Latest / The Diary of a CFO / Agentic AI for CFOs, with Vivienne Wei, COO of Salesforce Unified Agentforce Platform

Transcript

Agentic AI for CFOs, with Vivienne Wei, COO of Salesforce Unified Agentforce Platform · The Diary of a CFO

0:00
Joining us today, Vivian Wei few executives possess her unique perspective,
0:04
COO of Unified agent force platform at Salesforce, leveraging technology
0:09
to create massive business value.
0:11
An angel investor spotting the next trillion dollar trends.
0:14
And a keynote speaker who knows exactly what keeps global CFOs up at night.
0:17
She's here to deliver the strategy you need today.
0:20
Welcome back to the Diary of a CFO Podcast, the podcast with the lessons,
0:24
challenges, and wins that shape their careers as well as their organizations.
0:28
I'm your host, Wassia Kamon, and today I'm super delighted
0:31
to have with me, Vivienne Wei.
0:33
Welcome to the show, Vivienne.
0:35
So happy to be here.
0:36
Thank you for having me, Wassia.
0:37
Of course.
0:38
You are the rock star.
0:40
So I just wanna dive in into your amazing career journey and really
0:45
understanding how you ended up in your current role and what were
0:49
some unexpected turns along the way.
0:52
Yeah, you know, my journey started in finance on Wall Street where, um.
0:58
Discipline around numbers, accountability really shaped my worldview.
1:04
And then after I went to business school, came back and I was
1:09
like, Hey, I really wanna go from being a agent to a principal.
1:15
So that's when I joined corporate where I'm like, Hmm, you know, in any
1:20
of the acquisitions I had done, done from my Wall Street days, the most
1:25
amount of value actually come from how you successfully integrate them.
1:30
So that's when I started with eBay initially.
1:34
Acquired multiple businesses around the world.
1:38
I was in this business unit in eBay, classifieds acquiring them.
1:42
And then also I spent a lot of time integrating them working
1:45
cross-functionally, whether it's sales, marketing, customer service,
1:50
support, finance, different departments.
1:53
And then I just really, really enjoy the working with.
1:57
Engineering technology people.
1:59
I think there's something about, you know, when you're finance, you love numbers.
2:03
When you work with engineers, they are really into the details.
2:08
Um, using their logical thinking, which actually landed.
2:11
That's how I landed my role, um, at Salesforce.
2:16
Nine years ago in data center planning and what I did at the time, you know,
2:22
we had some operations people, modern monitoring, our compute that supports
2:27
the cloud growth, and I was working with.
2:31
A group of, um, former Microsoft bankers and consultants that were
2:36
hired into Salesforce to scale up our data center operations.
2:42
And this was when I had my first experience with ai,
2:46
which was predictive ai.
2:47
How do we predict compute using.
2:51
Data science models and we, I remember we had daily standups
2:57
with a couple of data scientists inspecting the model in the black box.
3:05
Right?
3:05
And then like our team was like, what are the main drivers?
3:09
'cause you know, in finance when you do a forecasting model.
3:12
There's always like main drivers that change the model.
3:17
The parameters and the weighting of the para parameters are
3:21
so important in some ways.
3:24
Data science predictive model are very similar in that sense, right?
3:29
Because I studied math and statistics and like data science is, um, an evolution
3:35
of, you know what my husband likes to say, he's a PhD in neuroscience and ML
3:40
engineer and he's like, it's what breaks Excel and he likes to make fun of me.
3:47
Okay.
3:48
Because I was like, oh, I gotta do this spreadsheet and do it perfectly
3:51
with the perfect formatting.
3:53
'cause I remember I used to spend like nights and weekends making
3:57
the perfectly formatted model with, you know, blue color and red color.
4:02
And he is like, you know what ml I break all of the spreadsheets.
4:08
So that's kind of the advanced model of ai and.
4:13
What it really did was, if you think about the complexity of predicting data
4:19
center compute, you wanna make decisions on spending millions and billions of
4:25
dollars into scaling that capacity.
4:30
You have to get really precise, right?
4:32
And how you build that model is a combination.
4:37
It can't just be a black box, first of all.
4:40
It has to be a combination of art, science, and business logic.
4:47
And that's what I really enjoyed from kind of going from finance into operations
4:55
that leveraging AI for us to scale.
4:57
Wow.
4:58
And when you say art, science, and business logic, please break it down
5:02
because most people learned about AI when chat, GPT came on board.
5:08
We heard about your GPT and that's when people started talking like Main
5:12
Street starting hearing about ai, but you learned about it like years ago.
5:16
So I'm curious to hear how you really developed that approach of seeing
5:20
it in those three steps, but also what was your first impression when
5:24
you saw what predictive AI could do?
5:26
Left alone generative ai.
5:29
You know, predictably, in some ways, like it's forecasting, right?
5:32
Look at the history and then you forecast into the future.
5:36
Generative AI is very different 'cause all of a sudden you can develop new ideas.
5:42
And sometimes, like when I use chat GPT or Google Gemini, it gives me new
5:47
ideas that I didn't think about before.
5:50
Which I found was just super, super powerful.
5:53
'cause if you think about, you know, we always talk about
5:56
human working with AI agents.
6:00
If I have an assistant that's going to augment how I think about
6:07
a problem with new possibilities that make me a more productive
6:13
worker, a more effective leader.
6:18
Because now I'm able to take the different components, not only with
6:24
data, but also the human layer, right?
6:29
Because you know, for example, leading large scale change management is really at
6:35
the essence of a gent AI transformation.
6:38
At any large scale companies, I think of as 30% technology, 70% leadership.
6:46
The reason for that is because when you are trying to change how a
6:51
company is wired, how people work, it's easy to implement a system.
6:59
But how do you work with people who lead with emotions?
7:05
We're all humans, I mean.
7:07
We're all humans and we have emotions, we have fear, we have, we don't wanna
7:13
change, and that's just human nature.
7:17
How do you recognize that and work with people to figure out,
7:23
okay, actually help them realize it's so important to embrace ai.
7:29
Because it's going to make you more productive, more effective as you,
7:33
whether you're an individual contributor or you're a leader in an organization.
7:38
There's a lot of news in the market driving AI fear, but it's so false.
7:45
So I think of like as a leader, that is a huge responsibility We have.
7:50
How do I help my team experience the same AI delight moment that I have every day,
7:58
and incentize them, get them, motivate them to use AI and just test it out.
8:07
What's the worst thing that could happen?
8:09
You get an answer you don't like.
8:13
Especially in the corporate context.
8:14
We have Gemini as a tool.
8:16
We have Cursor for engineers, and I've orchestrated rhythm and
8:22
cadence for hundreds of engineers, thousands of engineers to come
8:28
together on a weekly basis.
8:30
It's not mandatory, but people just show up on a no meeting Thursday.
8:36
Bring their best recipes of how they're embracing ai and
8:42
we all learn from each other.
8:43
'cause you know, we're still at this very beginning of ai.
8:49
Nobody has answers.
8:51
Some people are gonna come out and say that I have the answer.
8:55
I don't have answers.
8:56
I don't have all the answers.
8:58
I'm learning every single day, and I learn from young people.
9:04
I learned from other leaders in the industry.
9:07
I learned from AI engineers, developers.
9:10
I learned from COOs who are not AI builders, but they're driving adoption.
9:17
They see patterns.
9:19
There's different people, different perspectives.
9:22
This is the moment for us to learn and embrace ai.
9:25
Wow.
9:26
And I wanna come back to when you first learned about AI working with
9:31
those engineers and them telling you.
9:34
What the different type of AI were.
9:36
So for example, how would you define an AI agent versus an AI
9:41
native leader versus Agentic ai?
9:43
Generative ai, there's so many AI at the end.
9:46
AI feels like catch up on the fries right now.
9:49
So please give us some definition before we keep going.
9:54
You know, you know, so, okay, here's how I, I would break it down.
9:58
An AI agent is essentially a digital coworker.
10:02
That can act on your behalf, whether it's solving service issues or sending
10:10
various emails or running financial reconciliations, and depending on how much
10:18
power you wanna give to this AI agent, the agent can operate without oversight.
10:26
Wow.
10:27
Now you have to be careful, right?
10:29
You need to have like agent managers that are monitoring the performance
10:33
of those AI agents, and usually it's a human who have the subject matter
10:38
experts and need to look at, okay, is this agent like performing based
10:44
on what I think it should perform?
10:46
And then over time I can give it more volume, more responsibilities
10:51
based on your customer needs.
10:53
Okay, now.
10:55
An AI native leader or an AI native worker is somebody who doesn't just
11:02
adopt ai, but also really think about, um, the workflows around it and like
11:10
reimagine at a large enterprise.
11:13
How should I be thinking about the new AI powered system?
11:18
Earlier I talked about, hey, you know the people who just bring the technology.
11:23
But really you gotta think about, hey, what are the various people,
11:29
process and technology changes?
11:32
To drive the AI adoption, and that's what I call like an AI native leader.
11:38
Um, it's really like think about at scale, how do you shift from just
11:43
managing tasks to actually orchestrating outcomes for the enterprise?
11:50
And then the last one, you know, it's actually a really good question.
11:53
You asked about like, ag gentech, ai, there's a lot of
11:57
point solutions in the market.
11:59
And, but I really think about a Gentech AI as a complete system.
12:05
Um, it's a system of various interconnected, many agents
12:12
in the different functional areas to learn, adapt, and
12:18
collaborate across all functions.
12:21
And if you think about it, it's really not just.
12:25
Automation.
12:26
It's kind of like a limitless workforce that Mark Benioff talked
12:30
about during our Dreamforce last year.
12:33
And you know, one of the powers of what Salesforce has been
12:38
able to do as our own gigantic enterprise is because we have sales,
12:47
marketing, commerce for transaction.
12:52
Uh, for service support, it really has various touch points across
12:59
the entire customer journey.
13:02
And when we have agents on top, the different functional
13:07
agents can have visibility and orchestrates with one another.
13:14
So from a customer standpoint, when I go on the help.salesforce.com.
13:20
I'm exp interfacing with an agent asking questions about support, but really,
13:27
like in the future, if I want to buy a new product from Salesforce, learn
13:33
about a new product from Salesforce, the agent can be on the front end, but
13:39
at the backend they can orchestrate and share information with the sales agent.
13:47
So the experience for me as a customer is it's a perfect handoff.
13:53
I don't need to know, I don't need you to be shipping the org
13:56
chart to me as a customer, I want those agents to orchestrate
14:01
the actions amongst each other.
14:03
Similar to how when we work, you know, if you're a customer wasa, like you
14:08
wanna scale, you know, your podcast, and you come to me and ask, Hey,
14:12
how do I use your marketing agent?
14:16
I'm going to introduce you, give the context of our relationship
14:22
of your business to the sales leader in our organization.
14:28
The agent should be able to do the same thing, and in fact, the agent will
14:32
probably be able to do a better job because it's all within the same system.
14:37
A Gen KI system powered by, in our case, it's powered by our Salesforce data cloud.
14:44
Which has a unified view of the customer across the entire customer
14:50
360, and we're taking new signals based on the interactions you have
14:55
with, uh, across ev every single touch point with our entire company.
15:02
That is the power of like the AG agentic AI system.
15:06
Wow, that's impressive.
15:08
But also coming back to something you said earlier, you said, when you think about,
15:14
um, this whole idea of egen enterprise energetic ai, you have AI managers, so
15:22
humans really overseeing the work of AI and they have to be subject matters.
15:28
But you also implement it on the other end.
15:31
You need like 70% leadership and 30% tools.
15:34
So what does it mean for the workforce itself?
15:37
For somebody maybe entering the workforce or somebody listening, how do, how
15:42
do they learn to manage AI agent?
15:46
Like what, what kind of upskilling is needed there and how can an existing
15:51
leader bring it to their organization?
15:54
So
15:54
essentially with this agent force transformation, gentech AI transformation,
15:59
we've created a new category of jobs.
16:03
Okay?
16:04
And that's what I call agent managers.
16:07
Agent supervisors.
16:09
And it's actually a component of a few different roles.
16:13
So first you talk about generative ai, right?
16:18
And there's this, this subject called prompt engineering.
16:23
And really, I mean, this is one of the most isolating terms, I don't think it
16:27
should be called prompt engineering.
16:29
It should be called prompt designing because all it is is using your natural
16:35
language to put in a prompt into open ai.
16:39
Anybody can do it.
16:41
There's nothing to be engineered.
16:44
You don't have to be an engineer to write a prompt, Uhhuh.
16:49
Okay?
16:49
So really it's a prompt design.
16:53
Okay?
16:53
And, um, and then.
16:55
So, so there's like people who specialize in prompt design, and we actually have
17:01
a team who designed those prompts based on the subject matter expertise because
17:07
of the various industries experience we've had over the past 26 years.
17:13
So whether it's, uh, financial, financial services, whether it's healthcare,
17:20
there are very specific topics.
17:23
That's needed, right?
17:24
In those industries.
17:25
And we have people who design the prompts, okay?
17:30
So that our customers can use it out of the box.
17:33
But if you think about as, um, an agent manager, you can design the prompts.
17:40
You can also set guardrails.
17:42
What are my agents able to do and should not be permitted to do based
17:47
on the roles of responsibilities similar to humans, right?
17:50
You're not gonna.
17:51
As a finance profess professional, you're not gonna let, um, a sales
17:56
person see specific financial information that's company sensitive.
18:02
So that's what we call guardrails.
18:04
You set guardrails, and again, that's where you need subject matter expert to
18:10
go and set those guardrails in the tool.
18:13
So the agent.
18:15
Aren't just wild agents.
18:17
And then once the agents are built, you also need to figure out, you
18:21
know, the agent is only good at as good as the data it has access to
18:27
and the, and that's relevant in the specific enterprise context it has.
18:34
Mm-hmm.
18:35
So then you gotta figure out, okay, how much data do I feed this agent?
18:40
Like I said, you don't need to give the agent all of the data.
18:43
Right now we're getting to the governance, security, privacy, access,
18:51
control into the data, and we build it into our platform where we can,
18:56
where our customers can make that decision, right for the agent manager.
19:02
Now, once you build the agent.
19:04
You need somebody to be monitoring the agent.
19:08
And I joke with people, unlike managing humans, you actually have have to be
19:13
watching the agents like very, very closely, especially as you ramp over time.
19:20
You don't just give those agents full responsibility on day one, have them
19:25
handle all of the custom traffic, and you have to be looking at the utterances.
19:31
That's coming from the customers, evaluate how they're responding
19:37
and are they meeting the customer satisfaction and also tweaking them.
19:44
So there's a very rapid iteration of looking at the, how the agent
19:50
is performing, tweaking, making the agents better over time.
19:55
And then as you.
19:58
Assess their performance, you can give it more traffic.
20:02
Right.
20:02
Because we started with zero traffic, then we ramped to 5%, 10% on the
20:09
help.salesforce.com agent, and you know, now we're handling, the agents are
20:16
handling like 85% of case resolutions.
20:19
Wow.
20:23
Okay.
20:23
Right, and, and it's a wha over time.
20:26
So that's like the job of an agent manager.
20:29
Understand the basics.
20:30
Learn the language of data, ai, and how to manage an agent.
20:35
Get your hands dirty and
20:38
your managing digital worker.
20:40
Thank you so much for sharing that.
20:42
And it also makes me think about how maybe companies are going at it too
20:47
fast, because there was a survey that came out saying how the percentage of
20:52
AI project that are failing, because again, I feel like we're using, uh,
20:56
ai like ke like catch up on prize.
20:59
When you think about.
21:00
Scaling, utilizing AI at such a large scale like Salesforce?
21:05
Like how do you bring such a big project to the finish line,
21:09
such a large organization?
21:11
How do you get buy-in?
21:12
Like what are some of the things that you did to bring it in your role?
21:17
Yeah.
21:18
You know, it's interesting you mention it because I think right now.
21:22
The market is still asking the questions about generative AI and thinking that the
21:29
large language model companies can solve all the problems at large enterprises,
21:35
and it cannot be further from the truth.
21:39
And the reason why I say that is because if you think about those large L Lands,
21:43
they're trained on public company data.
21:47
Right.
21:47
And the agents are only, and we're getting to the phase of agen ai, not just
21:53
generative ai, because I mean, generative ai, it's more consumerable, right?
21:58
Um, and, and you know, look, we also use those generative AI tools, the
22:03
LLMs, but I mentioned before, there's a few really critical things about data.
22:11
You need trusted, well governed.
22:16
Enterprise data.
22:18
You need to control the access.
22:20
There's toxicity and there's bias in those models that essentially we've built the
22:29
trust layer to do some of those work, and we built an extensible like platform.
22:38
For, for the Salesforce trusted agent force platform so that we can help
22:45
the customers solve these problems.
22:49
A lot of the customers we've seen is like, they, you know, they spent a
22:53
few million dollars, sometimes tens of millions of dollars with a, a large
23:00
language model, and then they're like, oh.
23:02
So we're like, okay, you're going down a DIY route.
23:06
You gotta build your own security, your trust layer, bias detection.
23:12
How do you build the guardrails, right?
23:14
And we've built all of this inside the Salesforce Unified Agent force
23:22
platform so that if you don't have the thousands of engineers
23:28
to do this, we've done it for you.
23:32
And what we advise, I advise my co uh, customers to do is look at the end of
23:39
the day, it's about you creating value for your business, for your customers.
23:47
Figure out where your core competency is and use us as a trusted advisor
23:55
with the platform and you focus on the majority of the time.
24:00
Designing those agents specific to your enterprise context
24:06
and make your customers happy.
24:09
Bring more customers, expand your business, and run it effectively.
24:16
That's where a lot of the enterprises should be spending their time
24:19
on, not just to build technology for the sake of it, right, but.
24:27
Focus on the higher value creation chain.
24:31
I mean, you're, we're talking about finance people, like focus
24:35
on, okay, what is, what's going to help me with my top line growth?
24:41
Bottom line efficiency.
24:44
I would direct all the business users and technology thinkers
24:48
and builders towards that.
24:52
Don't just pick up any shining project and follow the herd.
24:56
Rather think about, okay, what is the business outcome?
25:00
What are the biggest business opportunities we should be
25:03
going after that our competitors are not going after right now?
25:08
And now think about how do I use AI agents to get to that outcome?
25:16
Wow.
25:17
What is the shortest time to value?
25:19
Right?
25:19
Like shortest time to value because you want your ROI not wait for five years.
25:24
Want your RI like in the next three months, pick up a use case and build it.
25:32
Experiment it, prove value and then over time you'll give them more money.
25:38
Right?
25:38
And resources.
25:41
But, but like, prove it first, you know, this is the finance like discipline.
25:47
I give you a little bit of investment.
25:49
You show me the money, show me the return.
25:51
Then over time, I'm gonna give you a little bit more,
25:54
and you continue to do this.
25:56
I'm gonna calculate my ROI.
25:57
Yes.
25:58
In your investment.
25:59
Yes, yes, yes.
26:01
But I love why you said it, because it's really a roadmap for people.
26:05
You know, with, with any function, not just finance, like make
26:08
sure you have the right data.
26:09
One that you're on the right platform and yours is going for the sake of going.
26:14
But like you said, you really take a step back to think about your core competency.
26:18
Where would you have the most value in the shortest amount of time
26:22
to get that ROI and keep moving.
26:24
That's brilliant.
26:25
Thank you so much.
26:26
Now I'm curious now for the finance people, so what could be the benefits of
26:30
agentic AI for CFOs and finance leaders?
26:35
I think about, for example, procurement.
26:38
Okay.
26:39
My gosh, right?
26:41
I mean, I spend time in data center operations.
26:45
The amount of cross collaboration between engineering than to
26:54
procure the equipment with.
26:57
The partners, whether it's a hardware provider or interconnection data
27:03
center provider for power and space, uh, or to the delivery, right?
27:10
Once you place the order, there's a delivery process where
27:15
could really need visibility
27:19
and then to build.
27:22
Like I think about that entire chain and from a finance standpoint, what's really
27:27
important, you wanna see the throughput and you wanna see the time from the moment
27:34
you place the order to the procurement.
27:38
Like when you're gonna pay, once the data centers receive the assets right
27:44
through the COA and then you forecast.
27:50
The cash flows and the impact on the income statement.
27:54
I think a lot about there's the agent who can orchestrate across
28:00
the entire supply chain, right?
28:03
Not only from the moment of procurement, but like the common kind of having
28:08
common visibility across the different departments, because sometimes you
28:12
may not be getting the signals.
28:14
Yes.
28:15
And that creates error in forecasting.
28:19
It actually creates a big amount of risk.
28:21
Yes.
28:22
Think about during COVID time when supply chain was down earned
28:27
and like everything was delayed.
28:30
What?
28:30
Then you scramble from a finance professional, like you
28:34
scramble to figure out, okay.
28:36
How do I manage the books now?
28:38
Because I've told my investors to expect X and then there's all these
28:43
surprise supply chain issues, right?
28:45
Yeah.
28:45
But I think about, hey, how do I use agents?
28:48
Effectively, we could use agents to create the interconnection, doing the um,
28:54
coordination across the different kind of functions as well as the communication.
29:00
Nice
29:01
agents can send emails.
29:03
To the different departments based on the visibility, the type of information
29:07
you need to see for the needs you have.
29:10
So in the case of finance, like you could use the agent to tell you, Hey,
29:14
you know, there's gonna be a shock in the system and I'm gonna tell you
29:17
before like it happens in my financial.
29:21
Exactly right.
29:21
Well, before maybe other partners who may not be seeing your perspective,
29:28
what's important to you, but because you wire the agent this way.
29:33
The agent can predict and tell you.
29:35
Systematically,
29:36
what's your favorite thing to do outside of work with all these AI going on?
29:39
I'm curious.
29:41
You know, I have two young daughters.
29:42
They're seven and I, and um, I really started to think about,
29:46
hey, how do I create that must do common connection and so.
29:51
I actually go to my parents', um, retirement community every
29:55
weekend now, and we're learning how to play golf together.
30:00
Nice.
30:01
And so we learn, we try, we don't hit the ball.
30:05
Okay.
30:06
Like, you know, we get another try.
30:08
A lot of this, I was like, this is the agen AI mindset.
30:13
Right.
30:13
Just experiments with something and learn and you improve over time.
30:19
Because, you know, to get to perfect, it's 10,000 hours.
30:22
So I started, I really appreciate doing that and what it also helps
30:27
me do is like I connect with.
30:29
Not only my daughters, but also my parents were there and it's so
30:33
important we spend the quality time.
30:36
Yeah.
30:37
And invest in those
30:38
relationships.
30:39
Awesome.
30:40
Keeping that you in touch.
30:41
Well, thank you so much for being on the show, Vivian.
30:43
It was super, super helpful to have you on.
30:47
Thank
30:47
you so
30:47
much.
30:48
It's been a pleasure.
30:49
Awesome.
30:50
And that's it for today's episode of The Diary of A CFO.
30:54
Thank you so much for tuning in.
30:56
If you enjoy the show, don't forget to like, review,
30:59
subscribe, and share with others.
31:01
Our website is the diary of a cfo.com.
31:04
That's where you can find all the episodes, access all the
31:08
guest profiles, see their vials and their social media links.
31:12
It is also the place where you can apply to be a guest on the
31:15
podcast and have information about how you can sponsor the show.
31:19
As always, if there is any topic you would like me to cover in the future, just
31:24
email me at ask at the diary of a cfo.com.
31:28
See you soon
31:29
wondering if you're ready for the CFO role?
31:32
Find out with a free CFO readiness score at thecfo.scoreapp.com.