To really make
all of this work,
we have to
understand how do we
get the data to these
agents, to agent for us?
How do we bring that
data together in a way
that they can
actually use it?
And the good news is
we've got the world's best
person here to
explain that to us
Please welcome to
talk about Data Cloud,
our VP of product
marketing, Sanjna
Thank you so
much, Patrick.
Now, you've seen
Agentforce in action,
but data is really
the beating heart
of how Agentforce does
all of that cool stuff you
just saw Sophie do in that
interaction with Patrick.
But in order to fully
understand its power,
I want to debunk
a little myth here
for everyone today
on how AI can
Now, what a lot
of folks think
is that I need to
train an LLM with all
of my business's
data in order for AI
And that's just
simply not true.
It requires specialized
developers and data
And more importantly,
as your data changes
and as dynamic as
your customers are,
the AI doesn't change with
it without retraining.
So this really
isn't the way
that we're going to get AI
to work for our business.
Instead, we're going
to flip the script here
and put the power
inside of our prompt.
Now, a prompt is nothing
but a set of instructions
and questions that we send
to a large language model
to teach it about
our business.
So everything you
need for an LLM
to know about
your business
can reside
within a prompt.
But what we
also don't want
is a prompt that's
hundreds and hundreds
of pages long
because it has
all of that context
that we want
So what we want
to employ here
is a new technique called
Retrieval-Augmented
Now, what RAG
does is it enables
us to put a ton of
context into that
prompt that searches
through all of the data
that I have
available to me
and retrieve that
data into the prompt
and teach our LLM
about our business.
This is how we get
AI to work for us
So you see here
that we don't
need to do any DIY
projects to get
AI to work for us, right?
But I know what
you're thinking.
It sounds great to
get all of this data
into an agent,
but my data is
sitting in totally
different systems.
We know that all
of our customers
have islands of
disconnected data sitting
everywhere, whether
that's in an external data
lake, several different
Salesforce clouds, or even
some legacy systems
that might not even
We understand this
pain very, very well.
And it's actually
the inspiration
behind why we created Data
Cloud in the first place.
And it's our
hyperscale data engine
that resides deeply
inside of the Salesforce
And the mission we had
in mind with Data Cloud
We want to enable you
to put all of your data
to work, no matter what
kind of data it is--
telemetry data,
health care data,
We want you to put that
data to work for a better
Now, to really understand
the magic of Data Cloud,
we got to talk about how
it works under the hood,
And we make it really
easy to connect
And we have a few
of these logos
here, whether it's
your Salesforce clouds
or different Salesforce
orgs that you have
It could be external
applications.
It could be
external data lakes,
like Snowflake
or Databricks.
Or, it could
even be systems
that are storing your
unstructured data,
like your PDFs, your
voice data, audio files,
video files, anything
you can think of.
Now, once you've
connected to this data,
we do this in a really
special way using our zero
Because we don't
want all of you
to incur the
cost of moving
your data every
time you want
to use it inside
of Salesforce,
And so what
Data Cloud does
is we virtualize that
data inside of Salesforce,
so you can use it
like any other object
and activate it across
the Customer 360.
Now, another vital
part of Data Cloud
is harmonization,
because once you've
connected to your
data, you still
have different versions of
the same customer residing
All of your data systems
speak different languages.
And what Data
Cloud does is
it enables them to all
speak the same language.
So you have that
central, unified view
of every single
customer that you
can activate in
any experience
You can govern this data.
You can govern
these profiles
with define policies
inside of Data Cloud.
And you can activate
this data, of course,
inside of an agent
or any other actions
that you want to drive
for a better customer
Now, Patrick mentioned the
Atlas reasoning engine.
And Data Cloud is
really at the heart
of the way that
agents reason
And retrieval-augmented
generation
is such a vital
part of this.
And we're going to show
you this in action in just
a couple of minutes here.
Now, Data Cloud really
is at the heart of what
And I think the easiest
way to think about it is
that so many of these
capabilities around
personalization,
cross-sell, upsell--
or even just
closing deals faster
with the right
information for your reps
it wouldn't be possible
without connecting
to all of those systems
with Data Cloud.
So I want to show you
this in action in a demo.
And I'm going to
have my friends,
Clare and Rebecca behind
the demo desk help me out.
Now, Patrick had his story
with Sophie and his sack
But I got a
story of my own.
And I think it's a
pretty relatable one.
I think we can all
relate to that feeling
when you buy an
expensive item.
And then it immediately
goes on sale
and you want
your money back--
a price adjustment
workflow.
Now, that type
of question,
if I ask it to
an agent here,
it might take a service
rep at Saks several hours
They'd have to know the
return policy practically
by heart to figure out
how to process my refund.
But the great thing
about Agentforce
is that a task that's
really complicated
for a service rep
is a perfect task
So why don't we
go ahead and see
how we can enable Sophie
with this new ability
to handle this
customer question?
Now, our journey here
starts, of course,
in Data Cloud,
in connecting
to the right
data that we want
to empower Sophie with
to solve this question.
Now, it looks
like we've already
gotten a head start here.
We've connected to our CRM
data, which is awesome.
But in order to really
answer this question,
I need my Order Management
system data that
sits inside of Snowflake.
So I'm going to go ahead
and create a new data
And Data Cloud makes
this really simple,
with all of these out
of the box connectors
for any data system
you can practically
So I'm going to go ahead
and select Snowflake
and create this new data
stream with my Order
And what you're
going to see here
is it's automatically
creating this data graph.
And it shows a
visual representation
of all of the
relationships
across my different data
systems to that individual
that we want to
solve this query for.
OK, so our next
step here is
going to pop back
into Agent Builder.
Now, Patrick
showed you this.
Agent Builder
is where we give
our Agentforce new
abilities, through topics,
actions and instructions.
So we already
have a topic here
around pricing questions.
And we've actually
drafted an action as well
And this action is
a prompt, right?
I told you before, the
power is in the prompt.
So why don't we go ahead
and open up Prompt Builder
and make this prompt a
little bit more powerful.
Now, Prompt
Builder has been
available to our Agent
Blazers for a year now.
And they're
absolutely loving it.
And they're loving
it for a couple
The first is
that we enable
you to select
any model you
want to use under the hood
with a simple dropdown.
You don't need
to write any code
or figure out a
complex evaluation.
All you have to do
is test it inside
Now, the second reason
that our Agent Blazers
love this is because this
is prompt engineering
It's completely low code.
So you can
create and revise
these prompts in
natural language.
So we've already
gotten a head
start here and paste it in
the draft of our prompt.
And I want to make
it a little bit more
intelligent with
some data that I
connected to
with Data Cloud
So I'm going to
go ahead and drag
in the name of the contact
to really ground that
prompt and the information
about my customer, as well
as some of that
Order Management data
that we just connected
to inside of Snowflake.
So when I drag
all this data in,
I can also test this
on a contact record.
So I'm going to test it
on my own contact record
here and see what the
response could look
So I want to call
your attention
to the resolution here
because I can actually
see how the prompt is
thinking and retrieving
the right data to
create that response
And I'd say, I'd give
this response like a B+.
It's giving me
some context.
It's giving me the
amount of the refund.
It's addressing
me by first name.
But I think we can do
a little bit better.
And we can do better
with retrieval-augmented
So what I've
done previously
is I've created what's
called a retriever.
Now, all a
retriever does is
it kind of does what
it sounds like it does,
which is retrieve the
right data into the prompt
So I'm going to drag in
my retriever for my return
And when I drag
that in, I also
want to give this
retriever some parameters
because I don't want
the LLM to waste time
searching through
every single document
every time I run
this prompt, right?
I want to give it
some specifications.
So I want it to
specifically look at price
Now, let's go ahead and
save this, and preview it
again, and see how we did.
OK, if we look in
the resolution,
we're going to see exactly
where this prompt is
looking for the
right context
and grounding our
prompt with it.
And our response
is much better.
So why don't we
see it actually
inside of the experience
that we started out with?
So I'm going to ask
my question to Sophie.
I want this
price adjustment.
I'm going to confirm which
product it is because I've
So we want to make sure
it's the right product
for the price adjustment.
And immediately,
what you're
going to see here is
that Sophie can not only
adjust the price
here, give me
the amount, all the
context that I need--
but can process
that refund for me
This is the power of Data
Cloud under the hood,
empowering us with the
right data for Agentforce
to really take action
for the customer.
And Saks was able
to do all of this
They were able to ground
their prompts, empower
Sophie with new
abilities, and do
all of this on
the platform
without training
a single model.
