In this episode of Digitally Curious, host Andrew Grill, renowned futurist and author, sits down with Shannon Scott, Senior Vice President and Global Head of Product at Airwallex, one of the world’s fastest-growing FinTech innovators.
Key Topics Covered:
- Shannon’s Journey:
From rural Victoria to leading global product strategy at Airwallex, Shannon shares how his background in computer science and mechatronic engineering shapes his approach to building next-generation financial products. - Engineering Mindset in Product Leadership:
Discover how thinking from first principles and understanding technology “under the hood” enables Airwallex to deliver seamless, global financial services and challenge industry assumptions. - AI’s Transformative Role in Financial Services:
Explore how AI is not just automating traditional tasks like fraud detection and compliance, but fundamentally transforming business workflows, onboarding, and financial operations — turning hours of manual work into minutes. - Agentic AI Explained:
Shannon demystifies agentic AI, describing how autonomous AI agents can handle complex, multi-step financial processes, from vendor onboarding to payment reconciliation, and what this means for both large and small businesses. - Trust, Explainability & Regulation:
The episode delves into the importance of maintaining trust and explainability in AI-driven finance, the role of human feedback, and why robust regulation gives financial services a head start in adopting AI responsibly. - Data as a Strategic Asset:
Learn why proprietary, high-quality data is the new competitive edge in the AI era, and how modular, adaptable data infrastructure is critical for future-proofing financial services. - The Future of Decision-Making:
Andrew and Shannon discuss the evolution of AI from an operational tool to a strategic decision partner, capable of suggesting best practices, optimising approval flows, and proactively managing risk. - Actionable Insights:
Shannon shares three practical steps for listeners to better understand and leverage agentic AI in finance:- Embrace podcasts and diverse learning sources
- Experiment with new AI tools and services
- Continuously question and seek better ways of working
Resources
Airwallex Website
Shannon on LinkedIn
Thanks for listening to Digitally Curious. You can buy the book that showcases these episodes at curious.click/order
Your Host is Actionable Futurist® Andrew Grill
For more on Andrew - what he speaks about and recent talks, please visit ActionableFuturist.com
Andrew's Social Channels
Andrew on LinkedIn
@AndrewGrill on Twitter
@Andrew.Grill on Instagram
Keynote speeches here
Order Digitally Curious
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
got a fast connection , but at the end I can see how
much percentage is uploaded.
It basically uploads a perfectcopy in the background as we're
speaking.
Speaker 2 (00:08):
Sounds good.
Speaker 1 (00:09):
Okay, cool, I'm just going to bring up my script and
we'll go from there Three, two,one.
Today in the podcast.
I'm delighted to welcomeShannon Scott, senior Vice
President and Global Head ofProduct at Airwallex, one of the
world's fastest growing fintechinnovators.
Shannon's journey is afascinating one.
With a background in computerscience and mechatronic
engineering, shannon brings aunique technical and strategic
(00:32):
perspective to the world ofglobal finance.
At Airwallex, he's at theforefront of developing cutting
edge AI and data drivensolutions to transform how
businesses manage internationalpayments, compliance and
financial operations.
Today, we'll explore Shannon'scareer path, airwallex's vision
and dive deep into thetransformative role of a genting
(00:52):
AI in financial services.
Welcome, shannon.
Speaker 2 (00:56):
Thanks, andrew, it's great to be here.
Speaker 1 (00:58):
Nice to have a fellow Aussie on the podcast yet again
.
And we share similar traitsbecause we both come from
engineering backgrounds and I'dargue that engineers think in a
slightly different way becausewe were taught to think of
things from first principles.
That might come out in adiscussion, but perhaps you
could walk us through yourjourney from growing up in rural
Victoria to becoming SVP,global Head of Product at
(01:18):
Airwallex.
Speaker 2 (01:20):
Yeah, absolutely Like .
I think it's been a wonderfuljourney.
Actually, I did grow up inrural Victoria but had the
opportunity to go to theUniversity of Melbourne.
I knew I was good atengineering.
I was very interested incomputer science.
This is around the time of, Iguess, the dot-com bubble
bursting.
A lot of hype, a lot of tearsperhaps for those already sort
(01:43):
of in industry, but I sort ofsaw it as something that was
going to be very exciting intothe future.
One thing that was very luckyfor me as I went through my
education is the graduateopportunity that I had following
my university was actually avery small business and it was a
combination of buildingsoftware for the insurance
(02:03):
industry and then also makingsure we were working very
closely with our insurancecustomers to ensure they got the
most out of that software.
And so it was a greatcombination of software
development, understandingwhat's possible to build,
understanding the models and howthey work under the hood, but
then also really understandingthe customer experience and what
it means to build software thatthey can actually understand
(02:24):
and use.
And I think over my careerjourney there's always been a
software element, but it's oftenalso been on the commercial
side, on the product side,really thinking about how you
can bring technology toconsumers, to businesses, to
users and make it as powerful aspossible.
Speaker 1 (02:41):
I had a chuckle there when you were talking about
going to university during thatdot-com era.
I was at Telstra during theTelstra dot-com era when they
were trying to become a Yahooand of course they realised they
were good at being a telco.
So interesting times around theturn of the millennium.
We talked about the fact thatwe're both engineers or come
from an engineering background.
How does your engineeringbackground influence your
(03:02):
product leadership at Airwallex?
Speaker 2 (03:10):
I think your comment about first principles is
actually a really important one.
I think the better that youunderstand what's happening
under the hood and can reallyreason with how this technology
or service is operating, you canultimately create a much better
product or much better productexperience.
So you know to give an example,I think probably our
wallet-extended financialservices.
We provide both acquiring andissuing, so both the card
provider and, where you know,your audience, would enter those
(03:33):
card details online to acquirethe payment.
People don't often think aboutwhat's happening under the hood.
There.
They're just like okay, mymoney's been deducted from my
bank account and it appears inthe business that I'm purchasing
from, but actually it's just acomputer network.
It has all the same sort ofquirks that you might find if
something doesn't necessarilyalways go right when you use
your own computer.
It has some strange behaviorsthat are there, maybe because
(03:55):
that was a legacy of cards beinga service that used to be not
digital, and so there's somethings that aren't perfect, but
some things you take for grantedhow they work.
If you really understand how itworks under the hood, you can
eliminate those things you'retaking for granted or those
assumptions and actually thinkabout how can I make this better
overall?
If you don't have that depth ofunderstanding, you're probably
(04:17):
not going to question thoseassumptions or understand what
can and can't be changed.
Speaker 1 (04:22):
Now you're inarguably one of the world's fastest
growing fintech companies, somaybe give us an overview of
Airwallex's positioning in theglobal fintech market and your
role scaling a company valued at$6.2 billion?
Speaker 2 (04:34):
Yeah, so this is actually our 10th year.
I think we're coming up for 10years towards the end of the
year.
We've reached 1,800 employees.
We're across 25 differentoffices.
It was actually founded inMelbourne, but we've very much
sort of grown out all over theworld with a strong presence
here in London, where I am today, and then a number of other.
I think the sun never sets onour Airwallex offices.
(04:56):
The product itself is helpingbusinesses who need to operate
globally like really easilyinteract with their financial
services, and so if you're abusiness in the UK, you have a
customer in Singapore, forexample, you want to receive
pounds, but your customeractually wants to use a local
payment method that is uniqueand understood in Singapore and
they want to pay in Singaporedollars.
(05:17):
We make that an incredibly easyexperience.
It doesn't matter that you'reoperating in two different
countries.
Actually, both parties aregetting the experience that
they're used to, and this meansa business doesn't need to worry
about setting up differententities all around the world.
They can just get started withAirwallex and they can pay
suppliers internationally, theycan hold different currencies
and they can sell to customersall around the world, bringing
(05:40):
together all of those differentfinancial networks having the
licenses to operate in all ofthose regions takes a huge
amount of energy and effort, butthe team has been working
really diligently to just keepbuilding out that network and
making it as easy as possiblefor our customers to get access
to those global services and Ithink that's been a testament in
how the product's been adoptedour customer growth over the
(06:01):
last 10 years and now around$800 million in annual recurring
revenue.
Speaker 1 (06:07):
So we're recording this middle 2025, and I've spent
the first half of this yeartalking about nothing but AI.
So I feel like, almostcontractually, we have to segue
into the world of AI.
So how is AI fundamentallyreshaping financial services?
Speaker 2 (06:22):
Yeah, I think there's probably one other point I
should make about AirWallexitself, and I think one thing
that's interesting is thatfinancial services often are in
aid of some other intent.
Right, for example, a customerwants to purchase something from
your website, or you want topay a supplier who's providing
services to you it could bemaybe cloud services.
It could be the materials thatyou then on-sell, be maybe cloud
(06:44):
services, it could be thematerials that you then on-sell,
and so the payment is just apart of that intent to purchase
from the supplier.
Whatever that might be.
Technology and fintechs aremaking it really easy to not
just support the financialtransaction, but also all of the
workflows that businesses needor they potentially might do in
a manual way.
We support all of thoseworkflows, from onboarding the
(07:07):
vendor all the way to sort ofmaking the payment and
reconciling the tax aspects, andso when we think about
technology transformingfinancial services, I think the
first step is it's a lot moreabout business process and
aiding that process than it isjust the transaction.
So when we think about AI,there's a lot of different ways
AI can actually slot into boththe financial service and then
(07:28):
the business workflow.
On the financial service side.
There's things that we've beendoing for many years.
So risk management,understanding your customer,
fraud detection, fraudprevention, regulatory
compliance all of those aspectsI think have been using a
variety of sort of modelingtools and then more modern AI
solutions to support thosetraditional finance aspects.
But one thing I find reallycool is when I think about
(07:51):
building a product that's agreat product for our customers.
How is that AI actually visibleand supporting those customers?
So often that workflow foronboarding a vendor all the way
through to payment is performedby a financial operations team.
It requires a lot of steps.
It requires a lot of steps.
It requires a lot of approvalflows, and AI can actually
fundamentally automate a lot ofthe things that used to be
(08:11):
manual and what used to takehours or perhaps a lot of man
hours to process many vendorsactually comes down to just
minutes.
So I think that's one of thereally exciting applications of
AI that customers directlybenefit from.
It's not just happening underthe hood.
One other thing I'll say aboutAI that I think is really
exciting is it's changing theway that humans interact with
(08:32):
computers.
So your operating system isyour iPhone and the apps on the
iPhone it's your desktop, it'syour browser, but it's
increasingly becoming your chatinterface and it's a very
natural way for humans tointeract with other humans.
You know we're talking throughchat now or using text on our
phones More and more.
I think you're going to usethat as a channel to interact
with the services that you wantto.
(08:53):
So we're going to start to seemore and more services,
including financial services,being accessed through these
more natural language-basedchannels, and I'm really excited
to see where that goes over thecoming years.
Speaker 1 (09:06):
And what I find interesting is that you know AI
isn't new.
It's been around for 75 years.
Financial services have beenusing AI under the hood for
years now and I'm sure you'vebeen using it since day one.
So, to cut through the hype,you know what transformative
impact you think AI will Read itagain what transformative
(09:26):
impact will AI have in financeover, say, let's read that third
time what transformative impactwill AI have in finance over
the next, say, five years?
Speaker 2 (09:35):
I think it's going to be a much higher trust from
customers.
Even though financial servicesrequire pinpoint accuracy and
certainty in what transactionI'm making or how the money's
been received and thereconciliation, I think there's
still going to be a hugedelegation to AI to automate a
lot of those different services.
And so when I think about theway a lot of financial services
(09:58):
are offered today, it's like,hey, this is a business product,
I'm allowed many users on theplatform.
Those users might havedifferent functions, like one
might submit a request for apayment, another might set up
the payment and a third partymay even approve the payment to
make sure it's legitimate andgoing to the right person.
So I've got different userswith different roles happening
here and we provide a servicethat allows you to configure
(10:20):
that in such a way.
To say, if the price is, or thetransfer amount is, greater
than $500, then it must besigned off by two people, not
just one.
We have all the tooling thatcan do that, but we put the onus
on the user to actually say youcan set this up however you
like.
Ai is going to come in and notjust automate a lot of those
steps, but it's actually goingto start to make suggestions and
(10:40):
operate on your behalf.
To say this transaction looksvery standard or legitimate.
This is something that's verycommon by the requesting user.
I've already been able toverify the account details for
you and they match the actualuser's bank account information,
and we have to have checked thename against the third party
database.
Therefore, I think this is avery standard transaction.
I'm going to recommend to autoapprove this, for example, and
(11:02):
just at some, at some stage ordepending on the amount, it may
simply actually auto-approve andoff it goes.
At other times, we may see thata user wants to just eyeball it
and say, yes, that's great, butwe effectively took what was a
series of steps by many usersand now it is just, in fact, one
step or one click to verifythat we have the right
information, and so it'sactually those financial
(11:23):
operations that are going to beautomated, collapse down and
give businesses a lot of timeback to focus on their core
business.
Speaker 1 (11:31):
I was talking to my friends at SAP Concur they do a
lot of expense management andexpense reporting and they were
saying in the future we mighthave, rather than expense
reports, we might have exceptionreports.
And you kind of detailed thatthere so that if it looks normal
and if AI can see the trendthat it looks like a normal
transaction, only the exceptionwill go for human approval,
which is going to save a lot oftime than looking at every
(11:51):
single transaction.
Speaker 2 (11:53):
Yeah, absolutely.
And even things like uploadingthe receipts or sort of
understanding, like taking aphoto of the bill that needs to
be paid, for example, like thatcould already be automatically
uploaded if we can sort of lookat the profile of the image or
the document and recognise thatit's actually part of this
(12:13):
business transaction and it goesstraight in.
So you don't even have to bedoing the steps that upload the
documents.
Speaker 1 (12:20):
So the advantage for a company like yours is you're
fairly young, so you've got atech stack that's fairly new.
Speaker 2 (12:38):
When we look at traditional banks that have got
legacy systems.
How should traditional banksadapt to this AI-driven FinTech
competitors like yourselfchallenge they're going to find
is not only are we relativelyyoung, but I would say we are
tech first businesses.
We are not financialinstitutions, we are technology
institutions, and I think,therefore, we understand those
sort of emerging technologiesmuch more effectively and are
sort of bringing that into thefinancial service industry,
Obviously not only being afinancial institution, but being
(12:58):
an incumbent and largeinstitution for any business.
I think when you're large, itis hard to actually, you know,
sort of turn the ship towardsthese new technologies.
One thing that gives me a lot ofconfidence, though, is that a
lot of the services that AI isenabling and, if we take
customer support as an example,there's a plethora of great tech
(13:20):
companies vying for yourbusiness to provide really
wonderful customer supportsolutions.
Right, and this includes themodels and the languages that it
uses to understand the customerand how to respond to them.
It includes the case managementand the triaging and the
prompting.
It includes the embedding ofthe chatbot into your website or
the sort of call center oroperations team that you have,
(13:40):
and so traditional banks ortraditional businesses they're
not going to need to reinventthe wheel.
I think a lot they're not goingto need to reinvent the wheel.
I think a lot of thosetechnologies are going to come
to them in packaged ways, justlike they're using technology
today to be able to access theseservices.
So in that regard, I thinkthey'll be able to be supported
by a lot of other great techcompanies to sort of carry them
(14:01):
into the next era of finance andcomputing.
Speaker 1 (14:06):
So what we've seen with the adoption of generative
AI and also agentic AI.
So what we've seen with theadoption of generative AI and
what we'll talk about a bitlater, which is agentic AI
regulators are really scramblingto keep up because the
technology is just moving soquickly.
So what role does regulatorycompliance play in AI adoption?
Speaker 2 (14:27):
quickly.
So what role does regulatorycompliance play in AI adoption?
Look, I think it's mostcertainly extremely important.
I actually feel very confidentin our space in financial
services that it's actuallyalready very well regulated.
There are very clear guardrailson how information should be
utilised and stored, what typesof customers that you can
support, what types of sort offraud systems that you have in
place, and because all of thisis quite well understood, I
(14:49):
think you're creating a greatframework and sort of set of
guidance for how the agents, themodels that you bring into the
business, should be operating,and they can be quite small
agents with very specificfunctions, so it's well
understood what that function iswithin the sort of guardrails
of regulatory compliance.
I think there's a lot moreinteresting questions that
(15:10):
actually sit probably outside offinancial services.
That may include access topublic information, that may
include, like IP and those typesof issues where there can be
some really interesting questionmarks for AI.
But in financial services, I'mvery confident that we already
have the guardrails and a strongunderstanding of the current
guardrails from which we can useAI to build on top of.
Speaker 1 (15:34):
Now, the access to quality data has always been
something that companies haveneeded to worry about, and when
we move into this AI era, thequality of data becomes more
important, and I've been talkingabout the need for good quality
data for some time now.
But how should companiesleverage data as a strategic
asset in the AI era?
Speaker 2 (15:52):
It's a great question , and I think for data that is
generally, you know, readilyavailable or that is perhaps
even generated by the AI.
If we're looking at, you know,media use cases, for example,
you know businesses are going tohave access to that information
, but so is every other business, and so is every other sort of
AI product that's available.
And so what is the data thatyou have?
That is that you understandwell, that is structured in a
(16:15):
way that supports your businessuse case.
That is perhaps proprietarydata or sensitive data,
certainly because your customershave trusted you with that
information.
Like for most financialservices institutions, like you,
have data that isn'tnecessarily available outside of
that public realm, and so thatgives you a great opportunity to
, I think, build AI tools thatprovide a unique or value-added
(16:39):
service that other partiesaren't necessarily going to
provide.
So when I think about buildingnew AI features or how Airwallex
can really provide a greatcustomer experience, it's like
what am I actually offering thatthe customer can't get through
some of the means, or that isactually going to genuinely move
the needle for them in a waythat's built into Airwallex
(17:01):
rather than just something thatthey can do without necessarily
the services of Airwallex.
There are, of course, a lot ofdifferent types of data sources.
So you know, the financial datais very well structured and
requires very high accuracy andvery strong reconciliation.
And then there are data sourceswhere you're just trying to
understand hey, what does mycustomer look like?
(17:21):
What are the services that Ishould provide to them?
Is this a legitimate customer?
Do I understand their sort ofbusiness cases?
And in that case you're sort ofcreating a synopsis of that
customer to give them a greatproduct or service.
You know, once they'reonboarded, often AI can create a
better picture of who thatcustomer is and what they need
than a human can.
Speaker 1 (17:42):
So let's talk from about future.
Proofing.
How would you build a datainfrastructure that support both
current and future AI needswhen we actually don't know what
the next AI state might looklike in three or four years?
Speaker 2 (17:56):
It's a really good question and I think that it
would be um naive to have toostrong an opinion here, because
because you never know sort ofhow quickly things things move
um and and sort of what servicescome down the road.
I think we are seeing aconsolidation of different
services, and you mentionedagentic ai, and while ai has
been around for a long time likethat term is relatively new um,
(18:18):
like mcp service, for example,and these are sort of the units
or APIs that different AI agentsinteract with to perform a
given service is also arelatively new concept, and you
see the market solidifyingaround these things and then you
can start to build on top ofthem and create really great
solutions.
I think any business should bethinking about experimenting and
(18:40):
keeping up to date with thelatest sort of AI technologies,
but then also really thinkingabout where am I investing?
Where am I using sort ofbest-in-breed products that I
can easily absorb into mybusiness or build tooling around
?
If that best-in-breed productactually a new version comes out
in a year from now or itactually becomes even better in
(19:01):
two years, can I swap that outquickly and bring in a new
solution, to give you an example?
So, on that expense managementside, we used to use a
third-party provider to use OCRand entity extraction to
understand the contents of thereceipt.
What were the items, how muchwas it, what was the total?
What was the tax, what was thedate?
Um, it worked okay and itprobably cost us about 20 cents
(19:22):
per receipt to actually uploadit.
Um, in the last 12 months weswapped that out with google
gemini.
Um, the cost went down by aboutfour orders of magnitude, so I
think it was like 0.002 cents orsomething similar, and the
accuracy was dramatically,dramatically better.
The product was the same, youknow, it was exactly what the
users wanted.
In fact, it was better becauseit was higher accuracy.
(19:43):
But with the same intent, butbecause we'd built quite a good
modular solution, we were ableto swap in the latest technology
and get a much better outcomewith very little effort.
Speaker 1 (19:55):
Just to jargon bust MCP.
What does that stand for?
Speaker 2 (19:59):
Model Context Protocol, I think, and I suspect
that didn't really help youwith any further information.
Is that fair?
Speaker 1 (20:08):
I'll look it up.
I'll ask my AI friend and I'lllook it up and I'll drop in the
right term post.
Speaker 2 (20:15):
I actually feel a bit it's a relatively new
technology that I feel a littlebit sheepish explaining to your
audience and so I hope I don'tget it wrong but AI agents, for
example, those using chat theymight have to call on an
external service.
So, for example, if you typeinto the chat hey, I'd like to
make a transfer to Andrew for$100 AUD, please, please, make
(20:38):
that transfer for me.
In the background, the chat hasinterpreted that request and
made a request to a differentservice to initiate the transfer
or for that service to say okay, well, in order to do that
transfer, I need you to enteryour password, for example, and
(20:58):
so that tooling that understandshow to take that instruction
and provide a response iseffectively that MCP server, and
these are different units thatoffer that different functional
pieces of work.
Speaker 1 (21:13):
Now, one thing I've been saying to audiences for a
while that seems to resonate israther than thinking of AI being
artificial intelligence, theyshould think of it as augmented
intelligence, because there'salways got to be a human in the
loop.
So what role does humanfeedback play in AI systems, for
example in fraud detection?
Speaker 2 (21:31):
Yeah, fraud's a great example.
I would argue that we don'talways need a human in the loop,
but I think it can actuallyalways create a much better
solution.
So if we look at fraud, it isactually a heuristic.
A transaction happens inreal-time.
Maybe you've made a purchase onyour credit card, maybe
somebody else is making apurchase on your stolen credit
card and we have to make anassessment at the time.
(21:53):
Do we approve the transaction?
Do we decline it?
We're looking at things likewhat was the amount and the
nature of the purchase.
Is this sort of typical of thatparticular individual?
Do I have extra information,like the IP address of where the
purchase was made, so I know ifit's happening in a country
that is different to where youare?
We're using a number of thesedifferent factors to understand
(22:15):
what's happening and we'remaking a heuristic decision that
may or may not be correct.
Now there's a couple of reasonswhy you find out if you're
correct, and feedback, whetherit's automated or from a human,
is going to continue to refineand improve the model.
So if you let it through and itturns out to be fraudulent,
you're definitely going to findout about it, because the
customer is going to let youknow, and that would be sort of
(22:36):
an unfortunate outcome.
However, what we've recentlyintroduced at Airwallex in our
product now is sometimes wemight decline a transaction that
we believe is fraudulent and itmay not be, and the customer
would be very frustrated ifsomething's been declined when
they were making the transaction.
So in real time we send them atext message that actually just
asked hey, this was the transfer, was it you?
(22:58):
And if they say no, it wasn't,it's like cool, we blocked it.
Or if we hadn't blocked it, wesay actually, in that case we're
going to cancel your cardbecause somebody else tried to
do it.
And if it was you, we saythanks very much, please try the
transfer again.
We'll make sure we let itthrough the second time.
And so not only is it a goodcustomer experience, because
they can sort of correct for aninaccuracy in the model, but
(23:20):
it's giving us great feedback asto the quality of the model and
we can use that to keepimproving our fraud
understanding.
Speaker 1 (23:27):
So the other side of the coin is if we're going to
trust the AI a little bit more,how do you ensure AI remains
explainable and trustworthy forfinancial decisions?
Speaker 2 (23:39):
Explainable and trustworthy.
I think we should cover thoseconcepts separately.
I think actually explainabilityis a really interesting science
because if we look at the fraudmodel, for example, we may be
able to infer oh yeah, okay, theIP address was in a different
country.
The AI can sort of inform usthe data that was used to make
(23:59):
the decision and we might beable to sort of intuit.
That's the understanding.
But we're probably never goingto know exactly all the
different heuristics that themodel used to determine that
fraud, especially if it'sgetting this feedback from
customers over time.
And actually that's probablyokay.
That's probably okay as long aswe also have a metric for what
is the accuracy of our fraudsystem.
(24:20):
Can we get a sort of generalunderstanding that this
heuristic is operating with 95%accuracy?
You're catching the majority ofthe fraud, for example?
So I think the explainabilitypiece depends on the context.
The second item you'll have toremind me, andrew.
Speaker 1 (24:39):
Trustworthy.
How do we ensure that AIremains trustworthy for
financial decisions?
Speaker 2 (24:43):
So if we thank you, if we look at a financial
decision, so, for example, Isaid, hey, make this transfer
for $100 to Andrew, and Isuggested we might just do that
in a chat bot, for example.
Now that's a very importantdecision.
It must have perfect accuracyand it must know exactly which
Andrew I'm talking about and itwould be really disappointing if
(25:06):
it went to a different Andrewor the wrong currency or
something similar.
And I think, where that trustneeds to be proven and based on
the severity or importance ofthe request, it's relatively
easy to prompt the user to saythis is my understanding of what
you've asked of me.
Could you please verify thatthis is the case?
And because we do have thattwo-way communication with the
(25:28):
user, it's relatively easy tomake that request and establish
not only that trust but theverification.
I think again, the smallerthings.
Over time the trust willincrease and we're going to let
the agent do that for us.
We're not even going to worryabout it, we'll worry about
exemptions only, like yourexample earlier.
The bigger things we're alwaysgoing to include, we're always
(25:50):
going to include a verificationstep and if I think about the
example around structuringreceipts, it's actually probably
going to take a while before weget to 100% accuracy.
It's going to take a few yearsperhaps, but 90% accuracy with
human verification is stilldramatically faster than it was
for a human to do this on theirown, and so we're still getting
(26:10):
like massive benefit, even ifthe system is not necessarily
100% accuracy.
I think users can easilyunderstand that that's a huge
benefit and they're going tohave that trust that the model
is still giving them a lot ofbenefit.
Speaker 1 (26:25):
So we've got ahead around generative AI and most of
my friends are now talkingabout chat, gpt, as they used to
when they were going to Googlesomething, which is interesting.
But now we've got this newconcept agentic AI or AI agents.
First of all, how do you defineagentic AI, because everyone
has a different explanation, Ifound yeah, so.
Speaker 2 (26:46):
So when I think of a, I think actually it's a really
great, great question andeveryone probably should have a
different opinion and and alsothink about the benefits of of
those agents.
But, um, an agent is performinga work function that uh can
benefit, benefit the user, andthere might be a process that
actually has many differentfunctions and we can string
together those agents to performa more complex task with many
(27:11):
smaller tasks.
So, if I take the receiptstructuring, for example,
there's going to be an agentwhose role is to actually
understand and decompose thecontents of the receipt, and
then we're probably going tohave an agent that actually
understands, uh, the math behindwhat's happened, going to
happen next.
So I have a.
This agent has a understandingof what is an account number,
what is a currency, what is asort of financial amount, and
(27:33):
then I'm going to package thatup into an actual transaction
object that I can take to thenext step, and then another
entity is actually going toperform the transaction itself,
and so if we have these agentsthat are performing discrete
units of work, we can both haveAI that is composable and
understandable and doesn'tnecessarily become this sort of
behemoth system, but we can alsoachieve very complex tasks by
(27:56):
bringing together many of theseagents.
Speaker 1 (27:59):
So how will a Gentig AI redefine financial workflows?
Speaker 2 (28:05):
I think that example.
Let's break it down.
So I talked earlier about howfinance is not just making the
financial transaction, and so ifI look at all the businesses
today, it's like, okay, I needto decide which vendor I'm going
to use to you know, be my newsort of it chat system.
I'm going to verify that thevendor is able to perform the
(28:29):
features I need.
I am going to verify that thevendor has the right
accreditation.
I'm going to verify that wehave the right information for
that vendor to make to make a.
I'm going to go through aprocess of approving that that
vendor is actually the vendorthat we chose.
This is the right amount ofmoney.
When they send me an invoice atthe end of the month.
That is the right amount.
(28:51):
I'm going to verify the accountdetails.
I'm going to schedule a paymentfor the end of the month.
That vendor happens to be in adifferent country, so I'm going
to have to make an FX conversionfrom pounds to US dollars, for
example.
I'm not sure about currencyfluctuations and I don't owe
them until the end of the month,so I'm going to lock in a rate
from pounds to dollars that istriggered at the end of the
(29:11):
month.
So I'm going to createcertainty in the rate.
That doesn't happen until theend of the month.
I'm going to schedule atransfer.
It's going to happen.
Then, once that transfer hasactually taken place, I'm going
to reconcile that with myaccounting system, and so all of
those steps require a reallysignificant financial back
office and a number of differentplayers to perform that.
Today we have a system thatallows you to do it all in one
(29:32):
place, which is very valuable.
I can see the lineage all theway from the payment through to
the vendor and who it was to andhow it was signed off.
That's very valuable.
But utilizing AI and thatenergetic finance, we're
actually going to see all ofthose things collapsed into a
highly automated process thatrequires much less manual labor
is actually going to increase alot more trust.
We are actually going to seethe accuracy come up over time
(29:56):
and I think it's going tomassively redefine not only the
amount of effort it takes forlarge businesses, but it's going
to give small businesseseffectively the muscle to be
able to do this themselves orget that benefit when they
otherwise wouldn't have had theresources to be able to perform
this kind of financial duediligence on their vendors.
Speaker 1 (30:17):
So I think you're right.
We've just really only startedto hear about agenda AI.
Speaker 2 (30:20):
We need to scratch the surface, so maybe you could
talk about some of the use caseswhere autonomous AI agents
excite you the most also aboutthe value of understanding your
(30:43):
customer, and so there are verystrict guidelines in the
financial services industryabout what types of customers
you can support and what typesof solutions you can actually
offer them.
Perhaps there is risk for thosecustomers as well, like, for
example, they might be targetsfor other fraudsters to be bad
actors against their business.
So we want to make sure thatthe financial services they have
can be utilized responsibly andthat they actually are not
(31:07):
taking on too much risk, and sounderstanding who that customer
is is a really important part ofproviding them with the right
financial services.
I think what we're seeing andwhat we've already implemented
is what used to take perhaps thecustomer to try to explain to
us themselves hey, tell us aboutyour business and then write a
few sentences on their business.
Then we might do our own duediligence, which is like okay,
(31:27):
let's look at your website,let's see what you're selling,
let's um, maybe even we mightvisit the site at times if
that's necessary, um, butactually, since we've replaced
that with ai agents, we actuallyhave seen that the agents
themselves can go and crawl thewebsite.
They can go and learn moreabout the business, they can
understand the language used bythe business.
(31:48):
For example, we shouldn'tprovide financial services to in
some regions, military goodsprovider.
But if an e-commerce retaileris actually selling a military
jacket because they're aclothing store, that type of
thing used to trip up the sortof less sophisticated systems,
but now AI agents can reallyunderstand.
Yeah, this is definitely aclothing store.
(32:08):
They're selling a variety ofthings that might include a
champagne dress or a militaryjacket, but I'm not going to get
tripped up by those types ofkeywords and therefore we have a
much better understanding ofthe customer and we're finding
it to be far more accurate thanhumans doing this themselves.
Speaker 1 (32:25):
So you're playing globally.
How will agentic AI transformcustomer interactions when it
comes to global payments?
Speaker 2 (32:34):
Great question.
I gave an example earlier inour discussion around you
wanting to, or a British companywanting to, actually sell to a
customer in Singapore, and Italked about the benefit of that
customer in Singapore usingtheir own local payment method.
Now, I don't necessarily wantmy British merchant to have to
understand every type of paymentmethod or say hey, customer,
(32:59):
here's a selection of 100payment methods.
Why don't you pick the one thatsuits you?
But using AI, we can look atthe available payment methods
that we offer.
We can look at where thecustomer is, we can look at the
currencies available and we cansuggest to that user hey, we
think that this would be thebest payment method for you, or
this would be the cheapestpayment method, or this one
would give you buyer protectionsand we can start to orchestrate
(33:24):
the best ways to utilizefinancial services around the
world.
And this is something that Ithink no business should have to
worry about trying to get theirhead around, and no buyer
should want to have to worryabout it too.
The buyer just wants it to beeasy and intuitive to them.
The business just wants to sellto the customer and provide a
great service or solution, andso AI can actually help us all
(33:45):
understand the complexity of theglobal financial services.
And then, of course, thingslike regionalization and using
internationalization of theproducts.
So we're changing the differentlanguages all of those
transaction services and otherthings far easier and far faster
to perform in AI than they werein the past.
Speaker 1 (34:05):
So we talked about the fact that the technology is
going a lot faster thanregulators can keep up.
So what challenges do you thinkwill arise in deploying
autonomous AI in regulatedfinance?
Speaker 2 (34:19):
It's a good question.
Speaker 1 (34:21):
Only ask good questions.
Speaker 2 (34:25):
Again, this is already a very heavily regulated
industry, and so I'm actuallyquite confident that we have a
strong grasp of what are theregulations and how to stay
within it.
I think we're not going todeploy these agents
irresponsibly.
I think that we talked abouttrust earlier, and understanding
(34:47):
your money and where it's going, and that all the numbers match
up, I think, is one of thehighest levels of trust required
, and so I think we're going toapproach this with the right
level of responsibility, thatwe're not going to let those
agents sort of go wild.
When I think about other usecases, if you're a media agency
or a marketing agency using AIto produce all of your different
materials, it's probably not asconsequential if you get it
(35:09):
wrong, but I would certainlyrecommend that you don't sort of
take it too far and end upproducing content that doesn't
make sense.
But in that particular case, itmight be okay to take some more
liberties to take advantage ofthe latest technology or
efficiencies.
I think in financial services,we're going to make sure that
we're performing the right sortof regulatory responsibilities.
(35:31):
There are going to be othersolutions that come forward.
If we look at cryptocurrenciesand stable coins, for example,
regulation is changing veryquickly.
Ai may actually be able to helpus understand that regulation
more readily.
It may help us to actuallyadapt to changing regulation all
over the world.
So in a lot of ways AI canactually help us to make sure
(35:52):
that we actually are beingcompliant with all the different
regulatory bodiesinternationally.
Speaker 1 (35:57):
So in all of my keynotes I'm trying to change
the mindset of my clients tothink of AI not just as a tool
that can craft an email orreview a document but how do you
think AI, and I'll say thatagain.
Or review a document, but howdo you think AI?
And I'll say that again.
So in all of my talks I've beenreally trying to change the
mindset of my clients, to getthem to not think of AI as just
a tool, something to craft anemail or to check a proposal,
(36:18):
but actually become a decisionpartner.
So how will data or AI evolvefrom operational tools to become
a strategic decision driver?
Speaker 2 (36:28):
When we talked earlier about the benefit of,
you know, setting up the rightapproval flows, for example for
for an expense, and we talkedabout how it used to be, that I
could configure this as a userand then I could actually have
ai configure it for me.
This, I think, is theinteresting strategic
opportunity, and so while we weprovide a little tooling to say
(36:50):
you can, for example, if thetransaction is above $500, get
your manager to sign off andthen get the finance team to
sign off, and if it's over$5,000, get the CEO to sign off,
but now, with AI and asAirwallex has grown to support
many customers, we can look atthe entire corpus of our
customers in an anonymous wayand say hey, businesses like
(37:11):
yours typically use thisapproval flow.
If it's above $50, do X, ifit's above $500, do Y, and if
it's above $5,000, do somethingelse.
And so, instead of just givingthem the tools to do it or
automate the tools, we canactually suggest the right
structures.
One thing that I've seen isthere's a lot of challenges in
(37:32):
the financial services industrytoday, and especially, it's true
, in the UK, where people aretricked into sending money to
the wrong place, for example.
There's two things AI can dothat can really help this
situation.
Firstly, they can actually tryand prevent that because it's an
unusual transaction and promptthe user to say, hey, I think
this is probably not what youthink it is.
You should probably not proceedwith this transaction.
(37:53):
But if AI was to suggestapproval flows for a business to
say, hey, we think that youshould always have two people
check the transaction before itactually goes out the door.
You can actually not only makeit a more secure system for
those users or make sure theyhave a lot of approval flows,
but you can actually make it amore secure system for those
users or make sure they have alot of approval for those, but
you can actually make it a morelike fraud resilient system as
(38:13):
well.
And so if AI can actually lookat similar customers, look at
their cash flows, look at theirunderstanding of hey, you know,
it seems that actually you havea lot of surplus capital and
you've got more funds coming inthan going out, why don't we
suggest that we put that into alike asset management product
for you so you can generate arate of return on the funds that
you hold in the interim?
Or in the opposite case, wouldyou like to provide lending
(38:37):
services that we think areachievable and within your realm
, to actually pay back but willreally help you with cash flow?
Those types of suggestions, Ithink, are going to become very
synonymous with AI and thereforeit's much easier for businesses
like Airwallex or other serviceproviders to create a much
better user experience andinfluence the right decisions
for our customers.
Speaker 1 (38:57):
So final question before we go to the quickfire
round what excites you the mostabout what you and the team are
working on at Airwallex at themoment?
Speaker 2 (39:05):
I think what I'm most excited about is many
businesses don't have themanpower to have a financial
operations team.
Large businesses might havedozens of people, small
businesses probably it's justthe owner of the business
themselves doing a lot of thisaccounting, doing a lot of these
transfers.
That represents your chieffinancial officer, that
(39:30):
represents your financialoperations team.
That didn't previously exist.
Then we're giving you access tosomething that you otherwise
couldn't have had access to, andinstead of you sort of having
to do all the heavy liftingyourself and understand these
systems, you have an agenticpartner who's actually saying
let me handle these transactionsfor you, let me close the books
for you at the end of the month, let me handle these tax
obligations, let me control thetransactions for you, let me
(39:51):
close the books for you at theend of the month, let me handle
these tax obligations, let mecontrol the card spend of your
employees.
I think that is a model that isgoing to be, or a future that
is not very far away.
That is going to be verypowerful for businesses.
It means all of their financialservices are handled and they
can actually just get on withdoing their core business, which
is not making transfers.
Speaker 1 (40:10):
So we're almost out of time.
Right to my favourite part ofthe show, the quickfire round,
where we learn more about ourguests iPhone or Android iPhone
and I really wish I didn't haveto say that I tried very hard
with Android for many years.
Window or aisle Window, yourbiggest hope for this year and
next Window.
Speaker 2 (40:33):
Your biggest hope for this year and next?
It's not a hope, but I'm veryexcited about AI always
continuing to scale over thenext two years.
Speaker 1 (40:38):
I wish that AI could do all of my.
Speaker 2 (40:45):
Recruiting.
Speaker 1 (40:48):
The app you use most on your phone Strava.
The best you use most on yourphone Strava the best advice
you've ever received.
Speaker 2 (40:57):
I think it was the recognition that some advice
should be ignored.
Speaker 1 (41:06):
Good point.
What are you reading at themoment?
Speaker 2 (41:14):
Great question, good point what are you?
Speaker 1 (41:17):
reading at the moment .
Uh, great question, can we skipthat one?
Yeah, sure, well, I canrecommend a great book called
digitally curious that I wrote,but uh, probably not.
Won't get you one in time forthat.
Speaker 2 (41:25):
Uh, I think the reason why it's hard, if I may
say.
Actually, it's because I'vebeen listening to.
I've replaced books withpodcasts, okay, and the podcast
that I've been listening torelentlessly recently is
acquired oh yes, I need to pickthat one up.
Speaker 1 (41:40):
So a follow-on question who should I invite you
?
Speaker 2 (41:48):
should talk to another ai founder.
I have one in mind.
Speaker 1 (41:54):
We can talk about it offline sure, how do you want to
be remembered?
I think, building great techbusinesses so, as I'm the
actionable futurist, what threeactionable things should our
audience do today to betterunderstand how agentic AI will
power the future of finance?
Speaker 2 (42:15):
I think podcasts are a fantastic way to learn in sort
of bite-sized information,which I think is why we've
shifted from books to podcasts.
I think setting up your habitsso you don't just always go to
Google, but you start actuallyspreading your questions around
and learning what works for youin terms of the different AI
services around.
(42:36):
And the third advice I would beexperimental, be experimental
and always ask the question canthis be done a better way?
Speaker 1 (42:50):
Shannon, this has been a fascinating discussion.
Speaker 2 (42:57):
How can we find out more about you and your work?
I think Airwallex is very easyto find.
We'd love you to go to thewebsite, if you're a business,
and learn about the servicesthat we can offer you.
It's a broad range of servicesthere.
My details are on LinkedIn andI talk a lot about Airwallex and
technology there, so thosewould be great places to start.
Speaker 1 (43:14):
Thanks so much for your time and safe travels.
Speaker 2 (43:16):
Thanks, Andrew.
got a fast connection , but at the end I can see how
much percentage is uploaded.
It basically uploads a perfectcopy in the background as we're
speaking.
Speaker 2 (00:08):
Sounds good.
Speaker 1 (00:09):
Okay, cool, I'm just going to bring up my script and
we'll go from there Three, two,one.
Today in the podcast.
I'm delighted to welcomeShannon Scott, senior Vice
President and Global Head ofProduct at Airwallex, one of the
world's fastest growing fintechinnovators.
Shannon's journey is afascinating one.
With a background in computerscience and mechatronic
engineering, shannon brings aunique technical and strategic
(00:32):
perspective to the world ofglobal finance.
At Airwallex, he's at theforefront of developing cutting
edge AI and data drivensolutions to transform how
businesses manage internationalpayments, compliance and
financial operations.
Today, we'll explore Shannon'scareer path, airwallex's vision
and dive deep into thetransformative role of a genting
(00:52):
AI in financial services.
Welcome, shannon.
Speaker 2 (00:56):
Thanks, andrew, it's great to be here.
Speaker 1 (00:58):
Nice to have a fellow Aussie on the podcast yet again
.
And we share similar traitsbecause we both come from
engineering backgrounds and I'dargue that engineers think in a
slightly different way becausewe were taught to think of
things from first principles.
That might come out in adiscussion, but perhaps you
could walk us through yourjourney from growing up in rural
Victoria to becoming SVP,global Head of Product at
(01:18):
Airwallex.
Speaker 2 (01:20):
Yeah, absolutely Like .
I think it's been a wonderfuljourney.
Actually, I did grow up inrural Victoria but had the
opportunity to go to theUniversity of Melbourne.
I knew I was good atengineering.
I was very interested incomputer science.
This is around the time of, Iguess, the dot-com bubble
bursting.
A lot of hype, a lot of tearsperhaps for those already sort
(01:43):
of in industry, but I sort ofsaw it as something that was
going to be very exciting intothe future.
One thing that was very luckyfor me as I went through my
education is the graduateopportunity that I had following
my university was actually avery small business and it was a
combination of buildingsoftware for the insurance
(02:03):
industry and then also makingsure we were working very
closely with our insurancecustomers to ensure they got the
most out of that software.
And so it was a greatcombination of software
development, understandingwhat's possible to build,
understanding the models and howthey work under the hood, but
then also really understandingthe customer experience and what
it means to build software thatthey can actually understand
(02:24):
and use.
And I think over my careerjourney there's always been a
software element, but it's oftenalso been on the commercial
side, on the product side,really thinking about how you
can bring technology toconsumers, to businesses, to
users and make it as powerful aspossible.
Speaker 1 (02:41):
I had a chuckle there when you were talking about
going to university during thatdot-com era.
I was at Telstra during theTelstra dot-com era when they
were trying to become a Yahooand of course they realised they
were good at being a telco.
So interesting times around theturn of the millennium.
We talked about the fact thatwe're both engineers or come
from an engineering background.
How does your engineeringbackground influence your
(03:02):
product leadership at Airwallex?
Speaker 2 (03:10):
I think your comment about first principles is
actually a really important one.
I think the better that youunderstand what's happening
under the hood and can reallyreason with how this technology
or service is operating, you canultimately create a much better
product or much better productexperience.
So you know to give an example,I think probably our
wallet-extended financialservices.
We provide both acquiring andissuing, so both the card
provider and, where you know,your audience, would enter those
(03:33):
card details online to acquirethe payment.
People don't often think aboutwhat's happening under the hood.
There.
They're just like okay, mymoney's been deducted from my
bank account and it appears inthe business that I'm purchasing
from, but actually it's just acomputer network.
It has all the same sort ofquirks that you might find if
something doesn't necessarilyalways go right when you use
your own computer.
It has some strange behaviorsthat are there, maybe because
(03:55):
that was a legacy of cards beinga service that used to be not
digital, and so there's somethings that aren't perfect, but
some things you take for grantedhow they work.
If you really understand how itworks under the hood, you can
eliminate those things you'retaking for granted or those
assumptions and actually thinkabout how can I make this better
overall?
If you don't have that depth ofunderstanding, you're probably
(04:17):
not going to question thoseassumptions or understand what
can and can't be changed.
Speaker 1 (04:22):
Now you're inarguably one of the world's fastest
growing fintech companies, somaybe give us an overview of
Airwallex's positioning in theglobal fintech market and your
role scaling a company valued at$6.2 billion?
Speaker 2 (04:34):
Yeah, so this is actually our 10th year.
I think we're coming up for 10years towards the end of the
year.
We've reached 1,800 employees.
We're across 25 differentoffices.
It was actually founded inMelbourne, but we've very much
sort of grown out all over theworld with a strong presence
here in London, where I am today, and then a number of other.
I think the sun never sets onour Airwallex offices.
(04:56):
The product itself is helpingbusinesses who need to operate
globally like really easilyinteract with their financial
services, and so if you're abusiness in the UK, you have a
customer in Singapore, forexample, you want to receive
pounds, but your customeractually wants to use a local
payment method that is uniqueand understood in Singapore and
they want to pay in Singaporedollars.
(05:17):
We make that an incredibly easyexperience.
It doesn't matter that you'reoperating in two different
countries.
Actually, both parties aregetting the experience that
they're used to, and this meansa business doesn't need to worry
about setting up differententities all around the world.
They can just get started withAirwallex and they can pay
suppliers internationally, theycan hold different currencies
and they can sell to customersall around the world, bringing
(05:40):
together all of those differentfinancial networks having the
licenses to operate in all ofthose regions takes a huge
amount of energy and effort, butthe team has been working
really diligently to just keepbuilding out that network and
making it as easy as possiblefor our customers to get access
to those global services and Ithink that's been a testament in
how the product's been adoptedour customer growth over the
(06:01):
last 10 years and now around$800 million in annual recurring
revenue.
Speaker 1 (06:07):
So we're recording this middle 2025, and I've spent
the first half of this yeartalking about nothing but AI.
So I feel like, almostcontractually, we have to segue
into the world of AI.
So how is AI fundamentallyreshaping financial services?
Speaker 2 (06:22):
Yeah, I think there's probably one other point I
should make about AirWallexitself, and I think one thing
that's interesting is thatfinancial services often are in
aid of some other intent.
Right, for example, a customerwants to purchase something from
your website, or you want topay a supplier who's providing
services to you it could bemaybe cloud services.
It could be the materials thatyou then on-sell, be maybe cloud
(06:44):
services, it could be thematerials that you then on-sell,
and so the payment is just apart of that intent to purchase
from the supplier.
Whatever that might be.
Technology and fintechs aremaking it really easy to not
just support the financialtransaction, but also all of the
workflows that businesses needor they potentially might do in
a manual way.
We support all of thoseworkflows, from onboarding the
(07:07):
vendor all the way to sort ofmaking the payment and
reconciling the tax aspects, andso when we think about
technology transformingfinancial services, I think the
first step is it's a lot moreabout business process and
aiding that process than it isjust the transaction.
So when we think about AI,there's a lot of different ways
AI can actually slot into boththe financial service and then
(07:28):
the business workflow.
On the financial service side.
There's things that we've beendoing for many years.
So risk management,understanding your customer,
fraud detection, fraudprevention, regulatory
compliance all of those aspectsI think have been using a
variety of sort of modelingtools and then more modern AI
solutions to support thosetraditional finance aspects.
But one thing I find reallycool is when I think about
(07:51):
building a product that's agreat product for our customers.
How is that AI actually visibleand supporting those customers?
So often that workflow foronboarding a vendor all the way
through to payment is performedby a financial operations team.
It requires a lot of steps.
It requires a lot of steps.
It requires a lot of approvalflows, and AI can actually
fundamentally automate a lot ofthe things that used to be
(08:11):
manual and what used to takehours or perhaps a lot of man
hours to process many vendorsactually comes down to just
minutes.
So I think that's one of thereally exciting applications of
AI that customers directlybenefit from.
It's not just happening underthe hood.
One other thing I'll say aboutAI that I think is really
exciting is it's changing theway that humans interact with
(08:32):
computers.
So your operating system isyour iPhone and the apps on the
iPhone it's your desktop, it'syour browser, but it's
increasingly becoming your chatinterface and it's a very
natural way for humans tointeract with other humans.
You know we're talking throughchat now or using text on our
phones More and more.
I think you're going to usethat as a channel to interact
with the services that you wantto.
(08:53):
So we're going to start to seemore and more services,
including financial services,being accessed through these
more natural language-basedchannels, and I'm really excited
to see where that goes over thecoming years.
Speaker 1 (09:06):
And what I find interesting is that you know AI
isn't new.
It's been around for 75 years.
Financial services have beenusing AI under the hood for
years now and I'm sure you'vebeen using it since day one.
So, to cut through the hype,you know what transformative
impact you think AI will Read itagain what transformative
(09:26):
impact will AI have in financeover, say, let's read that third
time what transformative impactwill AI have in finance over
the next, say, five years?
Speaker 2 (09:35):
I think it's going to be a much higher trust from
customers.
Even though financial servicesrequire pinpoint accuracy and
certainty in what transactionI'm making or how the money's
been received and thereconciliation, I think there's
still going to be a hugedelegation to AI to automate a
lot of those different services.
And so when I think about theway a lot of financial services
(09:58):
are offered today, it's like,hey, this is a business product,
I'm allowed many users on theplatform.
Those users might havedifferent functions, like one
might submit a request for apayment, another might set up
the payment and a third partymay even approve the payment to
make sure it's legitimate andgoing to the right person.
So I've got different userswith different roles happening
here and we provide a servicethat allows you to configure
(10:20):
that in such a way.
To say, if the price is, or thetransfer amount is, greater
than $500, then it must besigned off by two people, not
just one.
We have all the tooling thatcan do that, but we put the onus
on the user to actually say youcan set this up however you
like.
Ai is going to come in and notjust automate a lot of those
steps, but it's actually goingto start to make suggestions and
(10:40):
operate on your behalf.
To say this transaction looksvery standard or legitimate.
This is something that's verycommon by the requesting user.
I've already been able toverify the account details for
you and they match the actualuser's bank account information,
and we have to have checked thename against the third party
database.
Therefore, I think this is avery standard transaction.
I'm going to recommend to autoapprove this, for example, and
(11:02):
just at some, at some stage ordepending on the amount, it may
simply actually auto-approve andoff it goes.
At other times, we may see thata user wants to just eyeball it
and say, yes, that's great, butwe effectively took what was a
series of steps by many usersand now it is just, in fact, one
step or one click to verifythat we have the right
information, and so it'sactually those financial
(11:23):
operations that are going to beautomated, collapse down and
give businesses a lot of timeback to focus on their core
business.
Speaker 1 (11:31):
I was talking to my friends at SAP Concur they do a
lot of expense management andexpense reporting and they were
saying in the future we mighthave, rather than expense
reports, we might have exceptionreports.
And you kind of detailed thatthere so that if it looks normal
and if AI can see the trendthat it looks like a normal
transaction, only the exceptionwill go for human approval,
which is going to save a lot oftime than looking at every
(11:51):
single transaction.
Speaker 2 (11:53):
Yeah, absolutely.
And even things like uploadingthe receipts or sort of
understanding, like taking aphoto of the bill that needs to
be paid, for example, like thatcould already be automatically
uploaded if we can sort of lookat the profile of the image or
the document and recognise thatit's actually part of this
(12:13):
business transaction and it goesstraight in.
So you don't even have to bedoing the steps that upload the
documents.
Speaker 1 (12:20):
So the advantage for a company like yours is you're
fairly young, so you've got atech stack that's fairly new.
Speaker 2 (12:38):
When we look at traditional banks that have got
legacy systems.
How should traditional banksadapt to this AI-driven FinTech
competitors like yourselfchallenge they're going to find
is not only are we relativelyyoung, but I would say we are
tech first businesses.
We are not financialinstitutions, we are technology
institutions, and I think,therefore, we understand those
sort of emerging technologiesmuch more effectively and are
sort of bringing that into thefinancial service industry,
Obviously not only being afinancial institution, but being
(12:58):
an incumbent and largeinstitution for any business.
I think when you're large, itis hard to actually, you know,
sort of turn the ship towardsthese new technologies.
One thing that gives me a lot ofconfidence, though, is that a
lot of the services that AI isenabling and, if we take
customer support as an example,there's a plethora of great tech
(13:20):
companies vying for yourbusiness to provide really
wonderful customer supportsolutions.
Right, and this includes themodels and the languages that it
uses to understand the customerand how to respond to them.
It includes the case managementand the triaging and the
prompting.
It includes the embedding ofthe chatbot into your website or
the sort of call center oroperations team that you have,
(13:40):
and so traditional banks ortraditional businesses they're
not going to need to reinventthe wheel.
I think a lot they're not goingto need to reinvent the wheel.
I think a lot of thosetechnologies are going to come
to them in packaged ways, justlike they're using technology
today to be able to access theseservices.
So in that regard, I thinkthey'll be able to be supported
by a lot of other great techcompanies to sort of carry them
(14:01):
into the next era of finance andcomputing.
Speaker 1 (14:06):
So what we've seen with the adoption of generative
AI and also agentic AI.
So what we've seen with theadoption of generative AI and
what we'll talk about a bitlater, which is agentic AI
regulators are really scramblingto keep up because the
technology is just moving soquickly.
So what role does regulatorycompliance play in AI adoption?
Speaker 2 (14:27):
quickly.
So what role does regulatorycompliance play in AI adoption?
Look, I think it's mostcertainly extremely important.
I actually feel very confidentin our space in financial
services that it's actuallyalready very well regulated.
There are very clear guardrailson how information should be
utilised and stored, what typesof customers that you can
support, what types of sort offraud systems that you have in
place, and because all of thisis quite well understood, I
(14:49):
think you're creating a greatframework and sort of set of
guidance for how the agents, themodels that you bring into the
business, should be operating,and they can be quite small
agents with very specificfunctions, so it's well
understood what that function iswithin the sort of guardrails
of regulatory compliance.
I think there's a lot moreinteresting questions that
(15:10):
actually sit probably outside offinancial services.
That may include access topublic information, that may
include, like IP and those typesof issues where there can be
some really interesting questionmarks for AI.
But in financial services, I'mvery confident that we already
have the guardrails and a strongunderstanding of the current
guardrails from which we can useAI to build on top of.
Speaker 1 (15:34):
Now, the access to quality data has always been
something that companies haveneeded to worry about, and when
we move into this AI era, thequality of data becomes more
important, and I've been talkingabout the need for good quality
data for some time now.
But how should companiesleverage data as a strategic
asset in the AI era?
Speaker 2 (15:52):
It's a great question , and I think for data that is
generally, you know, readilyavailable or that is perhaps
even generated by the AI.
If we're looking at, you know,media use cases, for example,
you know businesses are going tohave access to that information
, but so is every other business, and so is every other sort of
AI product that's available.
And so what is the data thatyou have?
That is that you understandwell, that is structured in a
(16:15):
way that supports your businessuse case.
That is perhaps proprietarydata or sensitive data,
certainly because your customershave trusted you with that
information.
Like for most financialservices institutions, like you,
have data that isn'tnecessarily available outside of
that public realm, and so thatgives you a great opportunity to
, I think, build AI tools thatprovide a unique or value-added
(16:39):
service that other partiesaren't necessarily going to
provide.
So when I think about buildingnew AI features or how Airwallex
can really provide a greatcustomer experience, it's like
what am I actually offering thatthe customer can't get through
some of the means, or that isactually going to genuinely move
the needle for them in a waythat's built into Airwallex
(17:01):
rather than just something thatthey can do without necessarily
the services of Airwallex.
There are, of course, a lot ofdifferent types of data sources.
So you know, the financial datais very well structured and
requires very high accuracy andvery strong reconciliation.
And then there are data sourceswhere you're just trying to
understand hey, what does mycustomer look like?
(17:21):
What are the services that Ishould provide to them?
Is this a legitimate customer?
Do I understand their sort ofbusiness cases?
And in that case you're sort ofcreating a synopsis of that
customer to give them a greatproduct or service.
You know, once they'reonboarded, often AI can create a
better picture of who thatcustomer is and what they need
than a human can.
Speaker 1 (17:42):
So let's talk from about future.
Proofing.
How would you build a datainfrastructure that support both
current and future AI needswhen we actually don't know what
the next AI state might looklike in three or four years?
Speaker 2 (17:56):
It's a really good question and I think that it
would be um naive to have toostrong an opinion here, because
because you never know sort ofhow quickly things things move
um and and sort of what servicescome down the road.
I think we are seeing aconsolidation of different
services, and you mentionedagentic ai, and while ai has
been around for a long time likethat term is relatively new um,
(18:18):
like mcp service, for example,and these are sort of the units
or APIs that different AI agentsinteract with to perform a
given service is also arelatively new concept, and you
see the market solidifyingaround these things and then you
can start to build on top ofthem and create really great
solutions.
I think any business should bethinking about experimenting and
(18:40):
keeping up to date with thelatest sort of AI technologies,
but then also really thinkingabout where am I investing?
Where am I using sort ofbest-in-breed products that I
can easily absorb into mybusiness or build tooling around
?
If that best-in-breed productactually a new version comes out
in a year from now or itactually becomes even better in
(19:01):
two years, can I swap that outquickly and bring in a new
solution, to give you an example?
So, on that expense managementside, we used to use a
third-party provider to use OCRand entity extraction to
understand the contents of thereceipt.
What were the items, how muchwas it, what was the total?
What was the tax, what was thedate?
Um, it worked okay and itprobably cost us about 20 cents
(19:22):
per receipt to actually uploadit.
Um, in the last 12 months weswapped that out with google
gemini.
Um, the cost went down by aboutfour orders of magnitude, so I
think it was like 0.002 cents orsomething similar, and the
accuracy was dramatically,dramatically better.
The product was the same, youknow, it was exactly what the
users wanted.
In fact, it was better becauseit was higher accuracy.
(19:43):
But with the same intent, butbecause we'd built quite a good
modular solution, we were ableto swap in the latest technology
and get a much better outcomewith very little effort.
Speaker 1 (19:55):
Just to jargon bust MCP.
What does that stand for?
Speaker 2 (19:59):
Model Context Protocol, I think, and I suspect
that didn't really help youwith any further information.
Is that fair?
Speaker 1 (20:08):
I'll look it up.
I'll ask my AI friend and I'lllook it up and I'll drop in the
right term post.
Speaker 2 (20:15):
I actually feel a bit it's a relatively new
technology that I feel a littlebit sheepish explaining to your
audience and so I hope I don'tget it wrong but AI agents, for
example, those using chat theymight have to call on an
external service.
So, for example, if you typeinto the chat hey, I'd like to
make a transfer to Andrew for$100 AUD, please, please, make
(20:38):
that transfer for me.
In the background, the chat hasinterpreted that request and
made a request to a differentservice to initiate the transfer
or for that service to say okay, well, in order to do that
transfer, I need you to enteryour password, for example, and
(20:58):
so that tooling that understandshow to take that instruction
and provide a response iseffectively that MCP server, and
these are different units thatoffer that different functional
pieces of work.
Speaker 1 (21:13):
Now, one thing I've been saying to audiences for a
while that seems to resonate israther than thinking of AI being
artificial intelligence, theyshould think of it as augmented
intelligence, because there'salways got to be a human in the
loop.
So what role does humanfeedback play in AI systems, for
example in fraud detection?
Speaker 2 (21:31):
Yeah, fraud's a great example.
I would argue that we don'talways need a human in the loop,
but I think it can actuallyalways create a much better
solution.
So if we look at fraud, it isactually a heuristic.
A transaction happens inreal-time.
Maybe you've made a purchase onyour credit card, maybe
somebody else is making apurchase on your stolen credit
card and we have to make anassessment at the time.
(21:53):
Do we approve the transaction?
Do we decline it?
We're looking at things likewhat was the amount and the
nature of the purchase.
Is this sort of typical of thatparticular individual?
Do I have extra information,like the IP address of where the
purchase was made, so I know ifit's happening in a country
that is different to where youare?
We're using a number of thesedifferent factors to understand
(22:15):
what's happening and we'remaking a heuristic decision that
may or may not be correct.
Now there's a couple of reasonswhy you find out if you're
correct, and feedback, whetherit's automated or from a human,
is going to continue to refineand improve the model.
So if you let it through and itturns out to be fraudulent,
you're definitely going to findout about it, because the
customer is going to let youknow, and that would be sort of
(22:36):
an unfortunate outcome.
However, what we've recentlyintroduced at Airwallex in our
product now is sometimes wemight decline a transaction that
we believe is fraudulent and itmay not be, and the customer
would be very frustrated ifsomething's been declined when
they were making the transaction.
So in real time we send them atext message that actually just
asked hey, this was the transfer, was it you?
(22:58):
And if they say no, it wasn't,it's like cool, we blocked it.
Or if we hadn't blocked it, wesay actually, in that case we're
going to cancel your cardbecause somebody else tried to
do it.
And if it was you, we saythanks very much, please try the
transfer again.
We'll make sure we let itthrough the second time.
And so not only is it a goodcustomer experience, because
they can sort of correct for aninaccuracy in the model, but
(23:20):
it's giving us great feedback asto the quality of the model and
we can use that to keepimproving our fraud
understanding.
Speaker 1 (23:27):
So the other side of the coin is if we're going to
trust the AI a little bit more,how do you ensure AI remains
explainable and trustworthy forfinancial decisions?
Speaker 2 (23:39):
Explainable and trustworthy.
I think we should cover thoseconcepts separately.
I think actually explainabilityis a really interesting science
because if we look at the fraudmodel, for example, we may be
able to infer oh yeah, okay, theIP address was in a different
country.
The AI can sort of inform usthe data that was used to make
(23:59):
the decision and we might beable to sort of intuit.
That's the understanding.
But we're probably never goingto know exactly all the
different heuristics that themodel used to determine that
fraud, especially if it'sgetting this feedback from
customers over time.
And actually that's probablyokay.
That's probably okay as long aswe also have a metric for what
is the accuracy of our fraudsystem.
(24:20):
Can we get a sort of generalunderstanding that this
heuristic is operating with 95%accuracy?
You're catching the majority ofthe fraud, for example?
So I think the explainabilitypiece depends on the context.
The second item you'll have toremind me, andrew.
Speaker 1 (24:39):
Trustworthy.
How do we ensure that AIremains trustworthy for
financial decisions?
Speaker 2 (24:43):
So if we thank you, if we look at a financial
decision, so, for example, Isaid, hey, make this transfer
for $100 to Andrew, and Isuggested we might just do that
in a chat bot, for example.
Now that's a very importantdecision.
It must have perfect accuracyand it must know exactly which
Andrew I'm talking about and itwould be really disappointing if
(25:06):
it went to a different Andrewor the wrong currency or
something similar.
And I think, where that trustneeds to be proven and based on
the severity or importance ofthe request, it's relatively
easy to prompt the user to saythis is my understanding of what
you've asked of me.
Could you please verify thatthis is the case?
And because we do have thattwo-way communication with the
(25:28):
user, it's relatively easy tomake that request and establish
not only that trust but theverification.
I think again, the smallerthings.
Over time the trust willincrease and we're going to let
the agent do that for us.
We're not even going to worryabout it, we'll worry about
exemptions only, like yourexample earlier.
The bigger things we're alwaysgoing to include, we're always
(25:50):
going to include a verificationstep and if I think about the
example around structuringreceipts, it's actually probably
going to take a while before weget to 100% accuracy.
It's going to take a few yearsperhaps, but 90% accuracy with
human verification is stilldramatically faster than it was
for a human to do this on theirown, and so we're still getting
(26:10):
like massive benefit, even ifthe system is not necessarily
100% accuracy.
I think users can easilyunderstand that that's a huge
benefit and they're going tohave that trust that the model
is still giving them a lot ofbenefit.
Speaker 1 (26:25):
So we've got ahead around generative AI and most of
my friends are now talkingabout chat, gpt, as they used to
when they were going to Googlesomething, which is interesting.
But now we've got this newconcept agentic AI or AI agents.
First of all, how do you defineagentic AI, because everyone
has a different explanation, Ifound yeah, so.
Speaker 2 (26:46):
So when I think of a, I think actually it's a really
great, great question andeveryone probably should have a
different opinion and and alsothink about the benefits of of
those agents.
But, um, an agent is performinga work function that uh can
benefit, benefit the user, andthere might be a process that
actually has many differentfunctions and we can string
together those agents to performa more complex task with many
(27:11):
smaller tasks.
So, if I take the receiptstructuring, for example,
there's going to be an agentwhose role is to actually
understand and decompose thecontents of the receipt, and
then we're probably going tohave an agent that actually
understands, uh, the math behindwhat's happened, going to
happen next.
So I have a.
This agent has a understandingof what is an account number,
what is a currency, what is asort of financial amount, and
(27:33):
then I'm going to package thatup into an actual transaction
object that I can take to thenext step, and then another
entity is actually going toperform the transaction itself,
and so if we have these agentsthat are performing discrete
units of work, we can both haveAI that is composable and
understandable and doesn'tnecessarily become this sort of
behemoth system, but we can alsoachieve very complex tasks by
(27:56):
bringing together many of theseagents.
Speaker 1 (27:59):
So how will a Gentig AI redefine financial workflows?
Speaker 2 (28:05):
I think that example.
Let's break it down.
So I talked earlier about howfinance is not just making the
financial transaction, and so ifI look at all the businesses
today, it's like, okay, I needto decide which vendor I'm going
to use to you know, be my newsort of it chat system.
I'm going to verify that thevendor is able to perform the
(28:29):
features I need.
I am going to verify that thevendor has the right
accreditation.
I'm going to verify that wehave the right information for
that vendor to make to make a.
I'm going to go through aprocess of approving that that
vendor is actually the vendorthat we chose.
This is the right amount ofmoney.
When they send me an invoice atthe end of the month.
That is the right amount.
(28:51):
I'm going to verify the accountdetails.
I'm going to schedule a paymentfor the end of the month.
That vendor happens to be in adifferent country, so I'm going
to have to make an FX conversionfrom pounds to US dollars, for
example.
I'm not sure about currencyfluctuations and I don't owe
them until the end of the month,so I'm going to lock in a rate
from pounds to dollars that istriggered at the end of the
(29:11):
month.
So I'm going to createcertainty in the rate.
That doesn't happen until theend of the month.
I'm going to schedule atransfer.
It's going to happen.
Then, once that transfer hasactually taken place, I'm going
to reconcile that with myaccounting system, and so all of
those steps require a reallysignificant financial back
office and a number of differentplayers to perform that.
Today we have a system thatallows you to do it all in one
(29:32):
place, which is very valuable.
I can see the lineage all theway from the payment through to
the vendor and who it was to andhow it was signed off.
That's very valuable.
But utilizing AI and thatenergetic finance, we're
actually going to see all ofthose things collapsed into a
highly automated process thatrequires much less manual labor
is actually going to increase alot more trust.
We are actually going to seethe accuracy come up over time
(29:56):
and I think it's going tomassively redefine not only the
amount of effort it takes forlarge businesses, but it's going
to give small businesseseffectively the muscle to be
able to do this themselves orget that benefit when they
otherwise wouldn't have had theresources to be able to perform
this kind of financial duediligence on their vendors.
Speaker 1 (30:17):
So I think you're right.
We've just really only startedto hear about agenda AI.
Speaker 2 (30:20):
We need to scratch the surface, so maybe you could
talk about some of the use caseswhere autonomous AI agents
excite you the most also aboutthe value of understanding your
(30:43):
customer, and so there are verystrict guidelines in the
financial services industryabout what types of customers
you can support and what typesof solutions you can actually
offer them.
Perhaps there is risk for thosecustomers as well, like, for
example, they might be targetsfor other fraudsters to be bad
actors against their business.
So we want to make sure thatthe financial services they have
can be utilized responsibly andthat they actually are not
(31:07):
taking on too much risk, and sounderstanding who that customer
is is a really important part ofproviding them with the right
financial services.
I think what we're seeing andwhat we've already implemented
is what used to take perhaps thecustomer to try to explain to
us themselves hey, tell us aboutyour business and then write a
few sentences on their business.
Then we might do our own duediligence, which is like okay,
(31:27):
let's look at your website,let's see what you're selling,
let's um, maybe even we mightvisit the site at times if
that's necessary, um, butactually, since we've replaced
that with ai agents, we actuallyhave seen that the agents
themselves can go and crawl thewebsite.
They can go and learn moreabout the business, they can
understand the language used bythe business.
(31:48):
For example, we shouldn'tprovide financial services to in
some regions, military goodsprovider.
But if an e-commerce retaileris actually selling a military
jacket because they're aclothing store, that type of
thing used to trip up the sortof less sophisticated systems,
but now AI agents can reallyunderstand.
Yeah, this is definitely aclothing store.
(32:08):
They're selling a variety ofthings that might include a
champagne dress or a militaryjacket, but I'm not going to get
tripped up by those types ofkeywords and therefore we have a
much better understanding ofthe customer and we're finding
it to be far more accurate thanhumans doing this themselves.
Speaker 1 (32:25):
So you're playing globally.
How will agentic AI transformcustomer interactions when it
comes to global payments?
Speaker 2 (32:34):
Great question.
I gave an example earlier inour discussion around you
wanting to, or a British companywanting to, actually sell to a
customer in Singapore, and Italked about the benefit of that
customer in Singapore usingtheir own local payment method.
Now, I don't necessarily wantmy British merchant to have to
understand every type of paymentmethod or say hey, customer,
(32:59):
here's a selection of 100payment methods.
Why don't you pick the one thatsuits you?
But using AI, we can look atthe available payment methods
that we offer.
We can look at where thecustomer is, we can look at the
currencies available and we cansuggest to that user hey, we
think that this would be thebest payment method for you, or
this would be the cheapestpayment method, or this one
would give you buyer protectionsand we can start to orchestrate
(33:24):
the best ways to utilizefinancial services around the
world.
And this is something that Ithink no business should have to
worry about trying to get theirhead around, and no buyer
should want to have to worryabout it too.
The buyer just wants it to beeasy and intuitive to them.
The business just wants to sellto the customer and provide a
great service or solution, andso AI can actually help us all
(33:45):
understand the complexity of theglobal financial services.
And then, of course, thingslike regionalization and using
internationalization of theproducts.
So we're changing the differentlanguages all of those
transaction services and otherthings far easier and far faster
to perform in AI than they werein the past.
Speaker 1 (34:05):
So we talked about the fact that the technology is
going a lot faster thanregulators can keep up.
So what challenges do you thinkwill arise in deploying
autonomous AI in regulatedfinance?
Speaker 2 (34:19):
It's a good question.
Speaker 1 (34:21):
Only ask good questions.
Speaker 2 (34:25):
Again, this is already a very heavily regulated
industry, and so I'm actuallyquite confident that we have a
strong grasp of what are theregulations and how to stay
within it.
I think we're not going todeploy these agents
irresponsibly.
I think that we talked abouttrust earlier, and understanding
(34:47):
your money and where it's going, and that all the numbers match
up, I think, is one of thehighest levels of trust required
, and so I think we're going toapproach this with the right
level of responsibility, thatwe're not going to let those
agents sort of go wild.
When I think about other usecases, if you're a media agency
or a marketing agency using AIto produce all of your different
materials, it's probably not asconsequential if you get it
(35:09):
wrong, but I would certainlyrecommend that you don't sort of
take it too far and end upproducing content that doesn't
make sense.
But in that particular case, itmight be okay to take some more
liberties to take advantage ofthe latest technology or
efficiencies.
I think in financial services,we're going to make sure that
we're performing the right sortof regulatory responsibilities.
(35:31):
There are going to be othersolutions that come forward.
If we look at cryptocurrenciesand stable coins, for example,
regulation is changing veryquickly.
Ai may actually be able to helpus understand that regulation
more readily.
It may help us to actuallyadapt to changing regulation all
over the world.
So in a lot of ways AI canactually help us to make sure
(35:52):
that we actually are beingcompliant with all the different
regulatory bodiesinternationally.
Speaker 1 (35:57):
So in all of my keynotes I'm trying to change
the mindset of my clients tothink of AI not just as a tool
that can craft an email orreview a document but how do you
think AI, and I'll say thatagain.
Or review a document, but howdo you think AI?
And I'll say that again.
So in all of my talks I've beenreally trying to change the
mindset of my clients, to getthem to not think of AI as just
a tool, something to craft anemail or to check a proposal,
(36:18):
but actually become a decisionpartner.
So how will data or AI evolvefrom operational tools to become
a strategic decision driver?
Speaker 2 (36:28):
When we talked earlier about the benefit of,
you know, setting up the rightapproval flows, for example for
for an expense, and we talkedabout how it used to be, that I
could configure this as a userand then I could actually have
ai configure it for me.
This, I think, is theinteresting strategic
opportunity, and so while we weprovide a little tooling to say
(36:50):
you can, for example, if thetransaction is above $500, get
your manager to sign off andthen get the finance team to
sign off, and if it's over$5,000, get the CEO to sign off,
but now, with AI and asAirwallex has grown to support
many customers, we can look atthe entire corpus of our
customers in an anonymous wayand say hey, businesses like
(37:11):
yours typically use thisapproval flow.
If it's above $50, do X, ifit's above $500, do Y, and if
it's above $5,000, do somethingelse.
And so, instead of just givingthem the tools to do it or
automate the tools, we canactually suggest the right
structures.
One thing that I've seen isthere's a lot of challenges in
(37:32):
the financial services industrytoday, and especially, it's true
, in the UK, where people aretricked into sending money to
the wrong place, for example.
There's two things AI can dothat can really help this
situation.
Firstly, they can actually tryand prevent that because it's an
unusual transaction and promptthe user to say, hey, I think
this is probably not what youthink it is.
You should probably not proceedwith this transaction.
(37:53):
But if AI was to suggestapproval flows for a business to
say, hey, we think that youshould always have two people
check the transaction before itactually goes out the door.
You can actually not only makeit a more secure system for
those users or make sure theyhave a lot of approval flows,
but you can actually make it amore secure system for those
users or make sure they have alot of approval for those, but
you can actually make it a morelike fraud resilient system as
(38:13):
well.
And so if AI can actually lookat similar customers, look at
their cash flows, look at theirunderstanding of hey, you know,
it seems that actually you havea lot of surplus capital and
you've got more funds coming inthan going out, why don't we
suggest that we put that into alike asset management product
for you so you can generate arate of return on the funds that
you hold in the interim?
Or in the opposite case, wouldyou like to provide lending
(38:37):
services that we think areachievable and within your realm
, to actually pay back but willreally help you with cash flow?
Those types of suggestions, Ithink, are going to become very
synonymous with AI and thereforeit's much easier for businesses
like Airwallex or other serviceproviders to create a much
better user experience andinfluence the right decisions
for our customers.
Speaker 1 (38:57):
So final question before we go to the quickfire
round what excites you the mostabout what you and the team are
working on at Airwallex at themoment?
Speaker 2 (39:05):
I think what I'm most excited about is many
businesses don't have themanpower to have a financial
operations team.
Large businesses might havedozens of people, small
businesses probably it's justthe owner of the business
themselves doing a lot of thisaccounting, doing a lot of these
transfers.
That represents your chieffinancial officer, that
(39:30):
represents your financialoperations team.
That didn't previously exist.
Then we're giving you access tosomething that you otherwise
couldn't have had access to, andinstead of you sort of having
to do all the heavy liftingyourself and understand these
systems, you have an agenticpartner who's actually saying
let me handle these transactionsfor you, let me close the books
for you at the end of the month, let me handle these tax
obligations, let me control thetransactions for you, let me
(39:51):
close the books for you at theend of the month, let me handle
these tax obligations, let mecontrol the card spend of your
employees.
I think that is a model that isgoing to be, or a future that
is not very far away.
That is going to be verypowerful for businesses.
It means all of their financialservices are handled and they
can actually just get on withdoing their core business, which
is not making transfers.
Speaker 1 (40:10):
So we're almost out of time.
Right to my favourite part ofthe show, the quickfire round,
where we learn more about ourguests iPhone or Android iPhone
and I really wish I didn't haveto say that I tried very hard
with Android for many years.
Window or aisle Window, yourbiggest hope for this year and
next Window.
Speaker 2 (40:33):
Your biggest hope for this year and next?
It's not a hope, but I'm veryexcited about AI always
continuing to scale over thenext two years.
Speaker 1 (40:38):
I wish that AI could do all of my.
Speaker 2 (40:45):
Recruiting.
Speaker 1 (40:48):
The app you use most on your phone Strava.
The best you use most on yourphone Strava the best advice
you've ever received.
Speaker 2 (40:57):
I think it was the recognition that some advice
should be ignored.
Speaker 1 (41:06):
Good point.
What are you reading at themoment?
Speaker 2 (41:14):
Great question, good point what are you?
Speaker 1 (41:17):
reading at the moment .
Uh, great question, can we skipthat one?
Yeah, sure, well, I canrecommend a great book called
digitally curious that I wrote,but uh, probably not.
Won't get you one in time forthat.
Speaker 2 (41:25):
Uh, I think the reason why it's hard, if I may
say.
Actually, it's because I'vebeen listening to.
I've replaced books withpodcasts, okay, and the podcast
that I've been listening torelentlessly recently is
acquired oh yes, I need to pickthat one up.
Speaker 1 (41:40):
So a follow-on question who should I invite you
?
Speaker 2 (41:48):
should talk to another ai founder.
I have one in mind.
Speaker 1 (41:54):
We can talk about it offline sure, how do you want to
be remembered?
I think, building great techbusinesses so, as I'm the
actionable futurist, what threeactionable things should our
audience do today to betterunderstand how agentic AI will
power the future of finance?
Speaker 2 (42:15):
I think podcasts are a fantastic way to learn in sort
of bite-sized information,which I think is why we've
shifted from books to podcasts.
I think setting up your habitsso you don't just always go to
Google, but you start actuallyspreading your questions around
and learning what works for youin terms of the different AI
services around.
(42:36):
And the third advice I would beexperimental, be experimental
and always ask the question canthis be done a better way?
Speaker 1 (42:50):
Shannon, this has been a fascinating discussion.
Speaker 2 (42:57):
How can we find out more about you and your work?
I think Airwallex is very easyto find.
We'd love you to go to thewebsite, if you're a business,
and learn about the servicesthat we can offer you.
It's a broad range of servicesthere.
My details are on LinkedIn andI talk a lot about Airwallex and
technology there, so thosewould be great places to start.
Speaker 1 (43:14):
Thanks so much for your time and safe travels.
Speaker 2 (43:16):
Thanks, Andrew.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.