Generative AI for Customer Experience

hello everyone and uh welcome to today's
webinar on generative AI for CX or
customer experience I'm Michelle Gupta
VP of product at natomi
in today's webinar we will be covering
generally about GPT generative Ai and
llms I'll give you a quick brief
background on these these terms uh we'll
talk about how generative AI can be
applied to customer experience we'll
also talk about how generative Ai and
conversational AI can work in tandem for
customer experience I'll also share some
of our observations and we'll wrap up
with uh questions if there are any so
again thank you for joining and let's
get into it
so many of you may have noticed uh chat
GPT appeared on the cover of Times
magazine recently this goes to highlight
you know how popular this has become
very quickly I believe in a matter of
about five days there are like a million
or so users already using chat GPD it
also highlights uh the potential
importance this technology generative AI
in general could be playing in our lives
across different applications
for some basic context uh chat GPD comes
from the same family of gpdei models
from open Ai and these models are built
using massive amounts of of knowledge
and data and their purpose built to
generate human-like text so with this
type of capability lends itself nicely
to things like text summarization
language translation and then of course
generating texts for applications like
chat Bots
broadly speaking these Technologies fall
under the umbrella of generative AI so
you know by its very meaning generative
means it can actually produce content
it's a set of AI algorithms that can be
used to create actually create new
content so this new content could be
video audio actually even write code
text and even generate art as you've
probably seen things like Dolly and
stable diffusion do if you've caught up
with the news recently and what's
interesting about generative AI is that
it's it's almost mimics how human beings
would create and produce content and
that's what makes this so fascinating uh
chat GPT in particular is an example of
text generation AI focusing specifically
on conversational tasks
and uh just to step back also if you
look at the landscape it's gotten pretty
uh pretty busy with a lot of
applications out there leveraging
generally generated AI for lots of lots
of applications uh conversational uh
experiences in the context of customer
experience uh is at least in our opinion
going to be a massive application of
this technology as well so we expect to
see more application of generative AI in
in customer experience and a lot more
applications coming out as well
now what is behind uh the generative AI
there's the the ultimate source of this
information is what is called a large
language model or llm for short and
these are deep learning algorithms that
have been built using again massive
amounts of data so think you know
feeding it almost the entire internet if
you will and using this this knowledge
these models uh then learn and they can
recognize summarize translate predict
and generate text based on on this
knowledge uh just to give you an idea of
the order of magnitude gpt3 which is the
model from open AI was released in I
believe June of 2020 that has over 175
billion parameters so massive data sets
and again the the interesting thing here
is that it can actually using the
knowledge it has sort of gained sort of
like a human being produce almost I
would say original new content
um
in order to interact with an llm you
basically provided an input or a prompt
which is basically you know asking it a
question and the llm then using its
knowledge that it's gathered from from
all of the data that it has understood
and extracted its Knowledge from it
generates the the output so it's
basically a a prompt to an output
interaction
so you know what this text generation
capabilities
um a customer experience scenario
usually involves a lot of text
interactions going back and forth so an
application of generative Ai and
customer experience would seem like a
very obvious use of this technology and
we'll talk about how that can be done
there are also certain shortcomings that
we need to be aware of as we do consider
using generative EI for customer
experience so uh let's walk through what
some of those shortcomings are
so firstly these large language models
are more General built using the
available or publicly available
information so they're not specific to a
business in in this situation you know
it's hard for a general Standalone large
language model to to do anything that
would be specific for uh for a business
so it certainly has those limitations
secondly it's not connected to your
business systems so doing things like
completing transactions or doing
end-to-end resolutions for customer
service requests those really can't be
accomplished
and then it's not really plugged into
the interfaces that your customers are
using to communicate with you so be it
messaging channels email voice whatever
that might be they're not connected to
to the front-end interfaces uh where
your customers uh would be reaching out
to you and then lastly there's no real
collaboration with human agents so
there's no way for for generated AI to
just escalate to a human agents now with
all of these shortcomings the
counterpart of generative AI which is
conversational AI
addresses some of these in in very
specific specific ways so the the models
that are built using conversational AI
platforms are are specific for a
particular business you have
well-defined process and flows that get
customers to specific outcomes they are
connected and they can be connected to
your business systems so you have the
opportunity to drive contextual
conversations do full resolutions and
and complete transactions um like order
bookings flight bookings and things like
that
and then you can of course plug
conversational AI platforms into the
front-end channels where your customers
are interacting so again be an email
chat voice whatever that might be you
can plug in these systems into your
contact centers and other channels to
actually have ai participate um in these
conversations
and then lastly conversational AI
platforms give you the opportunity to
escalate to human agents who are needed
so the collaboration with agents is
there the ability to do agent Handover
and agent augmentation are things that
the conversational EI platforms are able
to provide
and if you look at the business outcomes
that need to be delivered that's
specifically what conversational EI
platforms are purpose built for right so
in be it the front-end channels
integrating with the systems using
well-defined user Journeys driven by
company policy in in being on brand the
quality of the conversation or the
specifics of that conversation all of
the necessary guardrails can be put in
place so your conversational AI actually
delivers the specific business outcomes
that you're looking for so you've got
you know the the fluid generative AI on
one side and you've got the more
structured conversational AI on the
other side so it would it would make
sense to have these two technologies
work in tandem to produce a net benefit
in the customer experience Journey
so let's look at a few ways in which
these two technologies can work together
for a net positive impact on customer
experience so one area is just around
acceleration so a lot of things that
conversational AI platforms require is
you know training the AI model providing
it input on different uh intent
sentences
lexicon entries or entities building
building user flows building
conversation uh building sample
conversations for simulation these are
things that generative AI can now do
very easily given the vast amount of
data it has so from an acceleration
standpoint this is a nice addition to a
conversational platform where it can
utilize or leverage generative AI for
for tasks like this so it it certainly
shortens the the time needed for for
these types of tasks as you're as you're
building virtual agents secondly you can
and leverage leverage con generative AI
for massaging responses that are being
sent out to your users you can also
instead of just relying on static or
hard-coded responses introduce a level
of dynamic uh response creation so this
helps virtual agents be you know less
robotic and it certainly introduces a a
more natural conversation that AI
virtual agents are able to have uh with
with your customers and then I would say
with structured knowledge basis if you
feed structured and well-defined
knowledge bases into generative AI you
can actually extract better quality
answers that can be served up to
customers so you end up building better
end interactions using generative AI
alongside conversational AI and then
lastly on the agent assistance side
it's a good use to rely on generative AI
to do things like sentiment and Analysis
where you've got agents that have an
easier access to how the sentiment for a
particular conversation may be
progressing also getting suggested
replies from uh from the generative AI
so they don't have to spend too much
time drafting or writing responses
themselves and then also getting summary
on point summary for long and verbose
conversations so they can pinpoint
specific areas that they that they need
to focus on
uh just using one example of how uh
conversational Ai and general Avi could
work together is is in cases in use
cases where there is uh sensitive data
involved so you know for example if a
customer is going to be providing Social
Security numbers credit card numbers
date of births you certainly don't want
to feed that directly into a generative
AI platform now if you have the
conversational AI platform what it
initially does is it removes sensitive
information uh and it creates what's
called like a sanitized prompt right and
alongside the sanitized prompt you can
add layers of you know the intent
recognition any of the conversational
context and the sanitized prompt can
then be fed into the generative AI to
actually produce you know a a
well-crafted natural sounding response
so you've handled kind of like the
Privacy side and security side with
conversational Ai and then you've
leveraged the generative ai's capability
to print use good quality natural
sounding conversation so it's a good way
to see these two technologies work
together in this fashion
so uh to wrap up then you know share
some observations uh that we have
um in this space and in the potential
impacts on on customer experience so
firstly I think uh you know anyone who's
interacted with chat GPD has been I
would say pleasantly surprised at the
quality of the conversation so we used
to look at AI Bots as being very robotic
monotonous in how they would have a
conversation but now chat GPD has
demonstrated that generally AI can
produce very natural conversation so
this this has elevated the expectations
for for everyone in general and what we
expect from automated AI virtual agents
is going to be those expectations are
going to be a lot higher in in the the I
would say the good thing here is that
the effort now needed to produce these
higher quality conversations is fairly
low so you know basically there would be
no excuse for for anyone to put reduce
unnatural or robotics sounding um AI
driven conversations when this can be
done fairly easily
uh secondly you know there's a lot of
entertainment value that we've we've
seen through these uh chat GPT type of
conversations but business outcomes are
always going to Trump the entertainment
value so again Standalone general
purpose uh llms that are not uh taking
into account uh business context they
won't find themselves in in terms of
being able to produce any productive
value so contextually correct generative
AI is is truly going to be game changing
um and then the third area is the agent
assistance this behind the scenes
productivity
behind the scenes activity that that can
drive a lot of productivity we feel is
going to be a significant area where a
lot of I would say a lot of good
Innovation and productive Innovation can
be done the interesting thing about this
is that because it is happening behind
the scenes you always have kind of like
a human in the loop that is acting as
the guard rail
so anytime there's uh suggestions being
offered by generative AI summarizations
being have being offered by the
generative AI there is a human being
that is looking at these
um and then deciding what to do next um
so again I think the behind the scenes
activity a lot of times we we don't
recognize the value that's being
generated there but uh this is a big big
area where we see a lot of opportunity
to do uh to do good work and and
introduce efficiencies for customer
experience
um and then lastly uh you know accuracy
and security is going to be very
important so a lot of the open domain
systems that are out there
um and a lot of folks have been using
them for a variety of purposes you know
one has to recognize that these models
have been fed a lot of data that is
generally and publicly available it is
full of biases it is full of
inaccuracies and that introduces risks
if you're attempting to use these
directly so as a standalone
um these these uh open open domain uh
llms are are not not I would say
recommended to be used as is you may
have read about some of the challenges
uh the the largest software or search
companies faced after they introduce
some of the generative AI into their
search algorithms
um it went sideways on a few occasions
so it makes it even more important to
make sure to make sure that the controls
and the guardrails are in place if you
are going to attempt to use these as a
part of your customer experience Journey
so with that uh that wraps up the
material that I was wanting to share and
present I hope you found this
information useful
um I'll just see if there are any
questions from the audience
when would it make sense to just use
generative EI
um as is in terms of having direct
conversations with with end users uh
so the short answer is that if there are
scenarios where you do not expect a or
or the risk or the impact to your brand
is minimal and this you have to assess
on a case-by-case basis then certainly
you could consider using it for what I
would call The Last Mile conversation
all right so between the platform and
the end user so it could be just general
Greetings or basic you know um basic
information or exchange but anytime you
perceive uh the risk is going to be much
higher where generative AI could
potentially provide information or
respond in a way that is not on topic
not on brand you certainly don't want to
leave generative AI on its own to
interact with your with your customers
directly
uh there is one other question around
just the agent productivity piece
uh just asking you know what what are
the areas where we're seeing the biggest
uptake for for generative AI
um so the I would say two two areas or
say three areas uh one is uh
summarization so in scenarios where the
AI is having a conversation uh initially
and now the AI has to hand off the
conversation to a human agent uh
creating a summary using generative AI
that focuses on you know specific points
of the conversation that the agent can
then reference very quickly that's one
use case we're seeing a lot of interest
in that we're also incorporating into
our product uh the other area is uh just
in paraphrasing or rephrasing some of
the responses as the agents are are
typing uh the response so they could put
in a short sentence and the generative I
can can produce a better sounding reply
um so that's that's the other area area
that we're seeing the assistance piece
come in and then also the third area is
just the you know out the gate suggested
reply so even before an agent types up a
response just based on the conversation
context uh the generative AI is able to
offer a few suggested reply options that
the agent can then pick and then quickly
modify and then reply back to the user
so thank you again uh I believe those
were the questions that we we had and I
hope this was helpful I certainly look
forward to sharing more information as
this technology continues to evolve and
emerge and we're excited to uh to be
offering some of these generative AI
capabilities in the new in the very near
future in our product as well thank you
for joining