How Enterprises Actually Get ROI From AI — With Globant CEO Martin Migoya

Channel: Alex Kantrowitz

Published at: 2025-10-21

YouTube video id: 79mqihfbGnI

Source: https://www.youtube.com/watch?v=79mqihfbGnI

Let's talk about how AI is changing the
business world and where it might lead
with Martin Mcgoya, the chairman and CEO
of Globent in an interview today brought
to you by Globent. Martin, great to see
you. Welcome to the show.
>> Great to see you, Alex. Thank you so
much. Thank you so much for having me
here.
>> It's great to have you here. You have a
view into many many businesses and a
question we've been asking on the show
is, is AI actually helping businesses?
And if so, how is it helping businesses?
So just to start off, where is AI doing
a good job in business today and where
could it be better?
>> I think
let's kick it off by the overall view on
the on the technology itself. I think
it's something that it's a massive
revolution. Um this technology came to
stay here for many many years and um it
promised that it will you know kind of
change a good portion of how things have
been doing today and um that doesn't
necessarily mean replace humans uh but
also it will enhance
uh on the other hand it will enhance how
humans perform you know the same
processes that we have today. So um I
think that uh all these next generation
processes uh that will come with this
new technology
um u can be divided I would say into two
big categories one one which is pretty
you know straightforward how to imagine
them I mean if you're talking about
customer service that's something that
that uh is quite obvious how we can
accelerate how we can concentrate on the
difficult cases and take out all the
easy cases to use the human you know uh
intelligence to solve the difficult
cases. Um in those things are is quite
obvious uh and similar things like that
are quite obvious. Then you have other
things uh or other cases inside
companies in which it's not that obvious
how to use AI uh to make it and there's
a lot of confusion and noise around how
to analyze data, how to you know use uh
these next generation technologies to um
have better understanding of of the data
itself. Um how to analyze
better how customers react to certain
offers. And I think that there's a bunch
of things in which AI must generate like
the next generation of you know
processes which are not AI used by
humans but AI running the processes
itself and then humans supervising that
specific process. Right? So I would say
that these two cases there are some
cases in which it's much easier to
imagine it and some of the cases which
are more difficult to imagine and it
requires humans rethinking the way
processes are being done inside
corporations. That led to a lot of
change inside corporations. And it would
take time and it would take a lot of
effort because you know implementing
these kind of um probabilistic tools
into deterministic engines like all the
corporate information systems. It
requires a lot of effort. We're used to
in the software industry you know when
we have a problem there's a bug
associated with that problem. we go and
fix the bug and then the problem gets
solved. Well, this is not the case any
longer with with when you're talking
with a probabilistic system. You don't
know which are the things to touch
whenever you know you are not getting
the answer you want. So that creates
like a whole new level of you know um
difficulties on how to put the answers
of that probabilistic engine in the
range that you need for that to be a
good answer. right an enterprise class
answer. So those projects it has been
demonstrated already in these three
years that are more difficult than what
expected that it requires the right
partners. It requires uh to have a lot
of to pay a lot of attention to
corporate security to you know how you
put the guard rails of those you know
probabilistic systems or LLMs uh when
while you're there they are creating the
answer um and and this is a complex
process by itself so I think that um
this next generation
uh wave of change will come uh it will
take a while and it will be a slow a
slower than expected adoption on this
complex side while the easy cases will
be solved quite fast very fast
>> and so can you tell our audience a
little bit more about globun and and how
you are tackling these problems
>> we believe that um
all these AI AI things will kind of
uh evolve in in a in a pretty I would
say um
fast way but also there will there won't
be in my opinion one model that will
dominate the whole landscape instead we
will have like different models having
different you know spe specialties and
um the secret here to to future
solutions uh don't rely on just paying
attention to one model but to understand
which is the best model to solve your
specific problem. And I will separate
the race for AI in two specific, you
know, uh, portions. The first portion is
who's going to get the best LLM or the
best model for the specific business
case. And that's one that's one race,
right? And then the other race which I'm
more interested in is how to apply those
things to make real cases and how to
apply those things to change the way
companies operate. uh on that second
race is our race is what we are doing
every day in front of our customers.
While then many know LLMs today we
counted about 140 different models plus
versions of the model uh that are
integrated into our platforms. Um so so
what I'm saying is all these
technologies here the difficult part for
me is creating the agentic you know
workflows uh to be able to solve the
problem. So our approach to that has
been to create a platform which we call
globan enterprise AI. On one side it
connects to 140 models in which you can
choose whatever model you want to use
for the specific case that you need to
solve. On the other side, it connects to
all the corporate information systems,
SAP, Salesforce, uh any kind of service
now or whatever corporate information
system, even proprietary information
systems, right? And then with those two
things, you start creating the agentics
workflow, the agentic workflow that you
need to solve the problems that each
company has. And then with those agents
solving specific problems then we'll
allow our customers in case they want to
publish those agents and for them to
help other organizations or other
companies to solve those those same
problems. So let's say we're solving now
a problem for for a big energy company
on procurement right and and but there
that that problem is not just for one
company but pretty much everybody has
the same problem. So once we have that
kind of procurement ready agent and we
already have it now now we can
extrapolate it to many other companies
to use it. So our approach our playbook
is that
you don't need to
uh get the paralysis analysis of
understanding how the com extremely
complex AI ecosystem is. But you need
instead to get one step below that. do
your things, connect your corporate
information systems and let the world to
evolve on which is the best system and
then with very simple you know changes
connect to the best model that you need
instead of just maring with one model
right so what we are offering to the
market is we bring a very simple
approach to change your company using AI
without the need of getting a single
partner right or a single model into the
equation and with all the enterprise
class things that you need when you need
to take care of security, of
accessibility, accountability,
traceability,
uh cost control. Well, all those things
are things that we take care in our
enterprise class uh platform which is
called glob and enterprise AI that has
to do with that with a playbook and a
platform to adopt AI like the golden
path for AI adoption and gen AI adoption
um for our customers. So uh this is the
way we see it uh moving forward to help
our customers to create that value and
to make that those AI savings uh really
tangible for our customers using a
pretty clear playbook on how to use the
next generation technologies and how to
navigate this extremely complex AI
ecosystem that every day becomes more
and more complex. And what I want to
figure out is how this is applying in
concrete ways with real companies. So
can you just talk us through like one
example of a use case where a company
would come to you with a problem and
generative AI has been the answer to
solve it?
>> We have many and and that's one of the
main things how we are growing these
days and u I would say that give you two
or three examples. Uh first on the
e-commerce site, massive e-commerce
site, they need to they have like a an
army of people answering calls from the
from their uh customers and they're you
know saying um okay so I bought these uh
three uh glasses and one came broken. So
what should I do? So so the system go
ask for a picture. Then the system
itself check that picture. If the
picture is real and it connects with the
same thing that they have on the back
end, then they go to the next uh if if
that's not the case, then it call a
human. Puts a human in the loop. But if
that is correct, they go to the next
step which is comparing the nine options
about full refund, total refund, partial
refund or full return, uh partial return
or no return. and then we analyze you
know with an LLM uh which is the best
option and then we suggest to the
customer a final answer. So that very
complex process that is slice into very
thin uh layers of small decisions is one
of the cases that we have solved many
times. Now uh also we are as I said
before big energy company, big
procurement company to be able I mean
big procurement problems to be able to
acquire the things on time for their
production and uh so so okay I need a
valve. Okay that valve u the first
question is do we have in the warehouse?
If not do we have a contract with some
vendor that can provide that valve? If
not we need to go to an RFP and if not
we need to go to somewhere else. So all
that process is being automated in terms
of having an AI agent running that
procurement process and understanding
from the moment we need the piece to the
moment we deliver the piece how we
shorten that time. Uh on the software
development space we have been using it
uh with our Koda for many many uh
clients. Koda is our coding agent uh in
which we have uh taken care of a lot of
the enterprise class problems like
repeatability, reuse of pieces of code,
uh how to uh connect with coding styles,
uh many of the things that corporations
really need when they develop software.
And we have helped our companies not
just to create new software products but
also to uh uh to migrate for example
from very old software to next
generation software in a fraction of the
time at a fraction of a cost and that
has been extremely successful in some of
our customers in the financial services
space uh in in in Europe and in the US.
Uh so those are three specific cases in
which we have been using AI and these
next generation models, next frontier
models uh as um a a real help, real
value for our customers.
>> And can I ask you so you talked in the
beginning about these are probabilistic
systems. So when do you know let's say
you're in the procurement phase that you
can trust the system to automate it?
Well, look, I think that all these
probabilistic systems must must be
matched with human supervision. I mean,
there's not any of those systems can
work without a human supervising
uh at enterprise class with humans
supervising what those systems produce.
So I think at the end of the day you
gain a lot of efficiency when you chain
the agents together but you can never
forget about humans just checking what
those agents are producing because
sometimes they could hallucinate they
could do something that is wrong. Uh so
I think human intervention as always is
very very important. Um but humans
become much more efficient supervising
those things than you know just running
the whole process by themselves. So I
think that this is um
um a quite interesting way of
understanding. I think also that with
time and with models becoming better and
I don't know if becoming better the the
largest problem that you will face in
any corporation is to put the right
context in front of them all and that's
the largest challenge. I mean one the
context is correct the probabilities of
hallucination
goes much lower. So the largest question
we need to answer to implement these
things and make it you know enterprise
class is about the generation of the
context and in pretty much every company
generating that context is extremely
complex and that's why I'm saying this
probability system when you meet mix
them well it's about generating that
context in the right mind in the right
way for the current models not the
future models current models to answer
it right so most of our work has been
how we prepare that context, how we
prepare that prompt, how we create uh
from the information we have the most
accurate prompt for the system to solve.
And one step more is to check
by human if that answer is correct. You
don't need to check every single answer.
And there's kind of uh a lot of
efficiencies that can be made on
checking one answer instead of checking
middle answers. Um uh so the final
answer sorry instead of checking middle
answer but those checking points are are
being you know generated uh connected to
having an enterprise class answer and
this is extremely important
>> right that that context piece that you
uh mentioned I think we shouldn't
underestimate it right it's these bots
are great when the context is right u
but then when you see them try to
incorporate a context that's too large
or expansive, that's kind of when things
start to mess up. I mean, I was just
speaking with um an Amazon executive
about this and talking about Alexa Plus,
and I think the conclusion that we
reached is that the hardest part of this
entire uh uh application is to, you
know, narrow down or to direct the bot
to the right stuff. And that's why we've
seen I I think that's why we've seen
delays there. And the Apple intelligence
issue is if you try to get all the data
uh into the thing at once and you're not
careful about the way that it looks for
it, you're going to have a mess.
>> Yes, absolutely. And and one thing more
I think that we have all been expecting
AI to solve problems
that basically are problems of of a poor
context
and uh the current technology we have
even if it doesn't gets better in the
coming years which is not the case. the
current systems we have, the current
technology we have, if you put the right
context, it gives you the right answer
and uh and you can decrease a lot
hallucination percentages uh just by
creating the right context. Now, is that
easy? No, it's not easy. It's it depends
on the it depends on the kind of work
you want to do. But um but that's one of
the most difficult things. So agree 100%
with with that Amazon executive that was
saying that. So you mentioned about like
we we their humans will be checking the
work of the of the bots. Um and it just
goes to the question of like what the
workflow of the future will look like.
You know a way we've teased it out here
is asking will we incorporate the bots
into the person to a human's workflow or
do does the human get incorporated into
the bots's workflow? And when you talk
about like the person a human employee
will have to check what the bot's output
is, it seems more of that second version
that it's you know the bot has a
workflow and we become effectively
auditors. Is that is that how you see
it?
>> Yeah. Um long answer short the answer is
yes. Now let me let me try to be more
specific. I will launch a new way of
engaging with us uh which is called the
AI pots.
uh those AI bots are kind of uh the
equivalent uh to produce software uh
with humans but now being done by
Agentic AI. Uh so one once you buy a
subscription to one of those AI pots uh
AI agents on the back will produce that
software that you want or that you need.
And of course in order for that software
to be enterprise class and to be up to
the standards that we have been used to
we need to have some human supervision
checking yeah the software that these AI
bot are generating is you know is
correct. Uh and that brings us to a
whole new model of how to engage with
you know a professional services
company. You need to solve a problem.
That problem gets solved by agentic AI.
You have a monthly limit of tokens that
represents like a proportion of the
effort that that took to be created. You
have full transparency on which are the
assets that has been created, how many
tokens each of those assets, you know,
required. Um, but someone needs to check
the integrity of those, you know, of
those pieces of software. Um and someone
must check and must use kind of a
supervisory
you know process to understand that what
we are producing with those agents makes
sense and it connects right with
environment and connects right with the
coding styles and and many of the things
that at enterprise class you need.
Sometimes it's funny to to hear some
other you know uh companies talking
about generating code. they're kind of
reinventing the wheel each time they
need to create something. And corporate,
you know, software development when
you're trying to blend into a a a repo
that has one billion lines of code is
not about that. A lot is about reusing
what you already have there. So when you
are creating that, if you're reinventing
the wheel every time, then you're not
doing the things right. So those human
supervisions are taking care of course
our koda take cares of that by itself
but it's taking care of that we're using
the right components each time we are
trying to develop something instead of
reinventing the wheel from the scratch
so I think this is one of the most
important things but to your question
absolutely yes AI pot for us is an
agentic
process that is being supervised by
humans to ensure at our cost that the
quality of what we are producing is in
line with what the company is expecting
from Globat. Um so uh uh yes it's is
aentic an agentic process being
supervised by humans. The initial
version of AI was humans accelerated by
AI. The next generation is AI supervised
by humans. It's like when you were
talking about internet in the early days
and internet was a tool to serve you
know uh someformational websites uh and
some emails and suddenly happens that
someone invented commerce on top of
internet and it was very successful and
then someone invented social on that you
know and then someone invented you know
an application like yuber so those next
generations ways of use AI are still to
discovered and it will challenge many of
the business models out there. Um I
don't know if you if you if you saw the
open AI agentic you know uh agentic
commerce solution uh basically you're
letting letting agents to buy for you.
So instead of driving the traffic into
let's say an Amazon,
they will send the send the agent to
Amazon, they would purchase and the the
guy that was interested in buying that
will not even know if they don't want
even know what has been where where they
buy the thing. They just getting you
know like a blue t-shirt that's it. So
um this is changing how brands how how
relevant are brands. This is changing
how commerce is being executed. This is
changing how uh massive entertainment
companies are driving users to their
websites. Uh how they are controlling
what is said about the brand. Everything
my friend is changing. Uh so this is one
of the most amazing times in the history
to have a company like us that is
overlooking all those things and trying
to make those things to work properly
for corporations.
>> And so for you selling your services and
technology, you've you've shifted
pricing, I think that's what you said.
Does that mean you go from let's say
like an hourly model to charging for the
output of the agents? Like how does your
pricing shift? Correct. Um we're we're
changing from I mean we're still doing
hourly model and fixed price models
which are uh the dominant models at
globant. Uh but we're seeing like a new
way of charging for it. It's like the
same way that open charge you is like
per tokens per consumption you know in
this case would be like supervised
tokens because you are sure that what is
being produced is the quality that you
need. So what we are charging is uh per
consumption is a monthly subscription
that includes uh a certain amount of of
supervised tokens and then you can go
faster and you can consume them faster.
You can create software more reliably.
You can create uh like a much more
AIcentric
way of of of creating technology. Uh and
we're not using just for technology.
We're using that for creative services
too. We're using that for um uh for
enterprise migrations uh let's say SAP
or whatever you know so we're using this
AI pot concept for pretty much all our
offering in the digital in the
enterprise and the you know uh creative
and and marketing services studio. So
this is the the next generation of
pricing I think is is is it will be more
similar to what we do than to what it
has been done in the industry for many
many years. And by the way, I I haven't
been Brian Globin since the last 22
years, right? And uh forever my largest,
you know, concern was how to scale the
company using technology. And we have
been using AI since the last 10 years.
You think we got a patent in 2020, uh
very similar to Copilo. We didn't have
enough money to to be able to scale it
up. But but we knew
>> Yeah. We knew the technology. Uh and uh
my big dream was at one one day I will
be able to change the way I do this
business and remove a lot of the
friction that is present when you're
trying to develop something. And let me
give you a parallel.
when Amazon started to sell servers
just by the click of some you know
things that you choose on a website and
you put your credit card you well before
that process was an extremely complex
process it was about buying the servers
getting the hosting getting the
connectivity once you get the hosting uh
all the capital expenditures of buying
the servers for the peaks right and
suddenly
next generation offer came to the table
and said you don't need to worry about
all this just you know buy whatever you
need and you will get a commission in a
few. So they remove friction from the
process of scaling up a company and my
aim is to remove that friction from the
process of creating technology in a
supervised manner and uh and that's
where we are going with our AI pots
offering.
>> Okay, I want to end on this. um you
you've used this word pastor. Um it
seems to be the word of the of the the
year the past couple years and you're
somebody again who you see businesses um
with problems that they want to solve.
Often they're coming to you to solve
them with technology and now you have
AI. Um how much faster is the business
world moving today with AI and how much
faster is it going to get once companies
start to figure out how to put this
technology into action effectively?
I think we're in the early days, my
friend. Um, we're in the early days. I
mean, things are moving fast in certain
areas as I described at the very
beginning and much slower in in many
others. And and still remember, I mean,
there must be five or six% of companies
that are technically savvy to implement
these technologies by itself. But then
for the rest of the people it's like
like a jundle that is growing every day
and doubling every day in which you need
to understand where to where h how to
where are the roots to get somewhere uh
and and and you know for 95% of the
companies or maybe more 99% of the
companies uh getting that real path is a
big problem right it's a big problem
because it's difficult to implement but
also difficult to move in that jungle of
AI complexity
So I think that the speed of
implementation of all these new
technology, the speed of adoption of all
these technology will be modulated by
how companies like Globant can you know
make that process faster for those
customers that are not savvy technology
savvy enough to make it happen. So I
think that that study from the MIT
saying there are some big portion of
these implementations that are failing
has a lot to do with that. Many
companies don't have the way to scale
this up. They don't have the playbook to
scale this up. And that's a problem
because then you start doing things and
you're not getting the results. And when
you don't get the results, well, I won't
invest on that. Right? So I think
there's a very good um healthy I would
say doses of realism that is happening
on the AI space right now that is
signaling that okay this technology came
to stay is not that easy to implement
it's not that easy to choose in that
jungle but is something that can change
our business so everything will depend
on how many or how good we are right
companies that go and are to bring that
thing to the market and create real
business change and real business
acceleration for our customers. So um I
don't know that that that's uh that's my
view on on that on that uh subject.
>> Okay, great. And so if people want to
check out Globant and see how you work
with companies that are looking to put
this into action, where would they go?
uh they can go to uh globan.com or
globan.ai
and uh they will get you know the
perfect playbook to implement all these
technologies uh and make it happen fast.
>> There's that word Martin. Great to meet
you. Thanks for coming on the show.
>> Thank you very much, Alex. Bye-bye.
>> All right, everybody. Thank you for
watching. We'll be back on the channel
with another video soon.