AI-Powered Intelligent Case Assignment
If you’re still manually assigning support cases, you’re behind. Join us as we dissect AI-powered case assignment, showing you how the right technology doesn’t just route workloads—it predicts the most effective resolutions. You’ll walk away questioning why you ever trusted a manual process in the first place.
You Might Also Like
Hello, everybody. My name is Tali. I am a product manager here at SupportLogic.
And
today I'm going to talk to you about the case assignment. So let me start. So
obviously
all of us doesn't want any cases. Let's start from the beginning, right? We don
't want the
cases. We want to solve the situation with the customer before they become a
case. But
once it's become a case, it's like a hot potato. We immediately want to assign
it, resolve it,
and make our customer happy. This is like number one priority in the situation
with the case.
So now who is the ones that are going to take care of the case, right? We want
to make sure
that we pick the right expert, the agent that knows to resolve the case. We
want to make
sure that we do it very fast. We assign the case very fast, so it will start to
be addressed
right away. We want to make sure that the expertise that we put into the
resolution of
the case is the right one. So the customer will be happy at the end, and
hopefully will
not open many more cases, and everything will be okay. But we also want to be
very fair
with our agents. We want to make sure that we balance our work so we not
overwhelm the
agents, so not the same agent getting cases over and over again. So for me, it
's like
if I think about it, it reminds me like kind of a human body. I have a heart
that is pumping
constantly. My cases are getting assigned every minute, every second, all the
time. But
then there is a brain here, like support manager, that needs to make sure that
the cases are
assigned appropriately. So it's kind of a balance that all the time happening,
and we're trying
to find solutions how to make it. So if we're looking holistically about how
the case assignment
made, we know that the case assignment appears across two main dimensions. We
have the manual
assignment, and we have the automatic assignment, and we also have it through
the push or through
the pull. So let's take a few examples. So manual assignment via push, it's
like support
managers that take a case and literally push it to the agent. You can imagine
that in the
scale it can be a problem, because support manager cannot all the time just
assign cases,
but some organization does using this manual push concept. Very fast, we know
that almost
all organizations have some kind of automation around the case assignment. The
common types
of automation are round robin, some organizations do skill-based assignment,
some of them using
the method. That of the case, kind of combination of all of these can be
possible as well.
So the final assignment through the pull is when we give our agent's ability to
cherry-pick
the cases they want. So you can imagine it also can be a little bit problematic
, because
agent can pick only the easy cases, or cases that he is more comfortable with,
so not a
very good option here. An automated pull is kind of more intelligent when we
help the agent
to bring the cases that are suitable for him, but he still can cherry-pick it.
So I wanted
to put out these charts, so we will remember a little bit better, understand it
a bit deeper.
What are the options for the case assignment when we go further? So I mentioned
a few methods
that we usually hear about them in the automated way, and those methods have a
lot of problems
and a lot of challenges. So some of them have bias in the assignment. For
example, the same
agents getting bombarded with a lot of cases. Some of them have misalignment
with the skills,
so we don't know exactly the skills, and we assign the case, and then the case
is stuck,
and agent is suffering, and the customer is not happy in all the subsequences
of the
event. Flexibility is obviously a challenge all the time. It's very hard to
scale and
make sure that the cases are assigned constantly to the agents in the right
volume. We're not
overwhelming the agents, but we're not stuck with the cases, so the flexibility
is a challenge.
Plus we have the limited scalability, as I mentioned, some solutions just not
build to scale when the
amount of cases is growing. So all those known challenges in the solutions that
we all familiar
with. So how do we do it in support logic? So you will see that the solutions
that we provide
is a scalable, very AI-driven solutions that combined with a lot of smaller
components,
and those components help us to take all the applicable agents that can be for
the particular
case and narrow it down step by step until we find these individual agents that
is the best
agent for the case. So here you will see some elements that I will show and
demo in a second,
but what you need to understand is that we scan the agents and we find the most
applicable agents
based on a lot of parameters that the modules consume. But this is not even the
best part,
so obviously we have the mechanism that can make our assignment easy, but what
is the best way of
doing assignment? So a lot of customers will ask for auto assignment, they want
to make sure that
all the cases are assigned all the time, but all of us know that in some cases
auto assignment is
not the best solution. Sometimes we need this manual touch for particular
customers, for particular
use cases, particularly SLA examples, etc. We will need this small channel of
manual assignment in
parallel, and this is the best thing that we at support logic trying to kind of
comprehend is to
provide both aspects of manual assignment and auto assignment that both
leverage this AI mechanism
and engine beneath, and in the combination of them we can provide the solid
platform of case
assignment for the organization. Cases that can be fully automated go through
the automatic route
and leverage the recommendation engine for auto assignment, and manual
assignment cases that
require a little bit more manual assignment will land on the board when the
manager can
go and manually assign those cases, but still have the agent recommendations
for the agents that
are best for those cases. So how do we do it? You're talking about all those
components and all this
stuff. So we in a support logic maintaining a five pillar kind of
recommendation, agent
recommendation infrastructure. Each one of them is packed with a lot of modules
and a lot of logic.
There is a lot of combinations also between those pillars which help us to
bring all this together.
There is also some small combinations and small modules that are coming around.
We didn't want to
mention them right away because it's just a lot, so we just want to focus on
the five main pillars.
Time overlap, skills match, customer experience, bandwidth, and assignment
balance. So let's
deep dive in each one of them. Time overlap. Time overlap is one of the main or
key features
that we need in order to resolve case successfully. We need to make sure that
the agents that is
taking the case is available for this case at the moment when the reporter is
asking for help.
We don't want to assign case to agents that will pick up the case a few days
later. We want to make
sure that whenever he's agent getting the case, he can ask the questions, reply
to the customer,
and collaborate in any possible aspect in order to resolve this case quickly.
So how do we identify this time overlap? For the agents that are working in our
organization,
we're using the shift management. When we assign those agents in the shifts,
shifts have assignment hours and this is how we leverage and they understand
what agent
supposed to be assigned over what hours. Sometimes we don't have this privilege
because the agents
are not assigned to shifts or there is any other combination, then we do some
prediction over it.
This is when the AI involved, we predict based on the agent activity what will
be the assignment
hours for this agent. Similarly, for the reporter, the customer, we're using
the prediction module
in order to understand when usually this reporter is applying the cases and
this how we merge between
the two. They hire the overlap, they hire the score and obviously we can do it
over the last
we are predicting next 24 hours. So if there is a weekend in the middle, we
will skip the weekend
as well. So we just want to make sure that we fully cover 24 hours of the
assignment.
We also take into the account the out of the office hours. So if agents are not
available during
period of time, we want to make sure that we're taking it into account.
The out of the office hours can be managed in support logic. We have a calendar
to manage this
but we also support it in the omni-channel for the self-force users. If we have
this information
in omni-channel, we can take it from there as well. So no need double work.
The second pillar is the skills match. So we all know that the skills is the
most important
thing almost for the agent to resolve the case good and knowledgeable. So we
want to make sure
that the agent with the higher percent of skills will get the case. We're
leveraging three different
modules to identify the skills. We have the case level skills, we have the
agent level skills
and we have the matching skill. Those modules help us based on the historical
data to understand
what skills are relevant for the agent and apply those skills, match them with
the case and apply
this logic together. Since the skills are something that is going and updated
constantly, we cannot
just do it once. So we're doing it on the monthly cadence and we're constantly
recalculating those
skills. This mechanism in place and the weightings that we have in the middle,
in the UI, we're showing
the top five skills that can be visible and can explain the calculation. But we
also hearing from
our customers that some use cases, it's not enough to have the skills from the
historical data.
One of the good examples is the new hires. When we have new customers come in,
new agents,
they usually onboarded into the system and doing some basic training and their
skills,
their original skills coming from those trainings. They don't have historical
data.
So a lot of customers are saying that those agents will be lacking on skills
until they will gain
the skills. So we heard this feedback many times and now one of the items that
we are currently
releasing is the manual skills management. So using the ontology file, the same
files that we
have today, we will allow to add or remove skills on the agent level directly
from the UI. In case,
we will need to adjust skills for some agents if they were not identified as we
expect or for those
use cases when the agents are completely new, we will be able to tag them with
the applicable
skills manually. Customer experience. Another very important parameter that we
want to not miss
is the experience that the agent has with the account that he's taken care of.
So when the
case is coming from account that is completely new to the agent, it can take
some time to the
agent until he will figure out all the details he needs. Therefore, we are
looking for cases
that already resolved from this account by this agent and trying to figure out
using AI
how much experience the agent has. The experience can be negative or positive
and we want to make
sure that we bring this together as well. We are using sentiments, we are using
the scores on those
cases etc. And eventually we bring it into the percentage of the match on the
customer experience.
The baseline is 20% just because we want to have also the negative impact. If
the sentiment is
negative in the cases that the agent already worked on, the percent will go
below 20. But if the
customer experience is positive because the agent worked on several cases and
those cases had a
good sentiment or a good score or a high score, the percentage will go higher
on those cases.
The first pillar is the bandwidth. So we obviously don't want to overwhelm our
agents by assigning to
them a lot a lot of cases. So we want to balance this. The balance is also
embedded into the
mechanism of case assignment. We are having a very configurable layer there. We
are using both the
priority and the status of the case to make sure that we can balance the
bandwidth that is applicable
for the customer. Every account, every customer has different variation of both
statuses as well as
priorities. And we want to make sure that we wait each one of individual
options there
appropriately. So eventually when we're looking at the bandwidth and the
calculation of the bandwidth
for the agent, it will apply the right weighting and right distribution. So we
will know that if
we're going to assign or we're going to make the bandwidth available, the agent
literally can take
more on his plate and he is the right one in this scale for the assignment.
The last point, the last big pillar is the assignment balance or what we call
sometimes we will see it
in the penalty. So the penalty is not in order to punish the agent. It's not
for this. What we're
trying to do is to again balance the assignment in order to not assign
constantly to the best
agents the same cases. So we don't want to make sure that agent is getting case
over case over case
and the other agents that are maybe a little bit less applicable and they're
not getting the cases.
So we're using small penalties that again configurable and weighted in order to
make sure that if case
assigned to the agent, the agent will get a little bit less weight and will be
going down the
recommendation agent. It's not that this agent will not be assigned. It just
means that we are
moving him a little bit down the row. So he will not be next in line for the
next case.
So eventually you can see those pillars in the UI. We're showing all of them
and you can see
those are the agents that are ranked. The AI modules all together will
calculate the
total weight. You will see it on the upper right corner and this is the first
in the list.
This is the most applicable agent for the case.
So how can we make sure that the steps that we are doing are for the case
assignment are actually
what will help us in the appropriate case assignment and not frustrate us
because
something is not working. We need to make sure that we have a strong keyword
and skill ontology
in place. Our support logic users can help with this. We have ontology files
that is available for
editing. We just need to make sure that the skills are in this file are up to
what the expectations are
and we now we can always edit them and now we will be able to add those skills
from the UI.
So just take a look and see that they are up to what you expect. Virtual teams
is the way of
grouping the agents. Obviously we can use the agents as is when we add them
into the shifts or
into the virtual queues. I will show it in a second. But virtual teams helping
us organize
those agents. So we don't need to deal with them one by one. We can create a
group of users and
put the users in the group. Shift is the way to manage our assignment. Every
shift that we create
we assign agents to it and we define the hours. I will show it in the UI in a
second. And virtual
queues those are the queues that help us to deal with the case assignment. We
can create as many
virtual queues as we need. Each virtual queue is a filter of cases. It's like a
list of cases
that are coming into this virtual queue and we can assign the appropriate group
of users into the queue
and this will mean that if the queue is automated those cases that will keep
coming into the queue
will be automatically assigned to the agents that are linked to the queue based
on the best available
agent. Auto assignment rules it's something that we added last year. Those are
the rules that are
helping to make our auto assignment a little bit more configurable. So we had a
lot of challenges
with just putting out the automation as is and trust the system. A lot of
customers come with
edge examples, edge scenarios that needed some twisting and tweaking. So we put
several rules out
to help with this definition. And the good news is that there are feedbacks
that come in constantly
more and more and we're going to enhance this and we're working on it right now
even more. So a
little bit more flexibility will come to the rules soon. So now just let me
switch to the demo.
You can see our assignment board. This is assignment board in the support logic
system and this is
the place when manual assignment happens. So on the left side you can see all
the unassigned cases.
You can bring them from the queues. Those are the queues we can design in the
system. Those are
just buckets of lists of cases that we want to interact with. And every case
that we will see in
the list will have those recommended agents coming right away on the screen.
You can see first three
but you can always open more options to see more. So if the queue has a group
of agents and you
want to see a little bit more than first three for any reason it is, you can
come here and see.
You also can, if you cannot find some agent and you don't understand why, if
you click show all,
you might bring more, some of them might be not available because they are
offline or out of
the office etc. etc. You can get to those, this information in the UI. So very
ex-explanatory.
So the best what you can do is for manual assignment, pick a case, select an
agent and assign the case.
While you are assigning you can send the message to the agent or to his manager
or to
yes to whoever you need, you can explain why the assignment is happening. If
needed
and that's it, like very easy. The recommendations are right in the screen. So
you're not doing it
blindly. You just pick in the virtual queue you want to focus on and go case by
case in order to
bring those recommended agents. It takes several seconds to bring
recommendations and then you can
assign. Shift management is the page when we define those shifts. This is the
one time work
that you don't need to do it constantly. Once you establish your shifts, agents
will be already
in the shifts and the assignment hours will be taken from there. When to create
a new shift,
you just go here, you give the sheet a name, you can decide the time zone for
the shift.
You can decide assignment hours, working hours. They can be the same. They can
be separate
and in the last step you assign your agent and again you can assign individual
agent or you can
use the teams. Once the shift is on the screen you can see you can always edit
it or delete it
and you can manage it. All the agents that are in the shift, they will be
getting those
assignment hours from this shift. Assignment queues is the page where we
managing the queues.
Some of our customers have CRM queues. queues that already exist in the CRM.
You totally can
use them. You don't need to create anything new. You just need to bring them
using this button.
You go here and you search for the CRM queue. It will come in the list and you
can edit and add
agents. Or if you want to be more flexible, you go to the second tab which is
called virtual queues
and this is where you create those virtual queues, those lists. Again, the
process is very similar.
You go to create and then you follow the wizard. You create your queue name.
You can start it from
the CRM queue. If you don't want to start it from scratch, you just grab the CR
M queue, you can
continue. You can add additional filters to the CRM queue. Those are the fields
and you can decide
what you want to filter. And then you also can, if you want to focus on
particular accounts,
you can do it from here as well. Once you have a virtual queue, for example,
here I created one
for the urgent cases. You can click on the agent column and add agents or teams
or organizations
or groups that you want to assign to this virtual queue. This means that agents
from those
groups or those agents will be assigned when the cases will fall into the
virtual queue.
It will rotate and find the right agent from those agents who is the most
available, most
capable. This is the one that will be suggested. Column auto assignment is the
columns it turns
on the assignment automation. So if you're not turning it on and you just
create a queue,
virtual or CRM queue, it will be available on the assignment board and as I
said, you can
leverage it for the manual assignment. If you want to leverage the auto
assignment,
you just simply turn it on and the auto assignment will kick off.
Here under the settings, you will have several rules that I mentioned them at
the last slide.
So you can explore the rules. Some people want to always have the queue cases
assigned.
The other one to do the opposite, never get them to assign. So there is an
option for this.
Some people will prefer to perform round robin then to go to the auto
assignment,
to the ML assignment. So we can leverage the round robin settings. We have
ability to limit the
capacity agents for a day. So for example, we want to make sure that each agent
that going to be
assigned will not get more than five cases a day. So this is the setting here
for it.
And we also can switch between the assignment and work in hours. As I mentioned
on the shift,
when we add the agent, we can assign assignment and working hours for the agent
. This is coming from
there. So the bottom line, if everything is there and everything is working,
you can see your last
three days assignment in the assignment board under the recently assigned tab.
All the cases that
were assigned, both from our system or from the CRM, will be here. And you can
always go to the
three dots and pull the case assignment timeline to see the process of the
assignment. So we see,
for example, here case was created, Aka is the auto assignment was kicked off,
case was auto
assigned to who and from what queue, etc. etc. Sometimes we are doing reass
ignment.
You can also scroll on the screen itself and come to the auto assignment. And
when you pull it out,
you will see the agents that was picked for the assignment. And you also can
see all the agents
that were there before. And we did not pick them. So you can actually check the
mechanism and see
why we picked one versus the other, etc. If you want to get a little bit more
information.
So with this, I will go back to the presentation and summarize these two last
slides.
This slide is about one of our customers, Koval and the numbers that we help
them for the case
assignment. As we all understand that the right case assignment, the successful
can save us a lot of time and a lot of money. It can make our customers much
happy. And we can see
Koval did reduction, 53% of reduction in time resolution. We have a 56%
reduction in escalation.
So the customer is much happier. The case is resolved much faster and better
quality.
And we have also the same day resolution increase of 31%.
So the key takeaways. Intelligent case assignment is
something that we all need. This is the bottom line. We need to make sure that
our cases
assigned intelligently, assigned quick, assigned to the right people. We want
to make sure that the
agents with the right skills addressing our cases. We not make sure that they
are getting it in the
right balance. We want to make sure that we have the automation in place, but
we also have some manual
option for the use cases that are more gentle and need some manual touch.
And we also want to make sure that we have some rules. So not everything is a
black box
and not everything is AI and ML. We also have some rules in place in order to
navigate
some stuff around and help the machines to do the right decisions.
Okay, that's it for my end. Thank you very much.
case assignment,