It's no secret that generative
AI is taking the world by storm,
and the opportunities made possible
by this technology are still developing.
We've only scratched the surface
of what businesses can do
when they embrace AI, but the truth is
it's already evolving faster
than anything we've ever seen before.
We've all heard influencers and thought
leaders say you need generative
AI or risk being left behind.
But that's easier said than done.
What’s good? I'm Bruce and welcome back
to “Salesforce on Salesforce.”
In this episode, Nahid Samsami, Senior
Director of Product Management
for Einstein Products, is here to share
five tips
for how you can get your organization
ready for generative AI now.
With examples
of how we've done it here at Salesforce,
you'll want to take notes.
So if you're ready, let's get into
it. Nahid, over to you.
we've been on a ten year journey
bringing artificial intelligence to CRM.
And one of the biggest questions we're
getting from our customers these days is
what should our company be doing today
to get value from generative AI?
It's a great question
and we have some thoughts.
Tip number one, “Start Experimenting
to Understand the Technology”
If we're going to be ready
for generative AI, then we need to start
with a solid understanding of how it works
and what it can do.
here is how generative AI works.
We start with a prompt,
we send that prompt to a large language
model, and we get a result.
Those are the three basic parts.
And why is this important?
Well, this basic framework
can guide our experimentation.
For example, let's start with that prompt.
What is the prompt and how does the data
we put into the prompt impact
Next, Where are we sending the prompt?
How do different prompts behave
with different large language models?
Finally, are we getting a good result
and how are we going to safely integrate
these results into our existing tools
and processes?
we've learned you can do a lot
with a great prompt
and we can use CRM data to enhance
those prompts.
The better the prompt,
the better the result.
And in many cases we don't need a custom
LLM if we have a great prompt.
But don't take my word for it.
Started experimenting
to understand the technology.
Tip number two, “Identify your Use Cases”
So what is generative AI good at?
Well, it's good at generating new content
based on a prompt.
It's good at translating content
and it's good at summarizing.
These are generative AI's superpowers,
which means we need to look for use cases
that align with those core capabilities.
Here are a few examples
that we found for our own products.
We can use generative AI to generate
personalized outreach emails.
We can also use it to summarize calls
we've just had with customers,
and we can translate developer comments
into working code.
Generation, summarization, translation.
Those are three
great use cases for generative AI,
so look for your own use cases that align
with these capabilities,
then experiment with different prompts
and different LLMs to see which produces
With this knowledge, you'll not only
identify your best use cases,
but you'll also identify the data
that's required by those use cases.
Tip number three, “Get your Data Ready for AI”
Through experimentation,
you'll find a generic prompt
leads to a generic response,
whereas grounding the model
by including more context into the prompt
will produce a more relevant result.
For example, if I say, “write me an outreach email.”
I'll get a generic and unusable result.
But if I add context about me, my company,
our product, our customer,
now I'll get a much better result.
But let's face it, data is messy.
It lives in different places.
It comes in different formats.
What's our strategy
for harmonizing this data
and using it to get better results
from generative AI?
Here at Salesforce, it starts with
building that single view of the customer,
which means we need solutions
for connecting all of our data,
harmonizing all of that data
and then building tools
that enable us to easily embed that data
into our generative AI prompts.
Tip number four, “Evaluate the Risks”
Generative AI involves
creating a prompt, passing
that prompt to a large language model
and then getting a result.
And there are risks to be considered
across all three of these dimensions.
Here is how we think about this
We already saw that
by putting more context
into the prompt we get a better result.
But some of that data might be PII
that we need to protect
for regulatory reasons, which means
we need a way of masking that data.
Then we need to think about where
we're sending that prompt
and how we'll protect the data
used in the prompt.
In our case, we have zero data retention
agreements with our AI partners,
which means all of this data
is instantly forgotten
and never stored outside the Salesforce
trust boundary.
Finally, there is the result.
Generative AI is impressive,
but it is not yet perfect.
It can hallucinate
and there is the possibility
of toxicity and bias.
To confront these risks
we've created automated tools
that evaluate
each response and provide toxicity scores,
and we ensure there is always a human
in the loop who can monitor the system.
Tip number five, “Integrate and Refine the Result”
The goal of this entire journey
is to get real value from generative AI,
which means we need to integrate
these results into our business processes
and measure the impact.
Well, at Salesforce,
we map prompts to actions.
So in this case,
we're mapping an email generation prompt
to a Salesforce action, which is why
we can not only select that prompt
but also integrate the results into our
tools. We’ll then use these signals
to further tune the system
and get more value from generative AI.
So these technologies are evolving fast,
but as you've seen, there is a lot
we can do today
to prepare our organizations
to get value from generative AI.
Thanks for watching this episode
of “Salesforce on Salesforce.”
If you enjoyed this episode
and you want to continue learning,
be sure to check out the “Take
a Deeper Dive” section down below.
Lastly, if you're low on time,
but you want to watch
the other episodes in the series,
try out the “My List” feature on Salesforce
Plus, just hover over the episodes
you want to watch later
and add it to your list.
Thanks for watching
and we'll see you next time.
