AI Usage Guide
AI Tools Utilization and Best Practices
Westmoreland Community Action
November 26, 2024
AI Tools Utilization and Best Practices
Westmoreland Community Action
November 26, 2024
I. Preamble: AI in Non-Profit Organizations
II. Responsible AI Practices
III. AI Tools Overview and Integration
IV. AI for Social Determinants of Health (SDOH) Communication
V. Effective Use of AI in Daily Operations
> Optimizing outputs as well
VI. Resources and Continuous Learning
Artificial Intelligence (AI) has emerged as a transformative tool for non-profit organizations, offering innovative ways to enhance services and support communities At Westmoreland Community Action (WCA), embracing AI is an opportunity to improve client interactions, streamline service delivery, and address challenges associated with resource constraints and complex program offerings
AI refers to technologies capable of performing tasks that typically require human intelligence. For WCA, AI adoption is centered on enabling client-facing staff and administrative teams to work more efficiently and effectively, focusing on the following key areas:
1 Rapid Information Access: AI can retrieve relevant program, policy, or community resource information in seconds, empowering staff to deliver prompt and accurate client assistance
2 Enhanced Client Support: By analyzing client needs, AI can match individuals with suitable WCA services, ensuring that support is tailored and impactful.
3. Data-Driven Insights: AI tools can help WCA track trends in service usage and outcomes, providing actionable insights for decision-making and program improvement
For WCA’s client-facing team members, AI offers several possibilities to overcome common non-profit challenges, such as limited resources, complex service offerings, and the need to demonstrate measurable impact Specific benefits include:
● Improved Client Interactions: AI tools enable staff to access a wealth of program information in real-time, ensuring conversations are informed and supportive
● Efficient Resource Matching: Future AI systems including Skilly could analyze client data or conversations to recommend the most appropriate programs or services, enhancing client outcomes
● Up-to-Date Knowledge: With WCA’s services evolving, AI tools could be used as “answer engines” to provide updated policy or program changes
● Streamlined Assistance: AI simplifies navigating WCA’s wide range of programs, ensuring staff can provide comprehensive support without being overwhelmed
● Social Determinants of Health (SDOH) Analysis: Our goal is to use AI to improve outcomes related to SDOH. This could be achieved through learning AI practices.
These capabilities ensure AI complements not replaces staff expertise, preserving the compassion and human understanding that are central to WCA’s mission
By leveraging AI for information retrieval and client support, you can enhance your ability to serve our community effectively This technology will assist you in providing more informed, efficient, and personalized support across all our programs, from Headstart to workforce development initiatives
As we integrate these AI tools into our work, remember that they are designed to augment your expertise and compassion, not replace them Your professional judgment and understanding of our clients' unique situations remain crucial in providing the best possible support
and data protection
Best practices for accuracy and privacy
Introduction
Integrating AI into WCA’s operations opens exciting possibilities for empowering on-site workers and enhancing client outcomes To achieve this responsibly, WCA must prioritize ethical use, privacy, and trust while addressing the practical needs of those using AI tools daily This chapter serves as a guide for WCA’s on-site staff, focusing on actionable steps, clear examples, and best practices for engaging with AI tools like Skilly
1. Questioning AI Outputs
AI tools like Skilly can provide quick answers and suggestions, but they aren’t perfect. Always verify outputs against your own expertise and available resources
● Tip: If a suggestion feels off, double-check with trusted sources.
● Example: If Skilly recommends a program that doesn’t align with a client’s needs, consider alternative solutions and provide feedback on the tool’s performance.
2. Transparency in AI Use
On-site workers should be prepared to explain how AI tools like Skilly operate and their role in client support This builds trust and ensures clients understand AI’s role in their assistance
● What to Say: “This tool helps me find the most relevant information quickly, but I’ll confirm everything with you before we proceed ”
● Why It Matters: Clients are more likely to trust AI-enhanced support when its role is transparent and clearly communicated
3. Ethical Data Handling
AI tools rely on data to function, but maintaining privacy is critical. On-site workers should follow these practices:
● Avoid Sharing Sensitive Data: Never input private client information into external tools like public chatbots.
● Why It Matters: Sharing individual client information with identifiable information is illegal, and public chatbots could use this to train future models which nefarious actors could use to squeeze leaked information out of.
AI tools can inadvertently reflect biases On-site staff should actively question outputs and ensure recommendations align with WCA’s values of equity and inclusion
● Regularly evaluate whether AI is producing consistent results across different client groups
● Ensure prompts and questions used in AI systems reflect WCA’s diverse community
● Action: Keep an open dialogue with colleagues about biases you encounter and how they’re addressed
Practical Example: Using AI Tools at WCA
Scenario: A client asks for housing assistance
Step 1: Use Skilly to quickly reference the latest housing programs and eligibility criteria.
Step 2: Cross-check the results with WCA’s program documentation (Skilly should be able to link this information) to confirm accuracy.
Step 3: Initial draft of the housing assistance application with a GPT model such as ChatGPT using a suggested prompt for housing assistance and the information you gatherer from Skilly (adhering to data safety practices)
Step 4: Refine the above step with your own expertise
Step 5: Discuss the options with the client, ensuring they understand the recommendations
Outcome: Faster support for the client while maintaining accuracy and trust
1 Keep It Simple
Only input general or non-sensitive information into AI tools unless explicitly authorized ○ Example: Instead of entering a client’s name, use general descriptors like “client A” when testing a tool
2. Ask Before Sharing
If you’re unsure whether data can be shared with an AI tool, check with your supervisor.
3. Use Approved Tools
Only use AI platforms vetted by WCA to ensure data safety and compliance with privacy policies.
To effectively use AI tools like Skilly, staff need ongoing training and support
● Initial Training: Learn basic features, best practices, and privacy protocols
● Annual Updates: Stay informed about new AI features and advancements through regular workshops
● Peer Learning: Share tips and experiences with colleagues during team meetings
Responsible AI practices empower WCA’s on-site workers to use tools like Skilly effectively, enhancing client interactions while safeguarding trust and privacy By questioning outputs, handling data ethically, and embracing ongoing learning, staff can confidently integrate AI into their daily work These practices lay the foundation for future growth as WCA continues to refine its AI tools and processes
In the next chapter, we’ll explore Practical Applications at WCA, focusing on real-world scenarios and success stories to illustrate the transformative potential of AI
Artificial intelligence (AI) tools can significantly enhance Westmoreland Community Action’s (WCA) ability to connect with clients, donors, and staff by automating tasks, providing instant access to information, and streamlining workflows. This section provides an overview of the tools selected for WCA, along with a framework for deciding when and how to integrate them into organizational processes.
1. General-Purpose AI Tools
These tools are versatile and can handle a broad range of tasks, such as answering questions, automating communication, and summarizing data. At WCA, the following tools fall into this category:
1.1. ChatGPT: Used for internal operations, such as drafting documents, generating responses, and supporting staff in decision-making and case management. However, ChatGPT can be incorrect and can even falsify sources and data, which requires additional follow-up on correctness for ChatGPT outputs.
1.2. Perplexity: A research and information retrieval tool for internal staff to find accurate, summarized answers to specific questions quickly, especially for program information or policy updates.
2. Specialized AI Tools
These tools are designed for specific use cases and tailored to meet the unique needs of WCA.
2.1. Skilly.ai: A user-friendly interface to provide a front-facing chat solution for clients and donors, ensuring prompt and accurate responses to common queries, program details, or donation processes.
The successful integration of AI tools requires thoughtful consideration of tasks, goals, and user needs A flowchart-based approach can help streamline tool selection and implementation Below is a conceptual guide:
Prompt engineering is the practice of crafting clear, targeted, and context-rich instructions to guide AI tools like ChatGPT For nonprofits like WCA, effective prompting bridges the gap between generic AI capabilities and the specific needs of the organization, such as addressing SDOH.
Without thoughtful prompts, AI responses may lack relevance or depth. This guide provides practical examples and techniques for leveraging prompt engineering to maximize AI’s potential as a productivity tool for staff and administrators.
AI models, like ChatGPT, process language based on patterns in their training data. However, they don’t inherently know your organization’s context or goals. Using targeted prompts helps the AI focus on WCA’s specific mission, programs, and objectives.
Key Benefits of Prompt Engineering for Nonprofits:
● Tailored outputs that align with organizational goals.
● Cost-effective and practical solutions compared to expensive AI fine-tuning.
● Streamlined operations and enhanced communication tools.
1. Use Context-Rich Prompts
Provide background details and specific instructions to get actionable and relevant results.
Example for Client Feedback
Simple Prompt:
“Create a system for capturing client feedback.”
Optimized Prompt:
“Westmoreland Community Action (WCA) connects individuals and families with resources to improve SDOH outcomes, including housing stability and healthcare access. Design a client feedback system that captures practical and accessible information, helping on-site staff make real-time adjustments and administrators plan long-term improvements.”
Result Comparison:
○ The optimized prompt provides detailed, actionable feedback relevant to WCA’s goals.
○ The simple prompt yields vague suggestions that may not align with specific needs.
2. Ask for Specific Outputs
Guide the AI to deliver focused results by specifying the format or structure of the output.
Example for Decision-Making Frameworks
Simple Prompt:
“Create a framework for improving SDOH outcomes.”
Optimized Prompt:
“WCA’s on-site staff address SDOH during client interactions. Develop a step-by-step decision-making framework that includes input from administrators, ensuring actions align with organizational goals. Provide practical examples for staff to follow.”
Result Comparison:
○ The optimized prompt generates a detailed framework, with actionable steps and role clarity.
○ The simple prompt produces a general outline that lacks depth.
3. Incorporate the Target Audience
Include details about the roles of staff and administrators in your prompt to ensure relevance.
Example for Explaining SDOH Impacts
Simple Prompt:
“Explain how housing programs impact client outcomes.”
Optimized Prompt:
“WCA’s programs address SDOH through housing assistance and mental health services. Explain how these programs improve client outcomes, focusing on specific examples like housing stability and healthcare access. Use accessible language and provide three relatable examples.”
Result Comparison:
○ The optimized prompt generates specific, relatable examples tailored to WCA’s audience.
○ The simple prompt yields generic responses without actionable insights.
Ensure that prompts clearly connect to WCA’s mission and operational objectives.
Example for Bridging Staff and Administrator Perspectives
Optimized Prompt:
“Compare the perspectives of WCA’s on-site staff and administrators regarding resource allocation. Identify five shared goals and three areas where their viewpoints diverge. Provide these in a table format, highlighting actionable solutions to align their efforts.”
Simple Prompt:
“Compare staff and administrator perspectives on resource allocation.”
Result Comparison:
○ The optimized prompt identifies actionable insights and fosters collaboration.
○ The simple prompt provides general comparisons without solutions.
1. Be Specific: Include relevant details about WCA’s goals, programs, or challenges in your prompts.
2. Guide the AI: Specify the type of output you need—tables, step-by-step guides, or examples.
3. Test and Refine: Experiment with different prompts to find what works best for your needs.
4. Stay Critical: Question the AI’s outputs and ensure they align with your expertise and organizational values.
5. Use Templates: Standardized prompt templates save time and ensure consistency across teams.
Task
Client Feedback Mechanism
Program Impact Explanation
Collaborating Across Roles
Optimized Prompt Example
“Design a feedback system for capturing client insights on SDOH outcomes, ensuring accessibility for clients and real-time usability for staff.”
“Explain how WCA’s housing program improves SDOH factors like housing stability Provide three examples using clear and accessible language.”
“Develop a table comparing on-site staff and administrator perspectives on program challenges. Identify shared goals and areas for improvement.”
Training Module Outline
“Create an outline for a training module that teaches WCA staff to effectively use AI tools like Skilly for retrieving program information and improving client support.”
● Prompt engineering is an essential skill for leveraging AI effectively in nonprofits.
● Optimized prompts produce outputs that save time, enhance productivity, and align with organizational goals.
● On-site staff and administrators can use practical techniques like adding context, specifying outputs, and testing prompts to make the most of AI tools like ChatGPT
By adopting these prompt engineering practices, WCA can empower its team to harness AI’s potential as a productivity tool, ensuring it supports the organization’s mission while addressing community needs effectively.
Communication
Leveraging AI for SDOH knowledge and communication
Onsite employees often need quick, accurate information about various programs, each with unique rules and guidelines. AI tools like ChatGPT and Perplexity can act as reliable assistants, saving time and improving efficiency.
Using AI for Information Retrieval
1 Quick Answers to Program-Specific Questions
● Example Use: Ask ChatGPT or Perplexity to clarify program rules, eligibility criteria, or procedures.
○ Prompt: “What are the eligibility requirements for the housing assistance program?”
● Tip: Be specific with your question to get a focused response.
2. Summarizing Program Details
● Use AI to condense lengthy documents into concise summaries for easier understanding.
○ Prompt: “Summarize the client intake process for the food assistance program in bullet points.”
3. Generating Checklists
● Create checklists to ensure compliance with program requirements
○ Prompt: “Create a checklist for enrolling a new participant in the digital literacy program ”
4 Preparing for Client Interactions
● Quickly draft client communication or responses to FAQs
○ Prompt: “Write a response explaining the steps for applying to our job training program ”
Best Practices
1. Clear Prompts: Frame questions to include program names and specifics for accurate answers.
2. Cross-Check Outputs: Verify AI-generated information with official documents or supervisors to ensure accuracy.
3. Data Privacy: Avoid inputting confidential client details when asking for assistance.
1 Practice Retrieval: Participants ask AI for specific program rules or steps and evaluate the clarity of responses
2 Create a Summary: Use AI to summarize a complex program guideline
By leveraging AI for quick, reliable information retrieval, employees can focus more on impactful client interactions while ensuring adherence to program rules
This involves defining clear tasks, providing context, evaluating outputs, and refining prompts to improve results. Below, we explore the four stages of prompt engineering to help WCA staff maximize the effectiveness of AI tools in daily operations.
Stage 1: Idea
"Create an idea, task, or question that you want the AI model to perform."
● Start by identifying the specific goal you need the AI to accomplish
● Examples:
○ Drafting an email to donors about a fundraising event
○ Summarizing a case report for internal review.
○ Creating an FAQ response for clients inquiring about housing assistance.
Best Practices:
● Clearly define the outcome you expect
● Avoid vague or overly broad goals (e g , "Help with programs") be specific (e g , "Summarize the top three benefits of WCA’s job training program")
Stage 2: Prompt
"Provide context and the content you wish to prompt the AI model to fulfill your task. The more information the better, except for confidential information."
● Build a detailed prompt that includes all relevant details about the task, audience, and desired tone
● Examples:
○ Poor: “Write about WCA programs ”
○ Better: “Create a social media post highlighting WCA’s job training program The tone should be inspirational, and include a success story.”
Key Elements of an Effective Prompt:
1 Context: Background information on the task or audience
2 Specificity: What exactly do you need from the model?
3 Format: Specify the desired structure (e g , list, email, paragraph)
4 Constraints: Any word limits or style requirements
Tip: Never include confidential or sensitive information in prompts.
Stage 3: Evaluate
"Examine the AI model’s response and determine if it fulfills your task. What is still lacking in its response?"
● After receiving the AI’s output, critically review it for accuracy, tone, completeness, and alignment with your initial task
● Questions to Ask:
○ Does the response address the task effectively?
○ Is the tone appropriate for the intended audience?
○ Are there any inaccuracies or missing details?
Example Evaluation:
Task: Create a response to a client asking about food assistance
AI Output: “We offer various programs to help people in need ” Evaluation: This is too generic It doesn’t provide specific information about food assistance programs or how to access them.
"Inspect your prompt, and see what contextual information is missing. Infer what could have led to the output you received and prompt the model once more."
● Use your evaluation to identify gaps in the original prompt and add details to guide the AI toward a better response
● Examples of Refinement:
○ Original Prompt: “Write about WCA’s food assistance.”
○ Refined Prompt: “Explain WCA’s food assistance program in 150 words Include who qualifies, how to apply, and one success story to illustrate impact ”
Iterate as Needed:
● Continue refining until the response meets your expectations
● Each iteration helps the AI better understand your requirements.
1 "Google AI Essentials" by Google: A comprehensive course on leveraging generative AI tools to boost productivity in the workplace
2 "5 AI for Work Tips and Tricks": Demonstrates practical ways to use AI for email management, data analysis, and more
3 "Artificial Intelligence in the Workplace": Explores how AI tools like ChatGPT are revolutionizing work processes and enhancing security measures
4 "Using AI Productively at Work, Home, & Job Hunting" by Jill Finlayson: Provides insights on effectively incorporating AI into various aspects of professional and personal life
1 American Cancer Society: Used machine learning to identify effective communication channels and messages for donors, resulting in a 400% increase in donation conversion rates
2 HIAS: Implemented AI to predict which email appeals were most likely to result in donations, leading to a 230% increase in contributions
3 International Rescue Committee: Utilized AI to process refugees' asylum claims more efficiently, expediting the process while improving accuracy
4 charity: water: Leveraged machine learning algorithms to predict donor lapse and intervene, significantly reducing attrition rates
5. National Kidney Foundation: Implemented AI for personalized email campaigns, doubling campaign open rates.
6. One Acre Fund: Used AI to personalize communication with clients, leading to an 18% increase in loan repayment rates and expanded services to smallholder farmers in multiple African countries.
7. United Way of New York City: Implemented an AI chatbot that helped donors learn about and donate to the organization, raising over $50,000 during the COVID-19 pandemic.
8. Hebrew Immigrant Aid Society (HIAS): Utilized machine learning to optimize refugee resettlement locations, potentially increasing refugee employment rates by up to 40%.
1 Grammarly: AI-powered writing assistant for grammar, spelling, and style corrections
2 Otterai: AI-powered transcription and note-taking tool for meetings and interviews
3 Calendly: AI-enhanced scheduling tool to streamline appointment booking
4 Asana: Project management tool with AI features for task prioritization and workload balancing
5 Zapier: Integration tool with AI chatbots for automating workflows across various applications
6. Trello: Project management platform with generative AI features for task creation and summaries.
7. Zoom Workplace Companion: AI tool for automating meeting summaries and action item assignments.
8. BeeDone: Gamified task management tool using AI to incentivize productivity.