How Generative AI Improves Support Experience
You think you know AI, but do you understand how generative AI can redefine your support organization? In this session, we’ll delve deep into advanced AI functions like case summarization and language translation, challenging you to rethink how you manage and interpret support data. Expect to walk away with a new perspective on what AI can truly accomplish.
You Might Also Like
My audible log there.
Hello, everybody.
Welcome back from the break.
We're going to be starting with our next session.
Today's quick introduction.
My name is Kratika Manohar, senior director
of product at Support Logic.
So I'm going to be walking you through the Gen AI-powered
workflows.
A little bit more in detail.
We saw an overview of a lot of those capabilities
during the keynote session from current this morning.
I'm going to be double clicking on all of those workflows.
And talking to a little bit more in detail,
we'll go through the demos and some under the hood details
as well.
Now, just a year ago, more than 50% of the C-suite
were the main blockers when it came to Gen AI adoption.
They did not really trust the maturity of the technology.
We just fast forward a year.
And now we see two in three executives really
think that Gen AI is going to be the next disruptive technology.
And 33% of them are already investing in it.
Now, if we look at the economic impact
of the different functions, customer support
and operations is third in line.
Now, there's significant potential to be unlocked.
There's about $400 billion worth of value
that we are seeing just from customer support and operations.
And that's because we are not only
able to increase the issue resolution,
but we're also increasing the productivity as well.
So there's an uplift as well as an increase
in the issue resolution with time gain.
So as a result, there's a huge impact and a huge interest
when it comes to Gen AI workflows in customer support
and operations.
Now, as far as our support logic journey goes,
you saw that summarization engine and answer engine
are some of the recent investments
we have been making in 2023 and 2024.
Summarization in general--
if we-- in general terms, LLM summarization
is you all know it takes a large document.
It could be a PDF.
It could be a knowledge-based article.
It takes it condensities and gives you a concise summary.
Now, when it comes to the CX world,
we are talking about agents call summary, a chat summary.
It could be a summary of an account
or it could be summary of an agent or a product.
There's a lot of potential over here.
Now, we have built a summarization engine
in which we have a very robust case summarization,
account summarization, and knowledge summarization
capabilities in the product.
Now, let's start with case summarization.
Now, at a surface, case summarization may seem very simple.
There's quite a bit of nuances that come with case summarization.
One, in general, when we talk about complex cases,
there's a lot of information that's involved.
There's a lot of technical details in the information.
And there's a lot of back and forth involved.
Being able to take all that information
and create a concise summary while still maintaining
the context is a complex challenge.
And striking the right balance becomes critical here.
We don't want to end up creating a very, very brief summary,
which is not useful to anyone.
On the other hand, we don't want to have a long cumbersome
summary, which nobody's going to read.
Another challenge is that these cases are long running
and they're evolving over a period of time.
So what started off as problem A may not be problem A.
It could have transformed into problem B.
So being able to carry that long running context
and maintaining the recency when we are creating the summary
is another tricky challenge.
Thirdly, there's going to be a lot of sentiments,
fluctuating sentiments as the case progresses.
There could be ups and downs, which we also
want to be able to accurately capture
as part of the summary.
And last is the sensitive information
that's there in the case.
It could be just four numbers, emails.
We want to be able to redact or mask that information
while creating the summary, but at the same time,
give a complete picture to the users.
Now, it's a port logic we have been
able to solve for each one of these challenges
with our enhanced case summarization.
Now let's take a look at the demo.
We're going to be starting with case summarization
from the context of the support manager.
Now, we briefly looked at this this morning.
So support manager, for me, support logic already
gives me the right cases.
It helps me be proactive by identifying the right cases.
Could it be a LTE case?
It could be a needs attention, negative case.
I don't want to be running through the entire case details,
especially if it has 40, 50 conversations.
I don't want to be reading through all of that.
Time is a sense for me.
So seeing this case summary is extremely helpful.
And what's more important is it gives me
the latest problem.
Like I mentioned earlier, the problem could have started as A.
And after a lot of troubleshooting,
it could have been discovered that there's
another underlying problem, Brie.
So our case summary is able to understand
where the exact problem is at the given point in time
and be able to give it to me as a support manager.
And two, we don't want to have the lost in the middle problem
where we are looking at the early part of the case
or looking at the conversations that
happened during the early stages.
We don't want to lose track of it,
but at the same time, we want to give more
weightage to the most recent actions.
So that current status is built to be able to remember
the context and still give you the latest current status
of the case.
And last, we have the planned actions.
So these are the actions that the agent has been--
based on the agent's conversations with the other agents
or the customer, there are some deducted next steps
that we have identified as part of the summary.
So as a support manager, this is extremely helpful for me
because now I know exactly what to do next.
I can bring in the right people, bring in a swarming team.
So it gives me the right actionable insights
to move the case forward.
Now, let's look at it from an agent's perspective.
Now, if I'm a support agent, I could be starting off
with a new case.
I could be starting with a case that somebody handed a word
to me, or I could be just picking up a case
that I was working on a week back,
it's just a long running case.
In either case, being able to orient myself to the case
and understand what the problem is,
understand where the current status is,
and knowing what has happened with the case
helps me jump into action right away.
Now, we talked about the complexities of case summarization.
We're seeing what is available right now,
but there's a lot more nuances that go behind case
summarization, kind of if we can switch to the--
all right.
So one is being channel aware, especially
with omni-channel cases.
A case could have come in as a web case,
and then there could have been, over into the complexity
of the case, the agent could have called the customer
multiple times, there could be blocks of wise conversations
inside a case, there could be a case which
started off as a chat case, and then transformed
into a wise case.
So it's possible that we have a case which
has a wise conversation blocks, chat conversation blocks,
and just general back and forth information
that the agent might have had with other employees
in the organization.
So being able to understand those units of conversation,
summarizing that, summarizing multiple of those
in the context of the entire case,
and then presenting the latest status on the next steps,
is extremely important for us to get a holistic view.
So we need to be absolutely channel aware.
And two is being context aware.
When I say context, we're talking about what
is the status of the case.
Now, is it-- maybe it's likely to escalate case
in which case the kind of workflows that I want to take?
And what kind of summary I want to produce,
where I want to present it, am I going to send it
as part of the escalation review,
am I going to present it as part of the alert?
So there's a lot of nuances that one could look for when we're
depending on the context of the case itself.
So two, we need to be context aware.
And three is being persona or role aware.
For example, as an escalation manager,
the kind of summary that I'm looking for,
will be very focused on, OK, why did this escalation happen?
What exactly did not work?
And what do I need to make it work?
Versus, if I'm a QA manager, the kind of summarization
that I'm looking for is going to be centered around the Asian
performance.
I need to know what kind of signals or behaviors went wrong.
Likewise, if I'm an executive, I'm
looking at a very high level summary.
I don't want to get into the nitty-gritty details.
So we need to be persona aware to be
able to generate the right summarization
so that we're giving the right insight to the right people.
Now, we have built the right level of configurability
under the hood to be able to present the right summarization
that is channel aware, context aware, and persona aware.
Now, under the hoods, we use anthropic cloud for case
summarization.
However, we do have a robust ML as a service that
is able to seamlessly work with multiple MLM models
and multiple LMM vendors.
And they are constantly evaluating models
for their performance, for scalability, for cost.
And also, depending upon the context channel or the persona,
the kind of prompts that we want to have
or the kind of LMM that would fit best for it
is going to change.
So on the one hand, we are constantly assessing it
because we have the flexibility to do it seamlessly.
And two, we are also able to do it in a configurable way.
So depending on the input or the need,
we are able to use different models.
Now, again, as a manager, getting a summary of one case
is great, but what would be even more useful
is if I can get a summary of a cohort of cases.
And when I say a cohort of cases,
it could be a group of cases tied to a certain topic
that's trending.
There could be a certain adapter failure,
some topic that is trending right now.
I want to see all the cases that are facing
that particular issue.
And I want to group them all together
and get a summary of that case.
Or it could be a team.
A team could be not performing very well.
I want to look at all the cases that are being managed
by that team.
And I want to get a summary of those cases.
Now, what we're doing here is trying
to look at the common patterns and underlying issues
to identify those common issues as well as emerging issues
and key signals and patterns that we are able to detect
using our ML engine.
Now, what we also want to do-- this is not something
that we have today-- but is based on the common and emerging
issues and key signals and patterns
to be able to detect the next best actions
and then automatically perform these actions.
Now, what you're seeing here is a generic way
of being able to summarize a cohort of cases.
Now, what we-- what Keren talked about earlier today
is account summarization, which is live in the product today.
Let's look at a demo.
It's doing exactly the same thing,
but just in the context of an account.
So given an account, we look at all the cases
that are tied to the account.
And we are able to generate a summary for that particular
account.
Now, here as account manager, there
are a couple of insights that I get just by looking
at the account summary.
Now, there's three sections here.
Now, the current status is really helping
me focus on those issues that are current burning
and something that we need to really pay attention to.
For things like there's active escalations,
there are production issues going on,
or there is business impacting high priority cases
that we need to really work on.
And then we have signals and risk indicators,
which kind of tell me, give me the warning signs such as,
there's a churn risk, or maybe there's
a lot of frustration signals or urgency signals,
or maybe there's confusion.
So there's a lot of signals that are elevated over here
at an account level.
So I know exactly what is going on with this account
so I can act up top directly.
And third is the overall trends and issues.
And this is where we are able to un-ert the overall common
patterns.
Again, these are not things which I may be able to resolve right
away.
However, by resolving this, we'll be
able to fix the health of the account as a whole.
Now, let's look at a little bit under the hood
on what exactly we do to get the account summarization.
So as I mentioned earlier, today the account summarization
is based on all the support cases that are tied to it.
Now, we feed a lot of facts about the case
to the summarization engine for it
to be able to generate all the current issues, key signals,
and overall trends that we discussed earlier.
Now, what are the facts that we feed it with?
The case trend, the signals that are detected on the cases,
the problems, the issue tags that are detected on the case,
and other factors like, what is the customer effort typically
involved when dealing with cases of this particular customer
and so on.
Now, what's coming next to what we're working on
is being able to do the same for account level interactions,
not just supposed interactions, but also
look at all your customer success interactions,
onboarding interactions, et cetera,
that are tied directly to the account,
and be able to take the facts from those interactions
and, again, feed into the summarization engine.
So we are able to generate a holistic summary for the account.
Now, one, just one, yes.
Similar to the cohort of cases, again, what will be really useful
here is to be able to look at the overall trends and patterns
that we already have, and then determine
what the next best actions are, and subsequently, what to be
even great is to be able to automatically
execute those next best actions.
And another important element here is to be able to look
at patterns across the accounts, while we are summarizing
a single account, just like how summarization of a cohort
of cases was useful, summarizing a group of accounts for us
to be able to determine the patterns across accounts
is another area that we will be looking into.
So now, moving on to the answer engine.
Again, we saw-- so we'll talk about the answer engine
this morning.
It is powered by our precision-guided RAG solution.
This answer engine is capable of taking in domain-specific,
long natural language questions, and it is able to look at
complex sources, data sources, and give us precise answers.
Now, there's a couple of different modalities through which
we expose the answer engine for different personas.
We have the agent assist, customer assist, and employee assist.
Let's start with agent assist demo.
So as an agent, we saw the case summary, which was great,
but for me as an agent, what's even more important is
knowing exactly how to solve this case.
And that's exactly what knowledge summary is giving me.
Now, this case could have been a simple case where a solution
has already been identified, in which case I would have been
presented with the exact solution, which I can take and give it
to the customer.
Or it could have been a very complex case where there is no
obvious solution, but there is enough information hidden in
the underlying Don Ledge base, or it could be information from
other cases, other engineering tickets.
The information from there is counted, and we are able to
present the right troubleshooting steps for the agent.
So the agent now knows that at least he needs to do A, B, and
C in order to progress the case to the next step.
And as an agent, again, they're going to not just see the
summary, but I'm also able to see all the reference knowledge
sources, because I'm an agent.
I want to read more.
I want more context.
So I can always click into those reference knowledge sources
and get more additional context and details that I want.
And if I want to do more research, I always have the knowledge
source through which I can do my independent research to figure
out how I want to go about resolving the case.
Now let's look at how exactly all of this is coming together.
We can go to the presentation.
All right.
So behind the scenes, what we do is we auto-summerize all the
cases.
So every time there is a change in the case, we auto-summerize it.
We are able to auto-tag the issues of all the cases.
So we have a repository of cases with issues that have been
automatically tagged on these cases.
And then an agent comes in, I'm opening up a case, and I have
a new case that I need to work on.
So we run the automatic summarization on this case as well.
Now we are able to identify the issue of this new case.
Now I have a repository of cases where the issues have been
tagged, and I have this new case that has come in.
And my case summary not only gives the issue tag, it also gives
me the problem statement.
So I take the problem statement of the current case.
I take the entire case context.
I just don't -- I don't pass simple parameters, but I take
the entire case as a context.
I pass this on to the answer engine.
And the answer engine is already powered with a lot of data
sources.
It has all the knowledge sources that you might have.
And it has all the cases.
It has all the engineering issues.
There's a lot of context, complex sources that it can look into.
And it also has the case context and the problem.
So looking at all of this, it is able to give me precise answers.
Now we take the precise answers.
We pass it through our support logic AI engine again.
And again we have other LLM summarization engines and other
ML capabilities that helps us to get to the knowledge summary.
Like I mentioned earlier, the under the hood, the LLM that we
use for one summarization may -- like in this case for auto
summarization, we use a different model versus the case
summarization we use a different model.
So it's fit for purpose.
So once it goes into our support logic AI engine, we are able to
get the knowledge summary that you show -- that we saw in the demo
along with the troubleshooting steps or the resolution.
And that knowledge summary can also be used for drafting a
response to your customer.
And then we also have the jinnai tools which will take the
draft response and you can apply the grammar checks.
You can do the tonality checks.
You can do the translation and all of that.
And all of this is great.
We saw it in the demo.
What's coming next is automatically creating knowledge
from the knowledge summary we just saw.
Once we have the automatic creation of knowledge, we are also
then able to tag that knowledge to the issue or the case where
the issue was found.
Now, by being able to create a pattern or make a correlation
between the case, the issue and the knowledge, we are improving
our -- the precision of our answer region.
Now, the value of this improved answers are more precise answers
is not only going to help me as an agent, but this is also
going to help with self-service.
It's also going to help customers that are coming in through
portal and they're interacting with the answer engine via a
portal.
So with that, let's go into the demo of customer self-service
on portal.
So this portal is -- as I showed is plugged into the same answer
engine that is used that is powering the agent assist.
So the same thing is leveraged here as well.
So it is one looking at the same complex sources.
It now has all the refined and great knowledge that it needs
to be able to precisely answer your customers' questions through
the portal.
As you can see, the question over here is quite nuanced.
I want to remove a case from escalated list, but be able to
get back to it later.
How do I do that?
Now, that's very domain specific and that's a very long
phrase.
We are able to take that and, again, scout through all the
knowledge that we have and produce the knowledge summary that
you see on the right-hand side.
And it also gives you the referenced article.
So as your customer can always go and look into the details
right from here.
Now, another area where the answer engine power -- another
persona that the answer engine powers is your employees.
Now, the employees can now access knowledge.
The knowledge is democratized by giving them access to the
knowledge co-pilot via Slack.
Now, here they can come and just use /resolve and type in
their question.
Very similar to the agent, very similar to your customer.
Your agents can also now use natural language.
They can type in very long questions the way they want to
phrase it.
And they are able to take that in, again, plug it into our answer
engine and give you the same kind of results that you saw in
the other two modalities.
So here, as you can see, the expine answer gives you the exact --
it gives you the summary and it gives you the sources where
this was -- where the information was referenced from.
Well, that brings us to the end of the demos.
Quickly takeaways for us.
So investment in Giniai has gained traction, as you saw.
Companies are now more willing to invest, especially when it
comes to customer support operations, there is a lot of
opportunity to unlock.
Now, support logic, Giniai-powered capabilities is large in
here.
We have the summarization capabilities.
We have the insights engine.
We have the answer engine, which powers the resolutions.
All of this is going to significantly improve your
agent performance and the support of operations efficiency.
And our commitment to AI continues, we are investing in
agent engine and as you all know, that is going to be
transforming the whole customer support function.
All right.
Thank you all for joining us.
[Applause]