Product Keynote: Redefining Post-Sales CX with AI
Showcasing AI solutions that are redefining support: from automatic case and account summarization to intelligent case routing, precise account health scoring, and seamless voice integration. This isn’t just about improving efficiency; it’s about transforming your entire support strategy. Get ready to see how these innovations can propel your operations to new heights.
You Might Also Like
Good morning everyone. I'm also joined on stage by my colleague, Krithika, huge
round
of applause for Krithika as well, please.
All right. First, I want to extend a heartfelt thank you to all our customers
and partners
that are joining us here today. As Joe mentioned, we have an incredible lineup
of speakers and
a power-packed session agenda through the course of the day, but hopefully will
inspire
you and empower you. But more importantly, we're here to learn from you, to
listen to
you and build connections. So I highly encourage all of you guys to stay
through the day, meet
all the experts that we have from SupportLogic here, also meet some of the
industry thought
leaders that we have here as well and make best use of this conference.
It's impossible to be at a product keynote in 2024 and not talk about AI, right
? But
we're not just going to talk about AI, we're actually going to show all live
demonstrations
of all the wonderful AI stuff that we've been working over the last few months.
And
we all know live demos can sometimes go wrong, right? I mean, I always say they
work perfectly
till the time you're going to show that they work perfectly, right? So if they
go wrong,
cheer for us, clap for us, pray for us, and we'll make them work. All right.
But before
I go into the live demos, I think it's important to touch upon the evolution of
AI at SupportLogic.
Krishna talked about it. We're extremely proud of this because, you know, from
the time SupportLogic
was founded, we've been pushing the boundaries of innovation by leveraging the
latest trends in AI.
In 2017, when Google announced their first large language model, Krishna talked
about it. But
we were one of the first early adopters in the valley, at least in the startup
ecosystem,
to be leveraging bot, to build classifiers and build our aspect-based sentiment
analysis.
This is the core of what we've done from the beginning. We've heard some
wonderful stories
today from some of our customers that use some of these components. So this is
the signal
extraction engine that we already built out in 2018. In 2019, we launched our
scoring engine,
which is essentially a heuristic-based model, which we leverage for sentiment
scores, attention
scores, QA scores, account health score, and so on. And this is also sitting on
a very,
very strong foundation of some of the mathematical models that I used in
digital signal processing.
And we did this way back in 2019. In 2020, we launched our flagship
predictive engine, which predicts escalations. We heard so many wonderful
stories
of how customers have used our prediction algorithms to reduce escalations.
This was done already in
2020. In 2021, we launched our multi-model recommendation engine that we use
for case routing. And this
works in actually two modes. It works in a fully autonomous mode, where based
on many different
criteria, we're going to look at all of this in the demos. It can automatically
route agents,
or route tickets to the right agents, or it can also work in a manual mode,
where it can help
a human being pick the right agent for the right case. Now, this is interesting
, because there is
all the hype around agent KI right now. This was our foray into agent KI way
back in 2021.
Then we built the alerting framework in 2022. Last year, we introduced the
summarization engine
that we're going to look at in detail during the demos. We do summarization at
an account level.
We do summarization at a case level. We do summarization at a knowledge level.
And we're going to look at all of that in the demos as well. And of course,
under the hoods,
we use the best set-for-purpose large-language model. We use entropic cloud,
sonnet-3. We use Mistral. We use AWS Bedrock services. But again, this is out
of the experience
that we have of which model is best for what kind of a use case. And then
earlier this year,
we launched our precision rag powered answer engine. Now, this is huge, and
this definitely
needs a little bit of an explanation. We all know in complex tech support, your
simple keyword-based
searches, or even for that matter, your common rag architectures do not work
because any internet
trained embeddings will not correspond with your domain-specific corpus. So
basically,
what our answer engine does is it's an implicit answering engine that
continuously learns from the
data. It combs through all the knowledge sources. It could be dynamic sources,
sources like your
cases, like your gyro tickets, as well as static sources like your KB articles,
your publicly
available documentation on the product or whatever, and then finds the right
solutions and the right
answers for the problems that you have. So this is huge, and we're going to
talk and show you
all of this in some time as well. In summary, we have two best-in-class engines
and models
for signal detection, for predicting escalations, for recommending the right
agents,
and we are also pushing the boundaries of innovation on the cutting edge,
agent AI, as well as precision rack. Now, combination of all these AI
innovations
combined with the richness of data that we have across all post-sales
interactions,
combined with the deep understanding of the business domain that we have from a
support context
perspective with the company, as well as very thoughtful and a purposeful
approach to bring
all of that together into a unified all-in-one supposed experience platform
gives us the portfolio
that we have that you see on the screen. Again, I'm not going to go through the
slides.
We're going to very quickly go into the demos, but just a quick recap, Krishna
had it on his
presentation as well. The core is meant for support operations efficiency. It
has the sentiment analysis,
it has the escalation management engine, it has the backlog management, and
then we have the
four add-on modules. We have the assign module to basically intelligently route
the right cases
to the right agents. We have the elevate module, which is essentially our
quality monitoring and
coaching tool. We have the assist module, mainly catered towards agent
productivity. We have the
expand module that Krishna announced today, which is meant for technical
account managers,
account managers and customer success managers to get a holistic score above
the accounts.
And then of course, we have the resolve Sx, which is essentially our precision
rack,
powered answering engine, which could be used both by the customers on the
portal, on the chat
bot, but it could also be used internally by any person in the organization,
and we will have a
good demo of that as well. With that, limit of truth, demos. All right, so
there are five
demos that we would like to show you today for the five different personas. We
're going to start
with the agent, then go to the support manager, quality assurance on an auditor
, account manager,
and what we're calling as the answer seeker. Let's start with the agent. Again,
we're going to look at how an agent can review and prioritize their case
backlog,
troubleshoot a complex case, get summaries, get the right answers, search for
additional
information that they may need before they respond back to the customer, and of
course,
in the end, craft response. Let's switch over to the demo now.
So what you see on the screen right now is essentially a view of support logic
that can be embedded
within your system of record. In this case, it's a sales force. It could be a
service now. It could
be a zendesk. It could be a Microsoft Dynamics or whichever CRM do you use. So
basically, what we
are doing is while the agent is looking at cases, we are able to bring
sentiments and signals
from support logic into this UI so that they can identify which are the right
tickets that you need to be working on. For example, the attention scores give
you
which are the most urgent cases that need urgent attention. The sentiment
scores,
as well as the sentiments that you see on the top, also give you a good pulse
on what are the
different sentiments across all the different cases. You can filter, you can
start, and so on.
Let's see, I've identified as an agent one of the cases that I want to work on.
I click on this,
and then I land on the case details page within your sales force. On the right
hand side, what you
see is essentially our embeddable widget. This, again, is currently being shown
in the context of
sales force, but you can embed this widget in your CRM as well. Now, the tab
that we're focusing on
right now is called the Resolve Assist. Think of it almost like helping an
agent troubleshoot a case.
Now, whether it's a simple how-to kind of a case where an agent just needs to
look for the right
knowledge article or a solution, or it's a more complex case where the agent
really needs to
troubleshoot, Resolve Assist basically helps you do that. How do we do that?
There are three things
that are extremely important once again. First, as we said, the simple keyword
basis searches
do not work, just because the subject of the case has two keywords you cannot
always assume,
at least in complex tech support, that you'll be able to find the right
solution based on those
keywords or any other metadata. So what we essentially do is this is built to
work for
almost case as a query, as opposed to keyword as a query. So we can take from
the case summary
that you have on the top of the screen, almost derive the context of what the
case is and pass
that almost as a query to our knowledge engine. Second, then we comb through
all the knowledge
sources that it has been connected to. Again, as I said, these could be static
sources like your
KB articles, or these could be complex dynamic sources like your gyro tickets,
like your past
cases and find the right solution. So on the screen, you can also see it shows
all based on the
implicit query, which was the case summary, it's able to find the right
knowledge sources or the
past tickets. From here, I can click on any of these and I can navigate to, if
it was a case, it will
go back into your support logic system, or if it was a knowledge article, it
will go back to
wherever your knowledge article is stored. On top of that, we also actually
produced the
knowledge summary. This is again a good effective use of a large language model
. We use Mistral for
this, where we essentially take all the possible right solutions that the
system has identified
as part of the trouble shooting, and we are able to compose a knowledge summary
based on
the reference knowledge articles. On top of that, I can also go to the
knowledge search tab on the
far right, and this is where outside of the case context, as an agent, if I
have an intuition here,
I've heard about this issue before, I can just look up for something, and again
, outside of the
case context, also find something else that could be helpful in providing the
right solution to the
customer. Now, before I respond back to my customer, maybe I also want to get a
good sense of
all the latest sentiments. So this is where I go into the case and side-stab.
This is where you
see that there is an extremely low sentiment score of zero, a very high
attention score of 100,
there is a lot of negative signals, including frustration. So maybe this is an
input that I
want to use as an agent to fine-tune my response when I go and respond back to
the customer. So,
by a single click, the knowledge summary along with all the articles that were
found as possible
solutions are just as a response on the result of this step. You also have the
next steps and stuff
like that. What the agent can also do is they can attach any other files that
they may want to
attach before sending the response to the customer. They can change the ton
ality view. We have prompts,
built-in in our platform for tonality. For example, and probably I want this
response to be a little
bit more empathetic because of all the frustration that the customer has been
facing. I could also
change the language of the response in case I'm responding back to a customer
that prefers a
different language or built-in in this productivity toolkit packaged for agents
. So this was the very
first demo geared towards how do you make agents productive by helping them
troubleshoot and find
the right solutions and quickly send those responses to the customers. Second
demo. This is my favorite
one because this honestly touches pretty much everything that we've done and we
are doing as a
company. This is a support manager. So what we're going to look at in the
context of a support
manager is typically support managers are either working on escalations or
working to prevent
escalations. So in this particular demo scenario what we will see is how a
support manager monitors
the support queue, focuses on the urgent and the escalated cases, finds a case
that is likely to
escalate because it's predicted by our predictable garden, gets a summary of
the case activity because
the support manager does not in the day-to-day works of working on the case
directly,
leverages AI to determine the next best action and also marks this particular
case
for audit which is something that we will show you as well. Okay, over to the
demo.
All right, so now a support manager can either be working on our home page, we
call it the console
for all our customers that use it today or the cockpit which gives a good sense
of what's happening
for a support manager but in this particular case for demo we've configured
certain alerts using
our alerting framework in the engine that I was talking about which is pro
actively notifying
the support manager that there is a case that is likely to escalate and this
could be in Slack,
this could be in email, this could be in MS Teams or whatever communication
channel you use in your
company. So in this particular case it's showing me you know one of those cases
is likely to escalate,
I can actually take actions here in Slack itself but I'm actually going to
inspect this case in
more detail and launch directly support logic UI from here. All right, so now I
'm looking at a
case as a support manager and this is a classic omni-channel case which may
have started on the web,
maybe there were some chat interactions, looks like there is also a voice
interaction here.
Krishna was talking about our integration with telephony systems, so in this
particular case
the voice recording from any of your telephony systems is brought into our
platform and then when
we run our sentiment and signal detection we also run it on the voice
transcript itself.
So as you can see on the top there are so many different signals that have been
identified
including signals that were detected on the voice call itself.
Okay as a support manager I have a pretty good sense of you know there's a lot
of frustration
and urgency and a churn risk signals. Next maybe I'm going to get a quick
summary of what has been
happening on this particular case. So I can quickly click on the case summary.
Now again when we're
talking about case summarization in the previous demo from an agent perspective
, the persona is
agent and the main context is troubleshooting whereas for a support manager it
's less of the
troubleshooting. It's more of what should I be doing as a support manager? Do I
need to bring a
swarming team together? Do I actually need to maybe even reassign the agent? So
here I get a
very quick summary for a support manager on what is actually happening on the
case.
In addition to that I can click on the start review and what start review is
going to do is
it's again going to invoke our escalation engine and give you the reasons why
our engine thanks
this case is likely to escalate. For example this could be because of things
that are happening on
the case. For example there are 33 conversations in this particular case there
are 12 negative
signals. It could be something to do with the agent activity. For example the
agent hasn't
responded in the last few days. There are a lot of cases that agent has on his
backlog. The agent
is actually actively working on 14 other escalations and most important the
agent does not have the
right skills to be working on this particular case. Or for example customer
activity for example
this customer has had five different escalations in the last 90 days. So all
these different
contributing factors are leading our engine to predict that this case is also
likely to escalate.
What the manager can also do is as you can see at the bottom right of the
screen there is the AI
assistant recommendation where AI is recommending that maybe you don't have the
right agent working
on this particular case. And then it basically shows you certain other
recommendations of other
agents who probably have better skill match. Probably they have better time
overlap with the customer.
Maybe they have had better experience working with the same customers on the
previous escalations.
And based on many of these different contributing factors the manager can
decide to reassign the
case to the right agent. All right. Next I as a manager I have a good sense of
this now. I see that
you know the quality score is 75 and QA auditor's world. This is not considered
to be a very high
score. Maybe this is something as a support manager I want to mark for review
so that
the auditor knows that this is a case they would like they should review. What
I can also do is I
can use the share option where again on an email on a slack or whatever
communication platform you
use you can essentially tag the auditor and trigger the review of the case from
here itself.
All right. So that was the second part of the demo focused on the support
manager.
Let's go to the third one. The third is in the day in the life of a quality
assurance on an
auditor. Here I mean what we're going to see is how they can review the
compliance scores
and trends across all cases. They can conduct a quality review on a complex
case we're going to
take exactly the same case that we looked at right now which was omni-channel
it had voice.
So the auditor can actually look at the auto QA results and our QA module also
operates in two
modes. There is the fully autonomous mode so that 100% of the cases in the
tickets in your platform
are always going to be auto QA. On top of that we provide the manual mode where
an auditor can say
okay out of maybe the 10,000 cases here are the 500 that I would also like to
manually audit.
So we are actually going to see how the auditor can do a manual QA on the same
case and from here
let's now launch our elevate product which is our quality or a trauma. All
right. So we're on the
same case case number 239. It's the same complex case it's omni-channel there
is voice.
The first thing that I'm actually going to show is click off you could click on
the language please
yes please. So as you can also see the original maybe your L1L2 support was
dealing with the
customer in Japanese. So as you can see you can see the original text. You can
also if you
as Krotaka has already actually highlighted Kathy is here the agent. We are
also able to detect
and score on skills and behaviors that you have defined also on the non-English
text. What we
essentially do is we get the non-English text we translate it and then we
essentially run our review
for the agent skills and behaviors on the English text. So as you can see first
of all there is also
a review happening on non-English text. Let's switch back to English.
Let's go to the voice call or maybe before the voice call. Actually as you can
also see John Brooks
is actually the customer. Here as an auditor I'm not just reviewing the agent
for their skills
and behaviors. I'm also able to get a good sense of the customer sentiments. So
I can very easily see
through the course of the conversation the customer was angry. There was a lot
of negativity so all
our sentiment signals are also available for the auditor in our elevate product
. Let's scroll
further down and now you have the voice call. And this rainbow picture actually
is very interesting
because first of all you know you can play back the actual call from here
itself. The different
colors actually indicate all the different sentiments and signals as well as
skills and behaviors
that were detected. And this is on two things. For example we use the actual
voice call and run
acoustic models on top of the voice call to be able to detect things like was
there dead air was
a too much whole time. At the same time if there was negativity, if there was
profanity, if there was
whatever other things that were detected you would be able to see on the score
card on the
right. So on the voice call as well you have all the criteria that you could
have defined as an
auditor. These are the skills and the behaviors that you want to essentially
review for every agent.
So across the voice call for all the skills and the behaviors that can be very
easily
configured in our platform you can essentially give a rating and you can add
detailed comments
on top of the ratings. Now what I can also do is in addition to the auto QA, I
as an auditor can
actually perform a manual review and I can do that in two ways. One I can say
okay I'm going to start
with the auto QA as a score as a baseline and then enrich that further so
photography would just
select yep and just proceed to review. So now it has taken essentially the auto
QA
scores and given me as the starting point I can say okay I listened to the call
and I think this
is not positive this should be actually negative so I can make all those
changes. Cancel and please
go back to the call or what I can also do is I can start the manual review but
start from scratch.
Clean slate I don't care about what the auto QA did I just basically want to
know for the entire call
or for the entire case when it's one of the review from scratch but again it's
the same score card
you can actually configure different score cards for auto and manual and you
can go through the
entire process and I can also of course add notes. So what we've seen so far is
an auditor
looking at a complex case that had multi-language that had different type of
interactions, voice
we are able to essentially detect the sentiments and perform a review across
all of those.
The other thing that I'm going to talk about is we have built in a lot of work
flows,
assignments is just one of them. What are assignments? Assignments are a
pragmatic and a
programmatic and a consistent way for auditors to define certain rule
conditions where you may say
every time there is a voice call, every time the CES the customer effort score
is between
whatever 10 and 30, every time there is a likely to escalate signal I always
want a manual QA to
be performed. So these are just consistent ways of enforcing compliance or
enforcing that certain
cases always get polluted. From here I can actually as an auditor trigger the
review itself. So
what Krithik other was she was in the MyAssignments and she clicked on, okay I
'm just going to go
through my daily workflow and directly land on a particular case that meets
that criteria and then
I can basically perform a review directly from here and again the process is
exactly what we
already saw. The other thing that we've built in is the disputes. Now whether
it's auto QA where
we are leveraging AI to be able to grade a case, to be able to grade an agent
or even if it's a
manual QA process it's never going to be 100% accurate. There could be cases
where I as an auditor
think that for example the agent did not have a good starting whatever skills
and behaviors I
may have decided but the agent does not agree to it. So this is where we have
the built-in workflow
right the dispute workflow where now we are looking at from an auditor lens. I
can for example open
up one of the disputes and yep the far right yep open up one of the disputes
here as you can see
the auditor maybe just scroll down and then the other way please perfect. So
for example the
auditor for the opening which is let's say one of the skills that you want to
review the auditor
things the introduction was bad. The agent in this case Krithika does not agree
to that so she has
basically raised a dispute and said well the introduction was done as per
whatever our best
practices. So maybe I can re-listen to the voice and I can potentially resolve
the dispute from here
and as an auditor I can say okay maybe I agree it was my mistake. So all these
built-in workflows
are extremely important because it's not just using AI to do something we know
there are always
workflows like assignments like disputes are always going to happen and our
platform enables
all of these workflows out of the box. All right let's go into the to the
fourth part of the demo
which is for an account manager. Here we are actually going to look at you know
a strategic
account look for any kind of commercial signals these could be positive
commercial signals right
there is a renewal opportunity there is an expansion opportunity or these could
be negative
renewal signals as well for example there is a churn risk I could check the
health scores and
the contributing factors I can collaborate with the team to address some of the
high
priority cases that are associated with one of the strategic accounts that's
probably up for renewal
and then I can also you know track some of the other account trends let's go
into the demo.
So let's say an account manager is typically living in your CRM system. So
again as we were
saying before these widgets can be configured and can be enabled in any of your
CRMs for the
for the demo purposes we have a sales force. So when I'm looking at my sales
force account
I can see that it's a high value account maybe just scroll down on the on the
left on the sales
force we are critical. So let's say it's a it's a strategic account the renewal
is up
in a month it's a high value account and then on the right you see the signals
from the support
logic system where it shows you it has a fairly low account health score it has
some active
escalations there are a lot of negative sentiments and signals and then there
is churn risk. So as
an account manager this is something that I need to address. So from directly
here I can again launch
support logic and I land on what we call as the account hub. Now as soon as we
land on
the screen you can see the account summary this is again where we are using AWS
bedrock services to
to create the account summary. There are essentially three sections in the
account summary.
The first one which is the current status is essentially giving you some key
insights
grounding you to some of the facts that you need to be aware of. For example in
this particular case
there are three active escalations that require an engineering fix as an
example
or there are six production issues that have been reported and all of them seem
to have a high
rate time. These are all grounding facts that you need to know as an account
manager. The second
section is the signals and the risk indicators. These are essentially warning
signals and sentiments
coming from our signal extraction engine. For example it's showing that there
are potentially
renewal risk because there are three-choned signals that were identified in the
last 90 days
or there is a lot of frustration signals found in 30% of almost all the open
cases. The last section
is essentially the overall trends in the issues and these are essentially
I would say potential next best actions recommendations coming to the account
manager. For example in this
case you know the cases with engineering issues seem to be taking a long time
than what typically
does so maybe it's indicating that there is a lack of follow-through. All for
example customer
has a wait time that has increased 10% in the last one month which possibly
could be indicating
that there is maybe resource allocation that the account manager could be we
need to be looking at.
So basically an account summary prepares an account manager on what are some of
the reasons why the
account has a churn risk, why the account does not have a good health score and
so on. And again
we are able to do this across all different kinds of four-sales interactions.
Currently we have the
support for support data what we are also adding in the coming days is support
for customer success
interactions, support for any of the onboarding tickets that we could also be
contributing towards
the overall account health score. Now if you scroll down you have a trend line
for the account health
score showing you over the last three months, six months, nine months how has
the account score
being trending. Now here again for the account score we used a heuristic model
that takes into account
many many different things to be able to calculate the account score. For
example if there are dips
or spikes in the cases if there are too many escalations what's the severity of
the different cases
or what's the quality of the service that we've been providing, what are the
different kind of
sentiments that may have been detected over a period of time. So all these
different contributing
factors holistically go into the heuristic model and come up with an account
health score. What I
can also do from this screen is I can drill down into the data for example I
see there are three
active enclosed escalations by clicking on that tab I have exactly those three
interactions
that are contributing towards that. I can actually for example the first one is
an escalated case
that has a potential churn risk. I can actually directly drill down as an
account manager directly
into that particular case again potentially use the get summary to get a very
quick overview of
what's potentially happening with this particular case and then maybe
collaborate with the support
manager or maybe directly with the agent to help resolve the issue. The other
thing that we can
also do is maybe just click on 16 negative signals perfect scroll down and you
can see now all the
16 interactions that have negative signals some of them also have churn risks
and I can exactly do
the same what we did before from here I can go specifically to that particular
interaction
and see what's going on. So in a nutshell the account hunt provides a holistic
overview to an
account manager on what's really happening at the account. It could be quant
ified within account
health score which is taking into account interactions across all different
four sales teams not to
support going forward customer success interactions and on boarding tickets as
well and gives you a
quick summary through the account summarization that we've built in for them to
get a quick
glimpse of what's potentially going wrong. Let's quickly also look at maybe the
positive commercial
signals right. For example in this case I've configured an email alert that is
picking on a
positive commercial signal. If I click on view in support logic I can directly
from here
go into a case. Now this is again very important and something that we all know
in the support
industry. Sometimes there is so much of this information and knowledge sitting
in the support
cases that never gets to the account managers. For example in this particular
case there is a
potential renewal opportunity that the customer was talking about in the
context of a case
and we are able to pick on some of these signals could be renewal could be
expansion
opportunities and then throughout alerting framework notify the account manager
or anyone
else who needs to be notified. With that we come to the last part of the demo
which is for
the answer seeker and this is where I would like to invite on stage my
colleague Sario
to come and talk about the answer seeker and more broadly the knowledge co-p
ilot. But before that
Pratika there were no claps, there were no cheers it looks like our prayers
worked. These are all
life demos my friends and please spend as much time as you want with us we're
going to be at the
demo booth. There are a lot of experts that we have here we have Kratika we
have Pali we have
Head of ML Alex we all have sessions here but we would also love to spend some
time with each
of you so please feel to do so. Thank you so much. Sario over to you.