How Generative AI is Transforming Post-Sales CX
You’ve heard the buzz - generative AI is revolutionizing the post-sales customer experience - but how are leading companies leveraging it effectively? In this panel discussion, industry experts will challenge conventional post-sales practices, sharing how AI-driven innovations are driving personalized support, faster resolutions, and deeper customer engagement. You’ll gain insights into how AI is transforming the way businesses deliver value after the sale, and leave questioning whether your support strategy is keeping pace with what’s possible.
You Might Also Like
So this is our first of several panels.
And this topic, AI, generative AI, all the components of AI is a complex one.
And we could unpack it for days.
We have a day, but this is going to be a high impact session.
We have a mix of great panelists representing different viewpoints in the
industry, from
builders, developers, and practitioners in the CX space.
So I won't steal John's thunder who will introduce the panelists, but I would
like
to welcome to the stage the host and moderator, John Ragsdale.
Many of you know him.
He is distinguished researcher and vice president of technology ecosystems at
TSIA.
Welcome, John.
And thank you.
Thank you, right to being here.
I have been such a fan and supporter of Krishna and support logic since it was
like five people
working in a Wii workspace on Stephen's Creek Boulevard.
So it's fantastic to be here at the first live event.
So if you think about it, you know, GPT came on the market just like a year and
ten months
ago.
And in my very long, long, long career, it's rare to see something have such a
quick and
transformative impact.
And I think that's because this is one of the first examples of AI that
everyone could
instantly understand.
And it affects you personally.
It affects you professionally.
And a survey that TSIA did found that 72% of B2B support organizations were
already using
GNAI for agent facing use cases and 50% were experimenting with customer self
service use
cases.
So adoption has been really, really rapid.
But I know that some companies are still early in their journey.
I talked to companies still shopping for use cases, trying to get budget.
But we're going to be speaking with three panelists today who are definitely
pace setters
around AI, ML, Gen AI.
And we're going to be talking about some of the potential for the technology,
but also
getting into the business value.
Because technology is cool, but at the end of the day, if we're not getting ROI
from it,
it's a problem.
So I'd like to welcome our panelists to the stage.
And first up, we have Sanjit Kaar Bali.
[APPLAUSE]
Thank you for coming.
So Sanjit is senior vice president of global customer support for N4.
And prior to N4, she held executive positions at Trellix, here technologies and
mobility.
And she is also on the advisory board for CSM practice.
So thank you for being here.
Our next panelist is Timur Yarnell.
[APPLAUSE]
[APPLAUSE]
[APPLAUSE]
[APPLAUSE]
[APPLAUSE]
[INAUDIBLE]
Timur is the head of AI and ML business development team for Amazon web
services startup.
So he is living at the center of innovation.
Previously, he was the CEO and co-founder of Neutronian.
And he has also held executive roles at Glean and Neva.
And our final panelist today is Bavech Metta.
[APPLAUSE]
[INAUDIBLE]
[APPLAUSE]
[INAUDIBLE]
[INAUDIBLE]
[APPLAUSE]
Bavech is senior manager of AI and automation for Uber.
He leads generative AI, conversational AI, and automation initiatives for
customer support,
directed in AI for strategy for Uber.
And previously, he held product engineering and strategy development roles for
Cisco.
So we opened the day with Krishna talking about really the silos of the post
sales customer
experience.
And I think it's fair to say, in the best of times, it's been a disjointed
experience.
And the worst of times, it's been incredibly painful for customers and for
employees.
So I want to start by asking each of the panelists, what do you see as the
single biggest challenge
for support organizations today?
Sanjid, could we start with you?
Yeah, I think, unfortunately, I don't have a very positive view on where we are
today.
So the leaders are leading, I would say, very overheated organizations.
The complexity is growing, especially in the complex product portfolio
companies.
The number of issues inquiries the volume is growing.
I think what is also happening is that the escalations are happening every day.
Customers are agents are dealing with outdated technology, not the right tools,
but then
to serve the customers right.
They're constantly in the firefighting mode.
So the whole view of us always wanting to get to that proactive level of
support is
still a laser, which is compounded by the fact that we don't have support
functions
as revenue function.
They still remain the cost functions.
So there's lots of challenges there.
And I feel that we're in the right place now with technology helping us scale
and creating
those positive experiences, but creating that bandwidth for our support people
to do the
right things for our customers.
So many challenges, but many opportunities to do.
Yeah, technology, culture, political, a lot of issues.
A lot of issues.
I think the escalations are basically the ones which are really really driving
the companies
to not fix their customer experience because they're constantly in that mission
cycle.
Yeah.
Yeah.
But Vash, what are your thoughts on top challenge for support?
Yeah.
I'll start by sharing a couple of examples, one from the enterprise business,
one from
consumer.
An enterprise that Cisco used to lead the data production engineering team
delivering
snapshot disaster recovery kind of solutions.
When you engage with customer, they're in a way already disappointed them with
your product
offering.
They come to the table, upset, angry, oftentimes, and oftentimes is the case
with enterprise
sales.
Somebody has put their job online to buy your product, especially let's say if
you're coming
from startup.
So it's not just about resolving that issue at that time.
There are a lot of emotions in play as well.
Similar on the consumer side at Uber, for example, if somebody's right doesn't
arrive
and they are on their way to the airport missing a critical meeting, food doesn
't arrive, the
family goes hungry.
Again, a lot of emotions in play.
How do you make them whole?
Your product experience, product defects have disappointed them.
In a way, not much has changed with respect to what people are expecting from
the support.
They expect you to have their back when things don't quite go as expected.
It may be a product issue.
It may be, in let's say, Uber's case, physical world issue, staffing, accidents
, and what
lot.
What is compounded with the situation is there is just enormous amount of, you
know,
technologies tools that people are using right now, Slack, WhatsApp, chat,
other chat interfaces,
voice.
What they don't want is they don't want to repeat the context.
Again and again, as you shift through different ways to help the customer, they
want fast,
accurate, context aware, and highly personalized resolutions.
The same resolution they may not be appropriate for.
Sometimes it's just acknowledging, hey, I hear you, I got your back and this is
how
we are going to resolve the issues.
As we head into AI, this multimodality is also going to become quite important
where
initially because of the business reasons, also we were guiding customers for
text,
ideally, optimally async support so that we are able to offload the operations
to maybe
cheaper regions.
Not anymore.
I think not only Uber operates in the economy, we as a society are operating in
attention
to the FASIC economy.
People want it quick, they want the issue resolved, they want to be on their
way.
So that's a big, big challenge but also opportunity that we are heavily leaning
into AI to solve
for us.
Fantastic.
Tim or what are your thoughts?
Well, I think Sanjid and Bhavash did quite well.
It's hard for me to say other than I think in the last presentation, showing
numbers like
77% or 81% of our customers want better experience of one faster resolution
that shows relatively
speaking what baseline we are.
I don't know, maybe it was part of speaking on this panel coming up and you're
aware of
it but I've had terrible support interactions in the last three weeks that I
can name.
Some with major airlines for example, I don't know how many people can say that
they've
got bad interactions in the last two or three weeks that come to mind for them.
So for me, I was trying to fly home from Seattle after we had this amazing
accelerator
kick off with Andy Jassy and Matt Garman and the height of technology and I
just wanted
to do a same day change on my flight.
And Delta put me into this infinite loop on my mobile app and on the call where
I was
not allowed to do a same day change to SFO because I was booked into SJC.
Fortunately I took the time and I pulled out my laptop and on my laptop, that
option was
there for some reason.
But it took me about 45 minutes of digging to get out of the infinite customer
support
loop.
Now just think about how basic that is.
By the way, I have over a million miles on Delta.
I'm a million mileer on Delta.
So with that as an anecdote of what we can point to, I am very excited by the
support
logic story for a number of reasons because I think the fundamental challenge
is how are
we going to use GNAI to augment human capability, all the gloom and doom
dialogue around replacing
humans and eliminating call centers to me so far fetched because we're starting
from
such a bad experience baseline.
So if we can use these signals to actually make the experience that much better
, I won't
dread calling support on equal par as going to the dentist.
And that's almost how I feel right now.
I mean I had another experience that I onboarded to new healthcare because I'm
only eight months
in to my role with AWS that was coming from GLEAN.
And I had two healthcareers going at the same time.
I had to call in and tell them which one was active.
And of course it took me six months to do that because I knew what was going to
happen
and I didn't have an hour and a half to waste because I'm onboarding to a new
role.
So with that being the upside, I think that's the challenge of just what we're
starting
with and we can use, I think a lot more we'll talk about using this to augment
human capability
is going to be so vital and enable so much better interaction with our
customers.
I think that you've all touched on really important challenges.
I think the biggest underlying challenge is something you mentioned that 60% of
B2B support
organizations are cost centers and they don't have the respect, the visibility.
We all know in this room the impact that a positive support experience has.
Sometimes getting company executives to understand that.
I think, and I'm just so sorry to interrupt you, but I think that's where the
difference
is because most of the companies are deciding on putting more resources and
investment in
customer success, and that's the right strategy.
But think about it, you cannot build a great customer success if you don't have
a solid
foundational support.
So it's like, we're not fixing support, you cannot fix success, you cannot
drive that.
So there is that tactical entity mode that we put support in and not seeing it
as strategic
is the biggest mistake that we are making.
And that's where there's a lot to change in terms of how we look at support in
terms
of its operating model.
We haven't revised that, we haven't gone back and looked into it.
We're still running support the way people have been running for 20 years.
So I think there's lots that we need to discuss and that's going to be probably
in our long
conversation.
Yeah, absolutely.
But I think that support logic is bringing that data to show the strategic
value of support
and instead of building all these health scores for success manager, building
health scores
to show the correlation between a positive support experience and the long term
customer
value.
So they were the very first company to build health scores for support.
Okay, well let's dive into some questions and Sanjid, I'm going to start with
you.
So comprehensive root cause analysis is basically nonexistent for most
companies.
And there's a lot of runs up the hill over the years.
Back in my early days running support, we had our agents stage a walk out
because we
were forcing them to use this decision tree that was so painful that they said
we would
rather lose our jobs than use this tool.
So could you talk about how AI can help with root cause analysis to not only
streamline
problem diagnostics but ultimately eliminate problems entirely?
Yeah, I think Krishna covered it in his presentation today.
If you look at the entire support logic functionality, it does address that,
right?
Like I always said and maybe I said it to Marie you as well that show me a
company's
RC and I'll tell you the culture and the capability and maturity of that
organization.
And so there are three parts to it which is culture, which is data and which is
bandwidth.
So we are all well-intended people wanting to do the right thing for the
business and
our customers but we don't have the bandwidth to do things like basic RC.
And if you don't do that, the systemic aspects of your products and other
experiences don't
get resolved so you're again in a constant vicious rush.
I think this case summarization, predictive analytics, escalation management,
data insight,
voice of the customer, all of these elements that support logic enables creates
that bandwidth
for us to look at this thing.
But the data on the ML side of things is about finding the right signals,
finding the right
patterns, right? The correlation and the data from different disparate sources,
all of this
intelligence is what makes it special for us to be able to do better job at RC.
Because RC is our painful extensive process but if you want to do it at scale
then you
really want to rely on these technologies to give you those patterns and those
signals
to device.
So I think this is the solution for creating that bandwidth and also creating
that data
and signals and patterns and inferences to do so.
And as technology grows more and more complex every year, it's really going to
be up to
AI to enable us to do this.
And think about the precision wrap that we talked about.
That is the resolution intelligence because in my B2B complex context, just
giving me
two knowledge-based articles does not help me.
And in fact, makes the agents say that I don't want to look at it because I
know this
stuff.
They want something that reduces that complexity and brings that information
that then enables
them to do better stuff.
So in regular B2B to be less complex product portfolio, that's a good way to do
the user.
But drag is the way for us to do more.
Maybe Timor could probably say better things with me but I feel like that's the
solution,
that resolution intelligence.
And the context continuity, that's really, really important.
And people don't have that.
Our agents today hop from five systems to six systems to just get basic
information to
solve something.
And by the time they get to that, they have two other escalations that are
popping up
somewhere.
So where do you get an RCA?
And then you don't do RCA, you don't have system issues.
So you are in that vicious loop.
Yeah.
And if I'm a relative to root cause analysis, oftentimes it's not even clear
whether it's
an issue.
And this is where we implement unsupervised clustering or anomaly detection
techniques.
Even flag.
Yeah.
So it might be something I'm here.
And it could be based on various system metrics that you have various business
metrics
that you may have.
For example, every consumer business venture into some sort of fraud.
At any time fraud is happening, you just don't know whether it's big enough or
whether it's
happening, where it's happening.
So you put some guardrails about, okay, you know what, I want to operate this
particular
line of business with this particular region.
And this is my budget.
If it goes week after week, a little bit beyond this, then it might be a cause
of concern.
So I think AI is already helping there.
And the second area, and I know support logic also has this feature.
When you have turn off defects coming in, to get to the root cause analysis,
you need
to lend it to the right subject matter expert.
And how do you route it?
Based on the intent taxonomy or the content within the support ticket.
I think that's also where AI is already helping us and showing data.
I think it's a great point to do.
I love what you said in terms of the delivery and the use of RAG retrieval
augmented generation.
And I guess just for show of hands, how many folks are at organizations that
are training
LLMs now or SLMs or some mode?
Yes.
Okay.
So it's pretty good.
I mean, heavy involvement there.
So I think that as you call it out, you know, getting that right data at the
right time
to the model is so critical.
And it's not being done well enough.
I know what you said of just showing two knowledge articles is actually
probably going to delay
the interaction that somebody screens through it versus getting right to that
search point.
And I think at AWS, and I am in the AWS startup's org, as we're seeing this
explosion in interest
here, what has really caught my eye about support logic, you know, besides the
connect
integration, and I think we've all seen kind of voice as that next frontier
kind of tied
to this.
So, is we're seeing just an explosion in RLE-HF companies.
So, you know, reinforcement learning with human feedback.
Guess what?
The humans are among the most important, if not the most important aspect of it
And companies that can provide RLE-HF, I mean, we've seen several that have
gone from,
call it, 5 million in revenue, 28 months ago, to over 100 million in ARR today
because they're
providing that tremendous service to verify, you know, hey, is the model
getting the right
point.
And what fascinates me about what support logic is doing, it's a bit of a
stretch and
obviously I don't know exactly what's on the roadmap.
But if you start thinking about your support agents is actually verifying, you
know, what
the model is bringing you in terms of a score.
You take those scores and actually start feeding that into some of your model
training.
I think that's a fascinating idea and I think it starts turning to are we
thinking about
our support folks as a cost center or as a real investment to train this
incredibly
valuable resource for the future.
So, that's part of what I'm quite excited about here and I don't have the depth
in terms
of support specifically that you all have but I think that's a very interesting
, you
know, pathway to explore.
And thank you for bringing up the RLE-HF because that was something I wanted to
ask about.
So, a lot of buzz about that right now.
But I should have a question for you.
Uber has a really unique set of customer profiles.
So you're dealing with drivers and merchants and customers.
So how is AI helping you really build the trust across all of these very
diverse contingents?
And especially, you know, high-stake interactions, a lot of emotion.
I can't tell you how many times I've left my cell phone in Uber and call from a
landline
panicked about it.
So how is AI helping with this?
Yeah, yeah.
Now this question, I think when it comes to the element of trust, I have a
favorite story
to tell it from my childhood.
Growing up, there was a chewing gum that was available.
I didn't particularly like the taste of it.
Moreover, the rapper included some statistics for the game of cricket and if
you read certain
number, you were in for a prize.
The shopkeeper there, he saw me day after day, you know, buying the chewing gum
, throwing
it in the trash, just collecting the rapper.
And after a few days, he's like, you know, what's going on here?
And I said, you know, I'm in just for that prize.
So he said, you know, let me save you all the trouble.
You know, here are the three options for the prize and you can select it.
You know, I was a, you know, a happy kid.
I think this, and then I happened to visit that shopkeeper a few years ago.
The area has significantly, you know, developed, gentrified, a lot of big malls
and, you know,
other fancy shops around.
That shop still exists with, you know, dedicated, loyal client base.
But that person did for that moment is identified pretty much proactively what
me as a customer,
I was looking into for the entire process and delivered that outcome saving me
money,
but also buying my loyalty.
Now, going back to the trust equation, you know, I think what customers are
looking for
is, you know, we talked about having your back, but they're also looking for
accountability.
When something goes wrong, are you taking ownership of the problem?
It doesn't matter where the issues, maybe I didn't follow your requirements
properly.
Maybe you know, something else happened in the physical world in Uber's case.
Are you providing consistency?
Meaning I will run into support issues from time to time.
Are you going to delight me every single time or is it a one-off thing?
And I think that's where the RLHF bit is also important.
How do you take experience from the most, you know, expert agents and trans,
using AI,
do it, make it accessible to every single person.
And so now, you know, how is AI helping?
You know, it starts with understanding.
You know, LLMs, they are out of the box, they are extremely good at
understanding pretty
much anything you say.
Now, of course, you know, previous generation of NAB models, you had to train
the, you know,
let's say, Google dialogue flow family of models, again, specific sort of
intense.
If it falls out of your training reporter, it may just say, I can't help you
and it was
very limiting that way.
Now, the LLM, now you have an opportunity to at least understand what the
customers are
saying.
After you understand, I think that's when you need to have a set of policies,
business
policies to respond to the intent in empathetic manner.
While you are resolving the issue, making sure you are communicating to the
customer that
throughout the course, you have their back.
And I will give you a very simple example of before and after of LLM.
Your food is running like very common defect for Uber Eats.
If you ask for the previous generation of NLP models when you train against
this intent,
they will say, well, you know, thanks for letting us know, we all make sure, we
all
to inform you that your food is still on track and here is a link you can track
lives.
Five minutes later, the same customer comes, ask, we'll say the same thing.
Now, this is, this has been the chief complaint against all the chatbots, no
matter who has
implemented them, it's too robotic.
Now, modern LLMs, they are quite nuanced.
They know that the customer had asked for this problem before.
The fact that the customer is bringing the same problem to me again, that means
anxiety
level is high.
You are talking about nuances of emotions.
So, it will respond, it will keep the facts the same, which is, you know, here
is the
driver, here is where you can track it, but explicitly acknowledge the
elevation in the
anxiety the customer has.
And the third element, which is out of somewhat unique to Uber, but I think
applicable to
support ecosystem in general.
A lot of time when, you know, financial transactions are involved, you know, we
have rider, we have
delivery partner, we have restaurants and we are trying to run three-sided
market place
in case of Uber, it's two-sided market place in terms of mobility.
While also keeping Uber's business interest in mind, it's impossible to
sometimes to
do what individual person has perceived right by them.
Driver and a delivery partner may want more share of the total revenue,
merchant may want
them more share of the total revenue.
And that's why explaining your decision making through LLM, that is
personalized, customized
to those personers will help them understand the effort that you are making to
do right
by them.
And if you do this consistently enough, they will believe you that you have
that back no
matter what.
So my next question is for Sanjid, these AI models are becoming much more
sophisticated.
We're looking at automating and streamlining, agent activity to boost
productivity, but
how do we do this and keep the customer experience in mind?
Yeah, I think we have to stick to the human in the loop approach, this over-rot
ating
on thinking that we'll be able to take X part of the business and just put it
on the digital
AI and supported kind of, I don't think that helps.
I think we need to keep human in the loop, we need to make sure that we are
doing this
continuous feedback and improvement mechanisms.
We need to also cut this narrative of that, you know, the cost cutting bit of
replacing
humans because we are relying on the same set of people to adopt these
technologies and
then posing this as a threat to their existence.
And I would say that we need to craft off mechanisms and workflows which
utilized, you
know, this digital ways of doing things, but always keeping the human empathy,
human interactions
in mind and not over-rotate on just making everything digital.
Because that model hasn't worked before, that model won't work.
There will be a lot of efficiencies that we're going to gain to be able to
invest more in
that high value conversation with our in human form as well.
We need to keep that balance.
So if support executives are being incented to cut costs, who owns that
customer experience?
Is that the CCO's job?
To make sure that you're maintaining an amazing experience?
I think we all have to take that responsibility.
I think it's a support leader's job to make sure that we are well-worzed and
have conviction
about our context.
We know what's good for business.
And I think it's a support leader's job also to make sure that we are
converting this cost
functions to revenue functions, increasing revenue, bringing more scalability.
And I think the bandwidth that we want to create is also to go and work cross-
functionally
to take these product and support interface and insights that we have to design
better
experiences.
So I don't think it's just a CCO's job.
It's a support leader's job as well.
And that's why what is it that we sell as ROI when we think about technologies
and solutions
like this?
Do we go and sell that?
Oh, we're going to have 30% cost reduction?
Or are we actually talking about the pain?
So the narrative around the pain has to be not just anecdotal.
It has to be data driven.
It has to also tell the impact it's having on our business, the negative impact
on the
business as well as customers.
And now it's closely tied to the retention and the expansion side of business.
Yeah.
And topic of whose job it is, I think Amazon certainly has the leadership
principle customer
obsession.
It is everybody's job.
And I think it has to start from the CEO.
Darai is a great story on this topic when he was running Expedia.
A lot of customers, they would buy flight tickets well in advance and they
would forget.
And they would be anxiety.
They would reach out to the customer support.
Now here was the problem of keeping the job limited to the support function.
There are so many inquiries coming in to have.
When is my flight, et cetera, of that nature?
Because the responsibility lied only with that support function, they solved it
by hiring
more agents.
Outcome was, we never addressed the anxiety party.
It just as the Expedia business scale, they had to scale the number of agents
to solve
this repetitive inquiry.
When Darai looked at the problem, he said, wait a minute.
Now why don't we apply tech to it and just send customers reminders, you know,
a couple
of days before, hey, I hope you are excited for your flight to New Mexico or
whatever.
And guess what?
First of all, it cut to the root cause of the problem where people may have
forgotten
that they had an upcoming flight.
And they were able to reduce the support cost at the same time.
This is why I think the teams will need to work together.
You know, product development team cannot think of supporters and afterthought.
Supporting the same way, they should not be in the business of solving the
repetitive defects
day after day without bringing back the product experience and fixing it on day
zero.
So Tamara, I want to ask you a data question.
So enterprises have oceans of data, AIs generating even more data.
So where do you think we are headed, particularly with unstructured data and
the ability to accurately
query with a generative AI solution that can really get at the meat of both the
structured
and the unstructured?
I mean, how do we avoid the garbage and garbage out problem?
Yeah, a great point.
And I think that ties back to points that both Babesh and Sanjid made as like
part of
it is we're dealing with with SaaS sprawl, right?
We've got SaaS sprawl.
So everyone's taking in so much data.
I still think 90% of that is going to come in unstructured.
So where are we headed?
I think dealing with that, I think LLMs have been incredibly good tools at
helping to make,
you know, get some signal from that noise of unstructured data.
But where we're headed that could be dangerous is a few things.
The data one, you know, I think an LLM is a university professor and you don't
need a
university professor in every engagement, right?
You can take a better trained model and potentially directed to an SLM, a small
language model
or open source because in that particular case, you know, a mechanic
interaction or
one you point to that.
So being able to refine that data down to train the model in a specific point
is very
important for managing costs and actually providing, you know, the best
interaction.
We've seen small language models really outperform large language models in
many cases.
I love the mention of Amazon Bedrock in that.
I mean, what we're trying to do is make, you know, model choice accessible as
broadly
as possible so that testing being continuous.
And then I think from beyond that, I love what Sanjid said about RAG and Retri
eval
at the LLM, right?
So, again, coming from a place like Glean, which is an ANIVA, so I have a bias
towards
search, I have a bias towards RAG, you know, we were able to do much of this
without doing
inference, right?
There's no reason to throw this very heavy compute at inference and that.
So that kind of all ties back to how do we make sense of this incredible mass
of unstructured
data, 90% of the world's data being unstructured.
And I think by being really smart along these lines and I think support logic
is really
can be a really key cog in that stuff that's going to help us get to the right
data at
the right time and make this, you know, Gini a profitable more quickly, right?
Now this is huge expense line.
So those are the key points that I would call out.
>> So could you quickly talk about this agentic AI that everybody's so excited
about at the
moment?
I mean, I don't want to run long here, but just what's the potential here?
Why are people so excited about it?
I'm sure you've all got thoughts, but can we start with you?
>> Yeah, maybe we can do a lightning round on it because I'm sure we could do a
whole
separate thing.
>> I think I'm looking at you for expertise.
>> Yeah.
Well, I think with agents, again, a voice, I believe voice can be that next
frontier.
It is perhaps the first frontier.
So again, the connect and we are all in on connect and also I mentioned our G
ini I accelerator
that we just launched and my team helped pick the North American cohort for the
startups
and we picked five out of the 20 are voice focused.
That's how much we believe voice is being that key generation.
That being said, I would also say that it's very early.
We are seeing wins in agentic, but it is quite early in terms of that happening
So in terms of the, you know, you take your calendly, very simple, you know, is
that triggered
by anything natural language or models?
>> Yeah.
>> I think agentic is one of those elements to make this work as well from a
profitability
standpoint that we'll have to have.
>> Yeah.
>> You've got a gist.
>> Okay.
>> I'll answer it in, you know, I'll add to it in from two angles.
One is, you know, sort of a simple example of agentic AI that we can all, we
could all
benefit from in future.
Now, today whoever is interacted with travel concierge from time to time, they
would help
you plan the travel itinerary, but you know, a lot of times you have to tell
them who you
are and what you like and your preferences including budget.
And they would fall short up until the point that, well, here is a 10-day
itinerary to
this place, but after that, you know, hotel booking flight, everything you need
to do
on your own.
Now, come agentic world, first of all, they will know who you are.
Just with few phrases, you'll have an itinerary for you to review.
If you're happy with it, they will go ahead and make the bookings.
Now this is the promise and of course in support runs, this is greatly
beneficial as well
because to answer or resolve support tickets, you'll need to access multiple
systems, you
need to read a bunch of information, you'll have to execute bunch of work flow
sizes,
sending the email, issuing refund, etc.
When you can trust AI to do all of that, all of this, you know, like end user
experience
can become extremely delightful.
One of the things that doesn't get talked about enough is imagine today,
imagine today
AI has all this capability to read any information from your private knowledge
base or public
knowledge base, do research, take actions on your behalf.
Are you really comfortable with it?
And it may be a personal choice, it may be a business choice.
So I think what we will need to come up with is sort of a risk modeling or risk
framework
about what is it that we, we as a business feel comfortable with, AI taking
actions on
our customers behalf versus not and that risk profile will need to be reviewed,
pretty
much on continuous basis as you build your own confidence into AI as your
customers
start demanding more of you.
But it's definitely the part of the future for when it comes to AI, I'm sure,
Amazon,
Uber, all of us have invested quite heavily into agent decay.
I think I don't know if you're seeing the same, but for me with our, we do have
bedrock
agents and other points of that.
Probably the most common scenarios we're seeing for agents so far is opening a
support ticket
when the interaction is going well.
So that's in step one.
And then interestingly, because I think we all do believe in the human
interaction, but
from the security standpoint, though, I think we're also trying to solve for
the, you know,
there are over one almost security flags that can happen.
So having agents who we trust to start doing some prioritization on that, that
's very
early.
But I think those are some of the areas where we see the biggest upside in the
near term.
Yeah.
So we are almost out of time.
We've got that dangerous before lunch spot.
We have a hungry audience.
So I would like to ask each of you to take one minute and give advice to people
who are
still early in their AI or gen AI journey.
What can they do to get started?
What can they do to make sure they're getting the most value out of their
investment?
So one quick take from each of you, Sanji.
So I would say start down.
Don't delay this.
You're already three to four, maybe five years late.
I think find the pain, find the real pain, find the right use case and get
started with
it.
Remove the friction points and focus on that and stay competitive in terms of
cost.
And just just realize the value of that particular use case before you get on
to the next.
But start now.
Great advice.
Yeah.
Pervash.
I would expand on that by saying just increase your innovation funnel.
One of the things we have seen at Uber and a few other companies that I'm
interacting
with worked really, really well is Hackathons.
And you won't believe the kind of ideas, kind of challenges people are applying
AI to and
finding benefits.
It's almost impossible I would say to dictate top down.
This is where you will be using AI and this is where you will be getting ROIs.
Increase your innovation funnel as much as you can.
And then create a culture of collaboration because you know this idea has come
from data
scientists, this product managers, engineers of course, how them work together,
create
pods and have sort of early success matrix in mind so that you know whether to
pull
the plug or double down on this.
And the third and most critical factor is also out-executive buying.
The few startups I'm involved in and I have seen that wherever executives have
personally
leaned into generative AI either by coding themselves using GitHub co-pilot, or
cursor
with cloud etcetera and or building some of this use case prototype themselves.
They have a much more grounded perspective on where AI stood and what it can do
for
your business.
Versus leaders who have just maybe read news about what AI can do and PR can
often be
more pompous or misleading.
Those projects are you know I would say internally working with leaders who are
personally leaned
in can give you much more rope, can give you much more investment and can
create a culture
and framework for you to succeed.
>> Jim, are you get the closing word?
>> The right port launch.
So I would say a kind of combo answer with generative AI, we're talking AI
specifically
I'd say stay horizontal for now.
One thing I love about support logic is the horizontal solution that can
transform your
business and what we're seeing at AWS and beyond is that as much as people are
saying
hey verticalize verticalize verticalize it's too early and genetic.
Do something that knocks out a big problem and then build off of that to look
at your
gen AI strategy and I like support logic for that is almost a platform to
attack broadly.
We saw the same thing when I was a gleaned, a rocket based on horizontal
knowledge management
and then combo with that I'd say take the start of mentality and eat your own
dog food.
Like do use this.
If you're not using gen AI yourself daily you're not going to have the
experience to
see how it's impacting your customers and getting your own tools and it's hard
to believe
it's only been 20 months but the day to day usage has just blown my mind and
has transformed
my work life in the last 20 months more so than I can other than mobile.
So those would be my point.
My final thought would be it takes the village to build a successful AI project
and we've
obviously got brilliant people here with a lot of ideas to share and I know all
of you
in the audience probably know at least one thing about AI that no one else
knows so I
would really encourage you while you're here to network as much as possible ask
each other
questions challenge each other.
You know we are the people in this room that are going to raise the support
organization
up to become a strategic sales.
This strategic is marketing as strategic as product.
So let's all work together and hopefully next year we can all get together and
talk
about our amazing successes.
So thank you all very much for joining me.
Thanks for having us today and have an amazing lunch.
[BLANK_AUDIO]