Blueprint for

Implementing AI
Underpin and amplify your AI capabilities
with best practices from MuleSoft.
Contents
Introduction 3

What is an AI app? 4

Jumpstarting your AI journey 6

Barriers to AI adoption 7

Four-step blueprint for implementing AI 8

1 Connect 9

2 Compose 10

3 Control 11

4 Action 12

AI success story: Iron Mountain 13

Conclusion 14
Introduction
AI creates visions of groundbreaking possibilities and innovations from healthcare to finance,
retail, and manufacturing. With 86% of IT leaders anticipating that generative AI will strongly
influence their digital transformation strategies, it’s clear that most are expecting AI to drive
monumental change across their organizations.1

The possibilities with AI today are vast. Generative AI and predictive analytics have allowed
us to personalize user experiences at an unprecedented scale, enhance decision-making
processes, and automate repetitive tasks across lines of business. Now, with the emergence
of capabilities like autonomous agents, apps can be built to understand complex instructions,
take action on the user’s behalf, and even recall past conversations.

Capabilities like these have the potential to revolutionize not just marketing, sales, and service,
but also logistics, finance, HR, and beyond. Think of it as giving every employee a personalized
digital helper who can respond to more than a single prompt. They can take directions in natural
language, create a list of next actions, and perform tasks independently so teams have more
time to do what humans do best.

And while AI sparks boundless imagination and potential, certain realities must be confronted first.

Today’s IT ecosystems weren’t designed to handle building and managing AI apps at scale.
To deploy intelligent automation with autonomous agents, you must design an IT ecosystem

3
to be able to handle the complexities of AI — and then extend its reach across your organization.

86% of IT leaders anticipate that

generative AI will strongly influence
their digital transformation strategies.

1
Salesforce State of IT, July 2023
What is an AI app?
There’s a lot of information about what an AI app is and how you should build one. Before
we delve into the specifics of setting up your IT infrastructure for AI, it’s important to understand
what an AI application consists of so you can properly strategize to address each component.

We’ve boiled AI apps down to their most essential components: the data, the user interface,
the models, and the outcomes. As you continue your AI journey, keep these elements
at the forefront of your strategy to ensure the best possible results from your AI initiatives.

Data
Data fuels AI by providing the context and information necessary for making informed
decisions. High-quality, diverse, and relevant data is essential for developing accurate
and useful AI solutions — and access to that data empowers AI models with the
best raw material they can have.

User interface (UI)

The UI is the gateway for users to interact with AI-powered systems, whether they are
customers seeking assistance or employees accessing internal tools. A well-designed

4
UI makes it easier for users to engage effortlessly with AI and make the most
of their capabilities.

Models
At the heart of AI apps lie the models, encompassing the algorithms, techniques, and
architectures that enable AI systems to process data, generate insights, and develop
a plan of action. In the case of large language models (LLMs), deep learning techniques
are used to understand and generate human-like text. However, to maintain trust and
compliance, organizations must be able to make their data available to these models
without compromising privacy or security.

Outcomes
In the end, every AI app must somehow improve the end user’s experience. Whether
fetching or updating data in real time, generating dynamic action plans, or executing
tasks autonomously, AI apps must deliver tangible value for the user through more
relevant answers, automated workflows, and actionable plans.
These elements come together to deliver AI apps that create value across your organization.
Discovering and implementing these elements might sound like a big undertaking, but luckily
you don’t have to do it on your own. The Salesforce Platform provides a head start for your
AI journey, bringing data, UI, models, and outcomes together for an effective AI foundation.

What makes Salesforce the #1 AI CRM?

The Salesforce Platform brings apps and data together on one integrated platform

5
to deliver AI-powered experiences across customer relationship management (CRM)
and beyond so organizations can automate and action insights across every application.

MuleSoft, Data Cloud, and Salesforce’s predictive and generative AI capabilities

are key components to helping organizations integrate, harmonize, and activate
all of their data securely to deliver next-generation AI initiatives.
Jumpstarting your AI journey
Any time you start a complex digital project, there’s no benefit to starting from zero if you don’t
have to. The Salesforce Platform offers a head start to AI app development by embedding data, AI,
CRM, and security into one comprehensive platform. It provides IT, admins, and developers with
an adaptable AI platform for rapidly (and securely) developing generative AI apps and automation.

The Salesforce Platform helps organizations hit the ground running with their AI projects with
out-of-the-box functionalities that make it easier to implement AI.

For example, Salesforce houses critical customer data needed for AI. With Data Cloud, organizations
can unify and augment data from all their CRM instances, providing valuable insights and a fuller
understanding of their clientele.

Salesforce also enables intuitive interaction with and actioning of AI capabilities with Agentforce
as a user-friendly interface. Agentforce is a set of tools to help helps users create and customize
AI agents — both autonomous and assistive — without requiring deep technical expertise.

Meanwhile, Salesforce’s Trust Layer offers native integration with models to ensure the safety
and security of sensitive information. Organizations can use pre-trained models or bring in their
own — whichever approach is needed to meet their specific business needs and requirements.

Whatever your objective is for your AI project, Salesforce lets you produce high-impact outcomes

6
across domains like sales, service, and marketing. Plus, the integration and automation capabilities
from MuleSoft help you action outcomes to any system or application across your organization.

From helping organizations manage customer relationships, forecasting demand, and providing
real-time service chat support, you can use the Salesforce Platform to drive business growth
and increase customer satisfaction. By embedding data, UI, models, and outcomes within
the Salesforce Platform, Salesforce equips organizations with the tools and capabilities needed
to get up and running quickly with AI.
Barriers to AI adoption
The Salesforce Platform greatly accelerates AI endeavors across CRM. However, activating data,
UIs, models, and outcomes outside of CRM — that reach across any application or workflow
in your organization — requires added attention.

Although Salesforce includes access to customer data in CRM, organizations likely have
a lot of other data scattered across hundreds of external systems. Shockingly, the average
enterprise uses nearly 1,000 systems. And often, connecting all that data is a top barrier
to AI implementation — 95% of decision-makers say their main roadblock to AI adoption
is integration.2

Disconnected systems frequently result in poor data access, quality, and structure inconsistencies.
Employees may find navigating and interacting with multiple user interfaces for different
applications challenging. Additionally, managing and maintaining numerous third-party apps
or user interfaces can be burdensome for IT departments, as updates, patches, and security
fixes must be applied to each application individually, increasing the risk of errors, vulnerabilities,
and downtime.

Meanwhile, the rapid proliferation of AI models (from just 30 commercially significant ones a year
ago to over 500,000 unique models available today) poses significant challenges for AI adoption
for the average enterprise.3

7
With such a vast array of AI models to choose from, many struggle to navigate the options and
select the most suitable model for their specific use case. Integrating multiple AI models into
existing systems or building your own model can be complex and time-consuming, with each
model having its own requirements, dependencies, and compatibility issues.

Lastly, many AI use cases will need the ability to extend beyond CRM. If a desired outcome
lives outside of Salesforce — for instance, your ERP — integration is needed to start an action
in one system and continue actioning outcomes in another.

For example, imagine an autonomous agent app that begins its actions in CRM by analyzing
customer data, sales trends, and interactions to predict future sales. It then takes these predictions
to the ERP system to automatically adjust procurement, inventory levels, and financial forecasts
accordingly. This kind of seamless integration between systems is necessary to enable a fully
end-to-end, hands-free sequence of actions for your AI apps.

2
MuleSoft 2024 Connectivity Benchmark Report
3
Hugging Face
Four-step blueprint
for implementing AI
As a critical component of the Salesforce Platform, MuleSoft extends the platform’s capabilities to
systems beyond your CRM so you can realize greater success with your AI initiatives. To help direct
you on your journey, we’ve created a holistic four-step approach to AI success, using APIs to help
you unlock your data and automate actions across every system and workflow in your ecosystem.

8
1 Connect all of your data for seamless access
Embedded in the Salesforce Platform, Data Cloud connects and harmonizes customer
data in Salesforce, eliminating data silos that would traditionally prevent AI applications
from accessing relevant data. Data Cloud offers native connectivity for all first-party
data in Salesforce and major cloud storage solutions like Amazon S3, Google Cloud,
and Microsoft Azure. It also allows for zero-copy data sharing and federation for major
data lakes and warehouses including Snowflake, Databricks, Google BigQuery, and select
SaaS applications.

Although your CRM data is valuable, it only provides a snapshot of one facet of your
organization. To build truly versatile AI apps, you also need to be able to incorporate
data about products, transactions, consumer behavior, and more. In addition to CRM
data, MuleSoft seamlessly unlocks data from other systems, including ERP, HR, Finance,
and homegrown or legacy systems to ensure the necessary information is accessible
and available for your AI applications.

Establishing seamless data flows from these systems allows you to feed your AI algorithms
with the most accurate, accurate, and up-to-date information — a process known as
grounding. Grounding helps ensure the AI apps you build are connected to reality in
a meaningful way. It involves connecting AI’s abstract concepts and algorithms to diverse,
correct, and current information from the real world. This allows AI to understand and
interact with the environment in a relevant and appropriate manner. Without grounding,

9
your AI systems might operate on outdated or irrelevant information, leading to errors
and unhelpful interactions.

Extending Data Cloud capabilities with MuleSoft

If you’re like most companies, you have essential data residing within industry-
specific systems such as FIS and Epic, across legacy platforms like SAP and
Oracle, in custom-built solutions like mainframe systems, and other cloud-
based applications.

Using APIs to connect disparate systems and applications, MuleSoft helps bring
all of your data — no matter where it resides — into Data Cloud so you can ground
your AI in diverse datasets and fuel more comprehensive outcomes across any
line of business.
2 Compose your data for specific
business functions and user interfaces
Unlocking your data and making it accessible serves the same role as the first step to cooking
a gourmet meal. The chef gathers all the necessary ingredients in the kitchen, making the
ingredients accessible for when they’re needed in the cooking process. However, without
the right tools, even the best chef would have difficulty turning those ingredients into
the ultimate gourmet dish.

Similarly, after unlocking your data to make it accessible, it’s critical to have an architecture
that lets you transform and serve that data to its destination — the user interface. By
composing your data in an API-led architecture and taking a three-layered approach
to organizing your IT APIs and assets, you enable the most efficient use of your data
for the specific business applications you’ve chosen for it, including AI apps. MuleSoft’s
Anypoint Platform is the leading platform for API-led connectivity.

Three-layered connectivity:
System, Process, and Experience APIs
System APIs
These APIs unlock the data stored within an organization’s core systems

10
of record. Examples of critical systems from which APIs can unlock data
include ERP, customer and billing systems, and proprietary databases.

Process APIs
By composing data surfaced from System APIs, Process APIs aggregate
and shape data for specific business purposes, such as gathering customer
or employee information or order details. Process APIs facilitate combining
data and orchestrating multiple System APIs for targeted business objectives —
for example, creating a 360-degree view of the customer, managing order
fulfillment, or tracking shipment status.

Experience APIs
Experience APIs expose data and logic to applications or interfaces. They
provide a business context for the data and processes unlocked and established
by System and Process APIs. Experience APIs ensure that the exposed data
is consumable by its intended audience, such as the AI user interface.
3 Control your data to enable trusted AI
Once your data has been unlocked and efficiently applied to an AI project, you need
a way to ensure reliability and security. You must be able to govern and secure your APIs
and the data used in your AI project to ensure sensitive information doesn’t become
exposed to unauthorized individuals or public AI models.

You can effectively control access and prevent sensitive information from being shared
with LLMs using an API management solution like MuleSoft, which provides several
key benefits:

Continued reliability: Once your AI app is operational, it needs continuous access to data
to power consistently useful output. MuleSoft APIs provide reliable integrations across
systems, processes, and experiences, so AI apps like autonomous agents have consistent
access to the data they need when they need it.

Governance and security: API governance ensures that APIs are managed effectively
by setting access controls, authentication mechanisms, and usage policies. It also allows
you to monitor API performance, usage patterns, and potential vulnerabilities to ensure
ongoing security.

Sensitive information protection: AI often involves sensitive data, including customer

records, financial transactions, health information, and so on. API management helps
prevent accidental or malicious leaks that could have severe consequences.

11
Organizational reputation and compliance: Compliance with data protection laws
(like GDPR) is non-negotiable. API management helps centralize and control privacy
and security requirements.
4 Action AI recommendations across your business
The framework’s final step is the ability to act on the insights, resolutions, or workflows
recommended by your AI application to drive the best and most relevant outcomes.

The reality is that many desired outcomes and actions will exist in numerous systems
that extend beyond Salesforce. While insights related to customers and marketing exist
in Salesforce, workflows attached to an e-commerce process, for instance, may happen
in other systems.

Suppose a customer purchases a product recommended by AI. In this case, cascading

events occur across different platforms and systems. The purchase needs to be sent
to the order management system and logged in the ERP. It then needs to be sent to the
order fulfillment system and posted to the accounting system when the product ships.

To enable a process like this, you need a way for your AI apps to access other systems
and trigger the next actions based on previous events. This capability allows AI apps like
autonomous agents to perform actions on your behalf, even if the data and workflows
involved reside in systems outside your CRM.

That’s where MuleSoft APIs and bots can facilitate real-time data movement based on events
that occur throughout this process end-to-end.

12
Extending Agentforce capabilities with MuleSoft
Agentforce is a suite of customizable agents and tools that seamlessly integrate
with an organization’s data and apps. Acting as a trusted advisor and helper,
it provides quick answers to the user’s queries and acts on their behalf, saving
them time and making them more productive. While Agentforce supports actions
within Salesforce, MuleSoft gives it the ability to act outside of Salesforce
as well, using APIs and robotic process automation (RPA) bots.

For example, you could ask Agentforce to update an invoice amount in your
ERP or place an order in your order management system. MuleSoft would then
enable Agentforce to reach into these external systems and perform the task
that instant, without requiring you to leave your workflow in Salesforce.
Applying the four steps
These four steps — connect, compose, control, and action — comprise successful AI
implementation. Through these four steps, you can ensure the accessibility, integration, security,
and reliability of your information from as many disparate sources as possible, while empowering
your AI and team to turn that data into actionable insights.

Let’s explore how one company used MuleSoft to integrate

its IT ecosystem and fuel AI innovation with Salesforce Platform.

AI success story:
Iron Mountain is a global information management services company that serves more than
225,000 organizations around the world, including 95% of the Fortune 1000. With a mission
to protect and elevate the power of its customers’ work and with a customer base of this size
and scale, providing an exceptional customer experience is a top priority.

Despite having massive amounts of customer data, Iron Mountain’s service representatives
couldn’t easily access it or use it meaningfully to address customer requests. Service reps
had to switch between multiple apps and databases to gather enough data for customer

13
inquiries. Plus, the order process was manual and information was located across numerous
places, forcing them to go back and forth between systems.

Iron Mountain turned to Salesforce technology to improve service representatives’

experience. It helped the company:
• Close cases faster: Using generative AI to automatically write personalized case replies
based on past cases and the knowledge base, service reps can more quickly find the relevant
information to close out cases.

• Reduce administrative work: Using Work Summaries, service reps can also move on from
a case more quickly, thanks to Work Summaries powered by the Salesforce Platform.
When a case is complete, Agentforce Service Agents automatically generate a summary
based on the case data, conversation, and transcript history, reducing the administrative
load that service reps previously managed.

• Integrate workflows: Iron Mountain uses the Salesforce Platform to pull the right data for each
case. This helps generate the right responses from AI and fully integrates their service workflows
directly in Service Cloud. Behind the scenes, connected data fuels an improved experience for
service reps. MuleSoft brings together account, billing, and engagement data all in one place,
so that orders can be automated and service reps have a holistic view of each customer.

Learn more on how Iron Mountain increases service representative productivity by unifying data.
Conclusion
AI apps like autonomous agents are set to revolutionize organizations’ operations, enabling
adaptive decision-making, automating workflows, and freeing employees for tasks requiring
a human touch. Before building one, however, you need an IT foundation that provides
unobstructed, secure access to data no matter where it resides.

MuleSoft makes this possible by removing barriers between data sources and streamlining
flows between CRMs and other platforms with robust API connectivity. From gathering and
shaping data to presenting it in the most powerful way possible, MuleSoft radically empowers
you to achieve more with your data and embrace building with AI.

Discover Webinar
Explore how the Salesforce Platform facilitates Find out how MuleSoft and Salesforce tools
fast AI development and automation. work together to accelerate innovation.

Explore the Salesforce Platform Watch the webinar

14
Blog Demo
Empower Agentforce to Act Across Learn how Anypoint Flex Gateway helps you
the Enterprise With MuleSoft APIs. effectively control and manage LLM APIs.

Read the blog Watch demo

Salesforce, the global CRM leader, empowers companies of every size and industry to digitally transform
and create a 360° view of their customers. For more information about Salesforce (NYSE: CRM),
visit Salesforce.com.

Any unreleased services or features referenced in this or other press releases or public statements are
not currently available and may not be delivered on time or at all. Customers who purchase Salesforce
applications should make their purchase decisions based upon features that are currently available.
Salesforce has headquarters in San Francisco, with offices in Europe and Asia, and trades on the New York
Stock Exchange under the ticker symbol “CRM”.

For more information please visit Salesforce.com, or call 1-800-No-Software.

MuleSoft is a registered trademark of MuleSoft, inc., a Salesforce company. All other marks are those
of respective owners.

About MuleSoft

Almost two decades ago, MuleSoft was the first to introduce the world to API-led connectivity. Today, we’re
a trusted solution in multiple categories. We power 7.56 billion customer integration transactions daily.
Additionally, we’ve been named a Leader in the Gartner Magic Quadrants for iPaaS, API Management,
and Low-Code Application Platform. Salesforce acquired MuleSoft in 2018.