Daily Learnings: Fri, Dec 27, 2024
We should all be thankful for those people who rekindle the inner spirit. — Albert Schweitzer
Salesforce AI Certification Exams
Salesforce has made all AI-based certification exams (Salesforce AI Associate & Salesforce AI Specialist) free through the end of the year, so I’ve set a goal to obtain these two certs before 2025. Today I spent about an hour studying for (and later taking) the AI Associate Exam. Here are my notes in preparation for the exam.
AI Associate Study Notes
AI Capabilities in CRM
- Einstein: the general bucket that all AI capabilities fall under
- A lot of the AI features are prefixed with “Einstein”
- 5 categories of AI tools
- Discover insights
- Predict outcomes
- Recommend Actions
- Automate Tasks
- Generate Content
- Einstein for Sales - benefits
- Boost win rates by prioritizing Leads & Opps most likely to convert
- Can be used to create Lead Scores in Sales Cloud
- Discover pipeline trends and take action by analyzing sales cycles with prepackaged best practices
- Maximize time spent selling by automating data capture
- Generate relevant outreach automatically with CRM data
- Proactive outreach seems interesting; in some of my webinars that I’ve attended, I heard that proactive AI outreach wasn’t supported yet…
- Boost win rates by prioritizing Leads & Opps most likely to convert
- Einstein for Service - benefits
- Accelerate case resolution by automatically predicting and populating fields on incoming cases
- Predicting and populating fields??
- Increase call deflection by resolving routine customer requests on real-time digital channels
- Used for case deflection in general through several different means
- Create tailored service replies, knowledge articles, and work summaries automatically with CRM data
- Practically, AI can do the following:
- Read through emails, perform case classification, and route Cases to the right people, all based on past inquiries
- Scan notes in real time and recommend relevant articles
- Accelerate case resolution by automatically predicting and populating fields on incoming cases
- Einstein for Marketing - benefits
- Know your audience more deeply by uncovering consumer insights and making predictions
- Engage more effectively by suggesting when and on which channels to reach out to customers
- Create personalized messages and content based on consumer preferences and intent
- Predictions that AI can help with:
- “Will a customer unsubscribe?”
- Some Einstein Products to consider
- Einstein Bots: Allow you build smart assistant into your customers’ channels like chat, messaging, or voice
- Use NLP
- Einstein Prediction Builder: Point-click wizard that allows you to make custom predictions on your non-encrypted Salesforce data
- E.g., “Will this customer attrit?”, “Will a customer miss a payment”
- Einstein Next Best Action: Allows you to use rules-based and predictive models to provide anyone in your business with intelligent, contextual recommendations and offers
- Einstein Discovery: Allows you to get a more full understanding of relevant patterns on all data in your company
- Needs at least 400 rows with outcome values to build a model, more than one driving variable
- The more relevant, historical data, the better
- Generative AI with Einstein: Allows businesses to generate personalized and relevant content by grounding LLMs in CRM data safely and securely, using NLP to help close deals and resolve service inquiries faster
- Einstein Engagement Frequency: Tool to help marketers know how often to reach out / right number of communications to send without going overboard, potentially causing an unsub to your marketing channel
- Einstein Predictive Forecasting: Helps Sales Managers to flag Opps that don’t line up with historical data stored in the CRM, giving them a place to focus on when coaching and meeting with their team
- Einstein Readiness Assessor: Can help you determine which tools have the pre-reqs in place so you know what you are closest to being ready for
- Einstein Bots: Allow you build smart assistant into your customers’ channels like chat, messaging, or voice
AI Fundamentals
- What is intelligence: The ability to acquire & apply knowledge and skills
- Specific kinds of AI that are good at specific kinds of tasks
- As of today: There are many different types of AI that are good at specific tasks, but no one AI that is great at all types of tasks
- Similar to humans today
- Types of AI/Categories
- Numeric Predictions
- Classifications
- “Is this plant edible or poisonous?”
- “Is this email spam or legit?”
- Robotic Navigation
- Ex. AI optimization of supply chain
- Language Processing (NLP)
- Ex. ChatGPT
- Key Terms
- Machine Learning: How computers can learn new things without being explicitly programmed how to do so
- Not specific algorithms
- Structured Data: Organized and formatted in a specific way, like in database tables or spreadsheets
- Unstructured Data: Not formatted in any specific way
- Supervised Learning: Model learns from explicit examples, usually prepared and reviewed by a human
- Unsupervised Learning: Process of letting AI find hidden patterns in your data without any guidance
- Neural Networks: computer system modeled on the human brain and nervous system.
- Utilizes interconnected nodes with multiple layers
- Natural Language Understanding (NLU) : Systems that handle communication between people and machines
- Processes data from unstructured to structured
- Natural Language Processing (NLP): Describes a machine’s ability to understand what humans mean when they speak as they naturally would to another human
- Is distinct from NLU
- Natural Language Generation (NLG): Data processed from structured to unstructured
- Named Entity Recognition (NER): labels sequences of words and picks out the important things like names, dates, and times
- Involves breaking apart a sentence into segments that a computer can understand and respond to quickly
- Deep Learning: Artificial neural networks being developed between data points in a large database
- Machine Learning: How computers can learn new things without being explicitly programmed how to do so
- Common Concerns about Generative AI
- Hallucinations
- Data Security
- Plagiarism
- User Spoofing
- Sustainability
Data for AI
- Data: Individual facts, statistics, or items of information
- Collection of data is a collection of facts
- Data is the “gas to AI”
- Quantity, completeness, and accuracy of data matters
- Data Cloud
- Data Literacy
- Qualitative Variables: describes qualities or characteristics, such as country of origin, gender, name, or hair color
- These are NOT numeric, and cannot usually be aggregated
- Used to categorize or segment data
- Nominal Variables: Characteristics that CANNOT be ranked
- E.g., types of fruit
- Ordinal Variables: Can be ranked
- E.g., Case Priority of high, medium, or low
- Quantitative Variables: describes measurable characteristics, such as height, weight, temperature
- These are numbers
- Can be aggregated
- Qualitative Variables: describes qualities or characteristics, such as country of origin, gender, name, or hair color
- Data Quality: Traits of “good data”
- High Volume: large amount of relevant, available data means that there’s a better chance you’ll have what you need to answer your questions
- Important Note: Relevancy matters, so don’t acquire data just to have data
- Historical: Data that goes back in time allows you to see how the present situation arose due to patterns that have occurred over time, such as looking at sales trends over the last 10 years to see increases or decreases
- Consistent: As things change, data should be adjusted for consistency
- E.g., Salary and pricing data adjusted for inflation
- Multivariate: Data should contain both quantitative and qualitative variables
- The more variables in the data, the more that you can discover from it
- Atomic: The more finely detailed the data, the more you are able to examine it at various levels of detail
- Granularity is important to consider, and often the more granular your data, the better
- E.g., If you wanted to understand bicycle riding trends in your state, it would be helpful to see those trends as impacted by county, city, and neighborhood
- Clean: Data needs to be accurate, complete, and free from errors
- Clear: Data should be written in terms that can be easily understood, not in code
- E.g., the housing type values “single family”, “two-family conversion”, and “end-unit townhouse” are much easier to understand than “1Fam”, “2fmCon”, and “TwnhsE”
- Dimensionally Structured: An accessible way to structure data is to organize it into two types:
- Dimensions (qualitative values)
- Measures (quantitative values)
- Richly Segmented: Groups, based on similar characteristics, should be built into data for easier analysis
- E.g., Data about movies could be grouped by genre
- Of Known Pedigree: To trust the data, you should know its background-where it comes from and how it has since been altered
- High Volume: large amount of relevant, available data means that there’s a better chance you’ll have what you need to answer your questions
- Data Aggregation Terms
- Sum: Arithmetic total of the numbers
- Average/Mean: Arithmetic mean of the numbers
- Median: The middle value in a data set in which the values have been placed in order of magnitude
- Minimum: Smallest number
- Maximum: Largest number
- Count: Number of rows
- Tableau Products
- Tableau Prep Builder: Used to clean, shape, and combine data so that it’s ready for analysis
- Desktop app or web app
- Validate & preview your data prep in Tableau Desktop
- Save cleaned and prepped data extracts to a file, publish to Tableau Server or Tableau Cloud, or write the output from your flows to external relational databases
- Save flows or publish to Tableau Server/Cloud to run on a schedule using Tableau Prep Conductor (which requires the Data Management Add-On)
- Tableau Desktop: Used to create & publish data sources and visualizations, to analyze the data, and to get insights from your data
- Tableau Server: Used to share what you’ve discovered so that others can view & analyze your data
- Allows you to author on the web by creating and editing your workbooks without having to use Tableau Desktop
- Web-based, customer-hosted
- Tableau Cloud: Same as Tableau Server, only hosted by Tableau
- Tableau Public: Create, save, and share visualizations using Tableau Public, a free, public instance of Tableau Cloud
- Tableau Desktop Public Edition: A free version of Tableau Desktop, which you can use to create and publish visualizations to Tableau Public
- Tableau Mobile: Used to get access to hosted data on the go
- Free app available for iOS and Android
- Tableau Prep Builder: Used to clean, shape, and combine data so that it’s ready for analysis
Ethical Considerations of AI
- Principles of Inclusive design of AI:
- Recognize Exclusion: When we solve problems using our own biases, exclusion happens
- We need to recognize exclusion before we can address it
- Learn from Diversity: Humans are the real experts in adapting to diversity
- Solve for One, Extend to Many: Focus on what’s universally important to all humans
- Recognize Exclusion: When we solve problems using our own biases, exclusion happens
- Guidelines for Trusted Generative AI
- Accuracy: Balance Accuracy, precision & recall
- Safety: Conducting bias, explainability, and robustness assessments, as well as red-teaming
- Honesty: Get consent to use data when training models
- Further, disclose when AI creates content or delivers it
- This is an interesting one that I feel many companies are hesitant to do
- Further, disclose when AI creates content or delivers it
- Empowerment: Automation vs. Augmentation, and finding the right balance
- Sustainability: Develop right-sized models
- Salesforce’s Trusted AI Principles
- Responsible: We strive to safeguard human rights, to protect the data we are trusted with, observe scientific standards, and enforce policies against abuse. We expect our customers to use our AI responsibly, and in compliance with their agreements with us
- Accountable: We believe in holding ourselves accountable to our customers, partners, and society. We will seek independent feedback for continuous improvement of our practice and policies, and work to mitigate harm to customers and consumers
- Transparent: We strive to ensure our customers understand the “why” behind each AI-driven recommendation and prediction so they can make informed decisions, identify unintended outcomes, and mitigate harm.
- Empowering: We believe that AI is best utilized when paired with human ability, augmenting people, and enabling them to make better decisions
- This is an interesting one that I feel is not actually communicated much at the Sales level. The idea of reducing headcount, or making additional hiring unnecessary is a regular talking point when Salesforce AEs are attempting to sell Agentforce or other Salesforce AI tools.
- Inclusive: AI should improve the human condition and represent the values of those impacted, not just the creators. We will advance diversity, promote equality, and foster equity through AI
- Ethical AI Practice Maturity Model
- Ad Hoc
- Someone raises their hand and starts asking not just “Can we do this?” but “Should we do this?”
- Informal advocacy builds a groundswell of awareness
- Ad hoc reviews and risk assessments take place among “woke” teams
- Organized & Repeatable
- Executive buy-in established
- Ethical principles and guidelines are agreed upon
- Build a team of diverse experts
- Company-wide education
- Ethics reviews are added onto any existing reviews, often at the end of the dev process
- Managed & Sustainable
- Ethical considerations are baked into the beginning of product development and reviews happen throughout the lifecycle
- Build or buy bias assessment and mitigation tooling
- Metrics are identified to track progress and impact post-market for regular audits
- Optimized & Innovative
- End-to-end inclusive design practices that combine ethical AI product and engineering dev with privacy, accessibility, and legal partners
- Ethical features and resolving ethical debt are a formal part of roadmap and resourcing
- Poor ethics metrics block product launch
- Ad Hoc
AI Associate Exam Results
After watching the linked video below slowly, and taking the notes above, I passed the test, missing only 1-2 questions. I feel like this material prepared me pretty well for the exam, though likely there are additional resources in Trailhead that would have helped me to not have to potentially guess here or there.