Thank you for joining us for our Asian
Forest Winter 26 release Q&A session.
I'm Sean Kane, your host for this session,
and I'm joined by our product managers
from Agent Force.
We're here to answer the questions
that you've asked in the community,
as well as some frequently asked questions
provided by our product managers.
Before we get started,
I wanted to remind you
to make all purchasing decisions
based on currently available technology.
And this format
is going to be super casual.
So we're going to provide answers
to your hard hitting questions.
So let's go ahead and get started.
All right so Corey starting with you here
an escalation
fallback will escalation fallback occur
every time a session is left in the queue.
Is there a set up interface
to determine when fallback occurs versus
There's not a set up interface.
It's all done through your harmony flow
that you configure
and connect connecting Agent Builder
through the connections tab.
It doesn't occur every time it occurs.
I tried to call this out in the demo.
a omni flow is finished without a root
element being called in the flow.
So in the flow, you have, an element
or an action called root work.
If that gets called, attempting
to root the work
out to a queue or a channel or something,
then we will not fallback
if that flow is completed and no root work
element is ever called.
Then you can, you can be confident
that we will fall back to the agent.
So the important part is just having
some sort of decision tree in your flow,
where you go through a different set
of parameters such as queue, wait time.
And if not, don't
call that root work element.
So let's do another question
for escalation.
Is there a new escalation topic
to have to configure, for this to work,
or do we need to change
my topic instructions?
The existing topic that you have in
your agent will work completely fine.
There's no need to add a new topic.
There's actually
no need to add new instructions either.
The work that was done
here is all on the back end.
So again,
we recognize when the flow is completed.
If there is no root work element called,
then we just have our messaging service
communicate back to Agent Force
and say hey, the escalation failed.
Please keep the agent in the conversation.
So, there's no, new topic.
There is some kind of, tailoring
you can do in the instructions.
So when fallback occurs, the escalation
topic is once again
called with the context
from our messaging service
that, hey,
we attempted escalation and it failed.
These kind of,
move forward with the conversation.
So you can include details
in your instructions, such as on fallback,
ask the customer
if they would like to create a case,
or ask the customer
if they'd like to do anything else,
or apologize
that we can't, communicate this time.
So you can use
your instructions to do that.
So another question here.
I saw another action called check
availability for routing.
Is this the same thing
or do they work together?
So check availability for routing,
was released by our service cloud team.
And they do have
slightly different functionality.
So our escalation fallback feature
kind of answers in the name here.
It's all about fallback IDs.
If all things go wrong and no one, answers
the next sort of, queue,
then we want to make sure
that your customer doesn't get left
abandoned, in a chat session or what?
However, they're speaking to your agent.
The check availability for routing action
developed by Service Cloud.
You can use this if you want, like really
tight tailoring on when you can escalate.
The reason being this action your mimic
your your your flow for escalation
and check everything
before the agent attempts to escalate.
And by doing that, you don't
get that message that you saw in the demo.
Like, sure, I'll escalate you.
Oops, sorry, I tried and I can't.
So if you want to do the check
availability for routing,
you'll avoid that message.
The only the only kind of difference is
you do have to have the same logic
in that check availability flow
that you have in your escalation flow.
So customers
kind of can choose what they want.
that more complex configuration,
then the agent can be more tailored.
Or if you just want to make sure that
if all things go wrong, it falls back.
That will be handled either way.
Let's go to the last question here
for escalation fallback.
Once escalation
fallback occurs, can the customer loop
and try to be escalated again.
So once fallback occurs we recognize that
from like a platform level.
And we say, hey, we try to escalate. It
failed.
Don't do it again in this session.
So in that customer session,
when the conversation falls back
to the agent for that next interaction
or that next agent's response,
it is siloed to the escalation
topic only.
So that's where if you add instructions
to that escalation topic,
you can be 100% sure
that they will be considered on fallback
because the agent is forced
to stay in that escalation topic
after that turn, or that interaction
or that agent reply.
It does open itself back up
to any of the topics in your agent,
but the escalation action itself is
blocked for the remainder of that session.
you know, service center opens at nine
and someone attempted this at 859
if they attempted to again,
a 901 escalation will be blocked.
That's
just how the platform works for now.
And we've considered
making updates to that in the future.
But that's just so that customers
don't attempt to do this, you know, ten,
20 times. And, kind of confuse the agent.
Let's go over to the next feature
here, Prompt Builder enhancements.
And Liz's first one's going to be for you.
Where can I find the model limitations
for multimodal okay.
So in the settings panel
which is on the left hand side,
you can see all of the model settings
as well as the model limits.
What this means is for each model
that you choose, you can see
whether that model supports
images only or images and PDFs.
And you can also see, token size limit
for that model.
You can also in that left hand panel,
configure what's called file
exclusion priority, which is if you hit
those limits for those images in PDFs,
you can tell the além
which ones to process and prioritize.
So let's go to the second question here.
Where can I go to set up dynamic search
filters.
So dynamic search filters are configured
in data cloud.
in Einstein Studio when you're setting up
or editing a retriever.
So ultimately that setting will be found
in Retriever Builder.
And that Pre-filter section
can be found in the retrieval panels
inside of Prompt Builder
when you're configuring your retriever.
But to set it up from the very start,
you need to start in data cloud.
Very good. And last question here.
Can I provide a custom output schema
schema for my prompts?
Right now we support default
HTML and Json out of the box.
But safe harbor at the end of October
you should be able to specify
if you want, custom output schema.
So not right now, but in the future.
Let's go to our last topic here.
We're going to be talking
about advanced connections setup.
So Abby, what are the exact requirements
for getting carousels to render properly
I mean, in other £0.03 channels?
as part of the new advanced, connection
setup, we have enabled this capability for
you to, render rich components
as part of your various channels.
So what you need to do here
is to be able to access this,
you have to go to your connections
tab in your agent builder,
and you will see a banner which says that.
Now you can switch
to the advanced connection setup.
Now you will see a completely new
connections pane,
and you will see a messaging connection
automatically attached to your agent.
Once you go into that messaging
connection, you will be able to see,
an option
called adaptive response formats.
And once you turn it
on, you will be able
to, start showing rich content,
as part of the agent response
specifically for carousels,
the way it works is you need to build
a custom action, and this custom action
should be able to output a set of things.
One is the name of the list option, then,
the the image URL,
what you want to show in that carousel.
And then
you also want to show a description
text and also the mime type of the image.
For example, Jpeg of what I mean.
So once you have this custom action built
in this specific format,
it could be an apex action
or be a flow action.
You will be able
to take the agent response
and then send all this payload,
through the adaptive response format.
And it would be able to render
as a, product.
So this is a very popular, UX control,
which is used for showing, product
information across a lot of our retailers
and various other customers.
And one thing to note about this product,
carousel says
it, it is not supported
in all the channels.
It is supported in the meow channel,
which is our first party channel, and also
Facebook channel, Facebook Messenger.
So, so this is the way
you can just make sure that the current
product carousels with rich experience
like images, product description,
how that is working
end to end in these two channels.
Let's go to the second question here.
Why are images inconsistently
rendering in external channels.
So example WhatsApp or memo
when using adaptive rich toys
responses from surfaces okay.
So the adaptive rich choice
response is what we have in
is basically a normalized
normalized format.
And we actually translate it
according to what
that individual channels are supporting.
So you could have a scenario, for example,
let's go back to the carousel.
And you want to send the,
particular image or a set of images
as part of the product
closer to that, specific channel.
But if, if the channel or the client
is able
to support that carousel component,
for example, Facebook Messenger
is able to support a carousel,
it will be able to display that currently,
whereas if a channel is not able
to support that carousel component,
it will either fall back to a text
only response
or it will just fallback
to just a button list.
So that is what is that is
what is probably what we can see here.
So it really is controlled by what
that particular channel
is able to support,
for that particular output.
So that's why you see
a little bit of inconsistency.
The same carousel could be shown as a list
picker inside WhatsApp.
Because we are not supporting,
carousels in WhatsApp.
So that is the underlying reason
why you see that inconsistency
Thanks. Let's go to the last question
here.
How can we restrict the adaptive response
format to only specific actions
like images or carousels, instead of
having it applied to all responses?
Okay, so that's a great question.
So one of the main reasons,
one of the main feedback,
what we have got is we want the customers
want to be able to specify exactly
when, carousel should appear,
when an image should appear,
or when a button should appear. Right.
So and we are actually adding
those advanced capabilities
as part of the next phase
of our innovation as part of the next gen
authoring builder, where you will have
more deterministic, controls.
you can still enforce some of these
conditions by the topic instruction.
For example, you
if you want to make sure that a particular
agent responds should be always,
rendered has buttons.
You could explicitly instruct the agent
within the topic instructions
to select a specific choice.
You could say something like,
hey, present these results as an option
that the user can choose from.
Or you can tell like from the user
to pick from one of these options or
So when you instruct the agent,
to pick from one of the options
automatically, you will be able to see
that the buttons is always used,
by the reasoner
to be able to render consistently.
So that is one way
you can force, specific, control.
Similarly,
if you don't want to use a button right.
Let's say you just want to show
a bulleted list, right?
And you don't want to see
this rich content.
You can again control that,
by the instruction.
So that's where you could,
say something like polite asking the user
to make a selection or refrain,
from asking the user to select an option.
Those kind of, instructions
will make sure that you and,
the reasoner automatically does not show
these options as buttons.
they will show that as a plain text
or bulleted text as you wanted to see it.
So. So that is the trick here.
And I do want to thank everyone
for joining us today.
And don't forget to keep your questions
and suggestions coming
via the idea exchange,
and we'll be back for the next release.
for more information on releases,
go to Salesforce.com slash releases.
Thanks again and we look forward to
seeing you next time.