From where to start the learning Journey: Trailmix, Trailmix, and more Trailmix!

That
should be your mandatory starting points – which has very rich set of theory, video learning
followed by hands on project. Make a most use of playground Orgs which you’ll get access
though this course.

https://trailhead.salesforce.com/content/learn/trails/drive-productivity-with-einstein-ai

Why you should not look other options to pass this certification: Most of us are aware
that many in our community clear Salesforce certifications using a few sets of questions
and answers. However, this particular exam is very unique. It is not merely an exam to
pass; rather, it serves as an entry point to explore the vast possibilities within Salesforce's
AI capabilities. I am confident that sooner or later, our clients will be interested in
implementing many of the features covered in each section. Therefore, it is crucial not only
to pass this exam but also to gain a comprehensive understanding of the material.

Returning to the topics, let's delve into the diNerent sections and their weightage, which
will ultimately help us clear the exam while also enhancing our understanding. The exam
consists of 60 questions, with a passing criterion of 70%, meaning you need to answer at
least 42 questions correctly.

Section Weighting No of Questions

Einstein Trust Layer 15% 9
Generative AI in CRM Applications 17% 10
Prompt Builder 37% 22
Einstein Copilot 23% 14
Model Builder 8% 5

1. Einstein Trust Layer (15%): You will get 9 questions from here and have to focus into
following areas.

Prompt Journey

i) Prompt: which comprise of clear instructions, contextual information, and

constraints help which is being sent to LLM.

Secure Data Retrieval with Dynamic Grounding:

ii) Relevant info from Salesforce Record is merged with prompt template
iii) Respects user permission, role hierarchy-based access, field level security.
 Data Masking: Sensitive data is masked; it supports multiple regions & languages.
You can change the settings of masking.

(i) How to enable data masking?

(ii) Where you can see whether masking is working as expected or not?
(iii) What kind of data retention policy Salesforce follows while getting
response from LLM?
(iv) When the data got demasked and how it’s shows to prompt user?

Prompt Defense: System policies help limit hallucinations and decrease the
likelihood of unintended or harmful outputs by the LLM.

(i) What is the role of prompt defense and on which stage in prompt life
cycle it got activated?

Response Journey

(ii) Secure LLM Gateway: Prompt to leave secure trust boundary by

passing through the Secure LLM Gateway to connected LLM.

Zero Data Retention: Salesforce partnered with Open AI and Azure Open AI to
enforce the policy:

(i) No data is used for LLM model training or product improvements by

third-party LLMs.
(ii) No data is retained by the third-party LLMs.
(iii) No human being at the third-party provider looks at data sent to their
LLM.

Toxicity detection and demasking:

I. It removes toxic, hateful, violent, sexual, identifiable, physical, and

profane responses. It dehydrates same tokenized data.
II. Toxicity scoring: Toxicity scores are logged and stored in Data
Cloud as part of the audit trail. The score for safety category
ranges from 0 to 1, with 1 being safest.
III. How responses will be shown if any response has 0 Toxicity score?
Auditing:

(i) Where this audit data stores? How to access this audit data?
(ii) Prompts, responses, and trust signals are logged and stored in Data
Cloud.
(iii) Feedback can be used for improving prompt templates.
(iv) Provides pre-built reports and dashboards for analysis.
(v) Do you need to have Data Cloud to be provisioned or not to store audit
data?
(vi) Enrich event logs with conversation data: setting is enabled in
Einstein Copilot, it allows an AI Specialist or admin to view session
data, including both the user input and copilot responses from
interactions over the past 7 days. This data is crucial for monitoring
how the copilot is being used, analyzing its performance, and
improving future interactions based on past inputs. This setting
enriches the event logs with detailed conversational data for better
insights into the interaction history, helping AI specialists track AI
behavior and user engagement.
2. Generative AI in CRM Applications (17%): This section is particularly challenging for me as
a non-CRM professional. Additionally, the questions related to AI in Sales and Service add
an extra layer of complexity. You will get 10 questions from this section and your target
should have at least 6-7 correct answer among those.

Following topics to focus here: -

1. Prepare with Einstein Copilot: for meetings by researching prospects,

summarizing records, and delivering recommendations for planning your day. It
provides selling guidance and close plans, auto-updates CRM data, and more.

2. Create AI-powered generative emails: Automatically generate personalized

and impactful emails at scale for prospecting, following up on meetings, and
sending proposals.
3. Leverage AI-powered call summaries and call support: Enable your
representatives to save time, enhance follow-up, and maximize the value of
every interaction with AI-powered insights, summaries, automated follow-ups,
and sales signals. Key mentions, transcripts, and action items are automatically
provided from recorded calls.
4. Build data foundation: Utilize automatic activity capture, opportunity and lead
scoring, predictive forecasting, account insights, and deal insights to make data-
driven decisions.
5. Coach Sellers: ONer instant and actionable feedback through immersive
simulations and personalized coaching programs.
6. Guide selling and grow relationships: Employ propensity to buy, next-best
opportunity, and product recommendations along with relationship maps,
recommended connections, and email insights to sell the right solution to the
right person at the right time.
7. Einstein Case Classification: Predict field values such as Priority, Reason, or
Type for classifying incoming cases based on the text provided by the customer
in the case Subject and Description.
8. Einstein Work Summaries: Generate a summary, issue, and resolution using
generative AI based on a chat conversation between an agent and a customer.
9. Einstein Case Wrap Up: Enable chat agents to complete cases quickly, with
greater accuracy and consistency. Check the diNerence between case wrap up
and work summaries.
10. Einstein Conversation Mining: Transform conversation data into service
insights and develop bot intents.
11. Einstein Service Replies: Use generative AI to draft and recommend fluent,
courteous, and relevant replies for your agents to review, edit, and post.
12. Einstein Reply Recommendations: Analyze chat transcripts to recommend
relevant replies during chat and messaging sessions.

Few areas where I got confused in 1st attempt from these sections -
1. What is the di,erent among Call Explorer, Call Insights, Call Summary.
2. Di,erent between Einstein Reply Recommendations and Einstein Sales
Emails. Which scenarios needs to use what?
3. What syntax needs to write in prompt to instruct AI to adhere with strict
guideline?
4. What is the role of Einstein Reply Recommendations and work
classification? In which scenarios it can be used?
5. Di,erence among Find Similar Deals, Create Account Plan and Create
Close plan regarding Al-generated plans containing guidance and steps
for closing deals
6. Which capability to access knowledge-based article and what are the
di,erent scenarios to use those for service agent.
7. Scenarios of creating Template-Triggered Prompt Flow.
8. Scenarios of creating Template-Triggered Prompt Flow.
9. Saving and versioning of prompt templates.
10. Enabling prompt builder for users.
11. Permission to create prompt template and use existing prompt template.
Check di,erence between Data Cloud Admin, Prompt Template
Manager and Prompt Template User
3. Prompt Builder (37%): This section should be your primary focus as it is relatively easy to
score well. With 22 questions in this section, aim to answer at least 18-20 of them
correctly.

Following topics to focus here: -

(i) DiNerent Types of prompt template.

(ii) What is the use-cases and usable objects of those prompt template.
(iii) Role of Prompt template Manager and User.
(iv) How to be grounding prompt template and which merge field needs to
use on which use-case cases.
(v) Unsupported objects to use related list during prompt grounding.
(vi) Best Practices for designing eNective prompt: Be clear and concise,
include context and directive, consider output, Provide Example
response(s), Break up complex task, Experiment and Refine
(vii) Prompt templates limits: You can expect at least 1-2 questions from here.

(viii) DiNerent types of Prompt Templates:

https://help.salesforce.com/s/articleView?id=sf.prompt_builder_standard_te
mplate_types.htm&type=5

(ix) How it works:

 Most of questions covered the following topics:

1. Standard prompt templates

2. Ground templates utilizing Salesforce resources:
1. Record merge
2. Flow merge
3. Apex merge
4. Related list field merge
3. Limits and limitations of the Prompt Builder
4. Scenarios for creating Template-Triggered Prompt Flows
5. Saving and previewing templates
6. Troubleshooting when field values do not display
7. Saving and versioning of prompt templates
8. Enabling the Prompt Builder for users

4. Einstein Copilot (23%): Einstein Copilot is a reliable conversational AI assistant for CRM,
seamlessly integrated into the Salesforce interface. It processes questions and requests in
natural language, delivering relevant answers sourced from secure, proprietary company
data. This section will contain 14 questions, and your goal should be to answer at least 12
of them correctly.

Following topics to focus here: -

(i) What is Co-Pilot Studio, how can it be enabled, and how to access it?
Refer to the two pictures below for an overview of each component of Co-
Pilot Studio to gain a comprehensive understanding.
(ii) Most importantly, these two pictures below were invaluable for me. It
provides a comprehensive understanding of the entire Einstein Co-Pilot,
preparing you to handle client questions eNectively.
5. Model Builder (8%): I really enjoyed this section, not just because of the BYOM (Bring Your
Own Model) aspect, but also because I had worked on it for a client before, giving me some
useful experience. This helped me score 5 out of 5 on both attempts. However, if you
haven't explored this before, keep the following points in mind for the exam.

Following topics to focus here: -

(iii) What is Einstein Studio and Which cloud will have this capability?
(iv) What are the diNerent options to build or bring the model?
(v) What is the diNerence between Regression models & Binary
Classification model.
(vi) What kind of permission set is required to access this studio.
(vii) What is ‘Build your Model from Scratch’ and it’s use-cases?
(viii) Last but not the least, below picture to get supported external models.
 Cheat Tips: Questions which I can remembered on –

1. How does the Einstein Trust Layer ensure that sensitive data is protected
while generating useful and meaningful responses?

2. Which step must needs take to make trust layer available to Org?
3. How should an organization use the Einstein Trust layer to audit, track, and
view masked data?
4. When the data get masked and when it got de-masked?
5. What is the data retention policy for LLM?
6. What is toxicity score means and what is the range of this score?
7. a copilot custom action using flow as the reference action type. However, it is
not delivering the expected results to the conversation preview, and therefore
needs troubleshooting.
8. How sales team to get insights into product and competitor names
mentioned during calls. And what the limit of total numbers of products to be
shown.
9. What is the role of the large language model (LLM) in executing an Einstein
Copilot Action?
10. How does Einstein co-pilot deletes any activities?
11. How to ground prompt template with Data Cloud’s data? Which types of
flows need to use.
12. What the role of plan in co-pilot life cycle.
13. What kind of permission set is required if a Data Scientist wants to create a
prompt template that needs to be grounded with external model data?

Statuary Warning: This is neither a dumps paper nor gives you guarantee to pass the
Salesforce AI Specialist Exam. But it will enhance your skill in the era of AI and boost your
confidence if you read this article post completing the below trail mix.

https://trailhead.salesforce.com/content/learn/trails/drive-productivity-with-einstein-ai