speaker: 00:19:45

what your customers or

your citizens actually want

Speaker: 00:19:48

and expect when it comes to AI.

Speaker: 00:19:51

I think we tend to assume

sometimes that we know

Speaker: 00:19:54

what the level of comfort is,

Speaker: 00:19:55

but I think it's important,

as we have tried to with some

Speaker: 00:19:58

of our recent research,

dig into what citizens

Speaker: 00:20:01

and customers of your agency

Speaker: 00:20:03

or your business actually

want, fear, expect

Speaker: 00:20:06

when it comes to AI.

Speaker: 00:20:08

And the second point I would

make is getting very specific

Speaker: 00:20:11

in the way that you have discussions

Speaker: 00:20:13

in governance forums,

in board meetings,

Speaker: 00:20:16

in risk management sessions

about use cases in AI.

Speaker: 00:20:20

In some cases,

that means upskilling,

Speaker: 00:20:22

playing around more with the tools,

Speaker: 00:20:24

but really having a sense of

Speaker: 00:20:26

what are the differences

Speaker: 00:20:27

between different specific applications

Speaker: 00:20:30

and what would those mean

in terms of the level

Speaker: 00:20:33

of risk upside and downside

that we need to manage.

Speaker: 00:20:36

- Miguel?

Speaker: 00:20:37

- The agentic AI

Speaker: 00:20:40

that we're seeing at the

moment, you know,

Speaker: 00:20:43

none of the assurance

frameworks were really geared

Speaker: 00:20:46

for that development.

Speaker: 00:20:48

And agentic AI requires,

again, another,

Speaker: 00:20:53

a different approach given

that it is, you know,

Speaker: 00:20:56

involves a level of autonomy and,

Speaker: 00:20:58

and so providing the

guardrails around

Speaker: 00:21:01

what agents are allowed to do

Speaker: 00:21:02

or the, the delegation

rights that they have

Speaker: 00:21:06

and how you manage that diffusion

Speaker: 00:21:09

throughout the organization.

Speaker: 00:21:11

So I think it's imperative at

the moment that you keep to,

Speaker: 00:21:16

you keep iterating and refining

Speaker: 00:21:18

your risk and assurance models.

Speaker: 00:21:19

It's not once and done.

Speaker: 00:21:21

We have to keep reviewing

Speaker: 00:21:22

whether they are fit for purpose

Speaker: 00:21:25

and adapting them and updating them

Speaker: 00:21:28

as the technology continues to evolve.

Speaker: 00:21:31

- Daniel, Miguel, thank you so much

Speaker: 00:21:33

and to you for listening.

Speaker: 00:21:34

If you'd like to read

Speaker: 00:21:35

BCG's latest "Trust

Imperative 5.0" report,

Speaker: 00:21:39

you can find the link

in the show notes.