1
Q
How would you explain what Claude is to someone who’s never used AI?
A
Claude is an AI assistant developed by Anthropic that can understand and generate human-like text. Think of it as a highly knowledgeable colleague who can help with writing, analysis, coding, research, and problem-solving. What makes Claude different is that it’s built with safety and helpfulness as core priorities — Anthropic designed it to be helpful, harmless, and honest.
2
Q
What’s the difference between the API and Claude for Work (Enterprise)?
A
The API is for developers building AI-powered applications — they integrate Claude programmatically into their products using code. Claude for Work (Team and Enterprise plans) is for organizations that want their employees to use Claude directly through a chat interface, with enterprise features like SSO, admin controls, audit logs, and expanded context windows. A nonprofit might use the API to build a custom grant-writing tool, while using Claude for Work so their staff can collaborate with Claude on daily tasks.
3
Q
When would you recommend a customer use the API vs. Enterprise?
A
Recommend the API when the customer wants to build a product or automate workflows at scale — like a nonprofit building a chatbot for donor inquiries. Recommend Enterprise when they want their team to interact with Claude directly for varied tasks — research, writing, analysis — with enterprise security controls. Many organizations use both: the API for their products, Enterprise for their people.
4
Q
What are Claude’s key strengths compared to other LLMs?
A
1) Safety and alignment: Constitutional AI makes Claude more reliable. 2) Long context window: Up to 200K tokens standard (500K on Enterprise). 3) Nuanced instruction following: Particularly good at complex, multi-step instructions. 4) Reduced hallucinations: More likely to say ‘I don’t know’ than make things up. 5) Vision capabilities: Can process images, charts, PDFs natively.
5
Q
What are Claude’s biggest limitations?
A
1) Knowledge cutoff: Doesn’t know about very recent events without web search. 2) Hallucinations: While reduced, can still generate plausible-sounding but incorrect information. 3) No real-time data: Without tool integrations, can’t access live databases. 4) Context window limits: Even 200K tokens has limits for very large document sets. 5) Non-deterministic outputs: Same prompt can yield slightly different responses.
6
Q
How does Claude handle hallucinations, and how would you explain that to a customer?
A
Hallucinations are when AI generates confident-sounding but incorrect information. Claude has improved significantly — it’s 3-4 times more likely to say ‘I don’t have that information’ rather than making something up. Coach customers to always verify critical information, use Claude for drafts rather than final sources of truth, and design workflows with human review for high-stakes decisions.
7
Q
What’s a context window and why does it matter?
A
The context window is Claude’s working memory — everything it can ‘see’ at once, including your prompt and its response. Claude’s 200K token window is roughly 150,000 words or 500 pages. For a nonprofit, this means Claude can analyze an entire grant application, review a full annual report, or process dozens of interview transcripts in one go.
8
Q
How would you explain system prompts to a non-technical customer?
A
A system prompt is like giving Claude its job description before a conversation starts. It sets the context, tone, and rules. For example: ‘You are a helpful assistant for a food bank. Be warm and supportive. Never provide medical advice.’ The user never sees this, but it shapes every response. It’s how you customize Claude’s behavior for your specific use case.
9
Q
What is Constitutional AI and why does it matter for nonprofit customers?
A
Constitutional AI is Anthropic’s approach to making Claude safe and aligned with human values. Instead of just relying on human feedback, they give Claude a ‘constitution’ — explicit principles drawn from sources like the UN Declaration of Human Rights. Claude learns to critique and improve its own responses based on these principles. For nonprofits, this matters because they can trust Claude won’t generate harmful content, and the values are transparent and auditable.
10
Q
How does Claude’s safety approach differ from OpenAI or Google?
A
Anthropic takes a ‘safety-first’ approach — it’s in their founding mission. Constitutional AI makes Claude’s values explicit and inspectable, whereas other models rely more heavily on human feedback which can be inconsistent. Anthropic also publishes detailed safety research and model cards. For risk-averse organizations like nonprofits and government, this transparency can be a deciding factor.
11
Q
What’s Claude’s knowledge cutoff and how do you handle questions about current events?
A
Claude’s knowledge cutoff varies by model — the latest models have training data through early-to-mid 2025, but the reliable knowledge cutoff is a few months earlier. For current events, Claude now has web search capabilities that can pull real-time information. Explain to customers that for time-sensitive work, they should either enable web search or provide Claude with current documents directly.
12
Q
What are Claude’s vision capabilities and when would a nonprofit use them?
A
Claude can process images, photos, charts, graphs, and technical diagrams. For nonprofits: analyzing infographics from research reports, processing scanned documents or forms, reviewing photos from field work, extracting data from charts in PDF reports, reading handwritten notes from community feedback. Particularly powerful combined with the long context window.
13
Q
What is tool use and how might a nonprofit leverage it?
A
Tool use lets Claude call external functions or APIs during a conversation. Instead of just generating text, Claude can take actions — search a database, run calculations, fetch current information. A nonprofit might: let Claude search their donor CRM, connect to their document management system, pull real-time data from program tracking, integrate with email or calendar. It turns Claude from a text generator into a workflow participant.
14
Q
How would you explain the different Claude models (Opus, Sonnet, Haiku) to a customer?
A
Opus is the most powerful — best for complex reasoning, strategic analysis, and tasks where accuracy matters most. Slower and more expensive, but worth it for high-stakes work. Sonnet is the all-rounder — excellent balance of capability, speed, and cost. Great for everyday tasks, coding, enterprise workloads. Haiku is the speedster — fastest and cheapest, perfect for quick tasks, high-volume processing, or real-time applications.
15
Q
When would you recommend a smaller model vs. a larger one?
A
Recommend Haiku (smaller) when: speed matters more than depth, tasks are straightforward, budget is tight and volume is high, building rapid prototypes. Recommend Opus (larger) when: accuracy is critical, tasks require complex reasoning, analyzing large amounts of information, cost of errors outweighs model cost. The key is matching the model to the task.
16
Q
What’s extended thinking mode and when would you use it?
A
Extended thinking lets Claude ‘think longer’ before responding — more step-by-step reasoning internally. Like asking someone to really think through a problem rather than giving a quick answer. Use it for: complex analysis or strategy questions, math or logic problems, situations requiring multiple angles, when accuracy matters more than speed.
17
Q
What’s prompt caching and why does it matter for cost?
A
Prompt caching saves money by reusing parts of prompts that don’t change. If processing 100 documents with the same system prompt and instructions, you don’t pay for those tokens 100 times. With caching, Claude remembers the static parts and you only pay for what’s new. For nonprofits watching every dollar, this can mean up to 90% cost savings on repetitive tasks.
18
Q
How does batch processing work and when would you recommend it?
A
Batch processing lets you submit many requests at once and get results later — usually within 24 hours. It’s 50% cheaper than real-time requests. Recommend for: processing a large backlog of documents, running analysis overnight, any task that doesn’t need immediate results. For a nonprofit doing annual report analysis or processing a year’s worth of donor feedback, batch processing is a no-brainer.
19
Q
What’s the Files API and how would nonprofits use it?
A
The Files API lets you upload documents once and reference them across multiple conversations or API calls. Instead of re-uploading a 50-page policy document every time, you upload once and reference the file ID. For nonprofits with large document libraries — policy manuals, program guides, historical reports — this makes it much easier to give Claude consistent access to organizational knowledge.
20
Q
How does Claude handle PDFs specifically?
A
Claude can process PDFs natively — it sees the actual pages as images, so it understands formatting, tables, charts, and even scanned documents with OCR. Huge for nonprofits dealing with: grant applications and reports, compliance documents, research papers, board packets, contracts and agreements. Claude sees the PDF as a human would, understanding context and visual layout.
21
Q
What are MCP (Model Context Protocol) connectors?
A
MCP is a standard that lets Claude connect to external data sources and tools. Think of it as USB for AI — a universal way to plug Claude into your existing systems. Connectors exist for Google Drive, Slack, GitHub, databases, and more. For nonprofits, this means Claude can access your actual organizational data rather than just what you paste into a chat window.
22
Q
What’s the difference between Claude.ai and the API console?
A
Claude.ai is the consumer-facing chat interface — you log in, have conversations, create projects. The API console is the developer platform — manage API keys, monitor usage, test prompts in the Workbench, handle billing. A nonprofit might have staff using Claude.ai for daily work while their IT team uses the console to build custom integrations.
23
Q
How does Claude handle multilingual content?
A
Claude is genuinely multilingual — it can understand, respond in, and translate between many languages. For nonprofits serving diverse communities or operating internationally: communicating with beneficiaries in their native language, translating program materials, analyzing feedback in multiple languages, supporting multilingual staff. Claude understands cultural context and nuance.
24
Q
What’s the Artifacts feature in Claude.ai?
A
Artifacts lets Claude create interactive content alongside the conversation — code that runs, documents you can edit, visualizations, even small applications. For nonprofits: build a quick data visualization, create an interactive budget calculator, draft a document you can edit in real-time, generate a presentation outline you can export. Turns Claude from text output into actual deliverables.
25
How does Claude Code differ from regular Claude?
Claude Code is a command-line tool for developers — it understands your entire codebase and can help with complex coding tasks directly in the terminal. For nonprofits with development teams: building and maintaining custom applications, debugging issues, onboarding new developers to existing code, automating development workflows. Bundled with Team and Enterprise plans.
26
How would you explain Claude to a nonprofit executive who's skeptical of AI?
Start with outcomes, not technology. 'Claude helps your team do more of what matters — serving your mission — by handling time-consuming tasks like drafting, research, and analysis. It's not replacing anyone; it's giving your people capacity back. Unlike some AI tools, Claude is built by Anthropic, a company founded specifically to make AI safe and beneficial. Your data stays private, and you control how it's used.'
27
How would you explain Claude differently to a developer vs. a program director?
To a developer: 'Claude's API is RESTful, supports streaming, has a 200K token context window, and includes tool use for function calling. You can integrate it into your existing stack in a few hours.' To a program director: 'Claude can help you analyze beneficiary feedback, draft reports, and prepare for funder meetings. Imagine a research assistant who's read every document your organization has ever produced.'
28
A customer asks 'why Claude over ChatGPT?' What do you say?
Focus on their specific context: Safety and values (Claude built by company whose mission is AI safety), Context window (handles more information at once), Enterprise features (better admin controls, data privacy), Nuanced instruction following (better at complex, multi-step tasks), Honesty (more likely to say 'I don't know'). But the best answer comes from their specific use case.
29
How would you explain token limits to a non-technical user?
Tokens are like the 'words' Claude sees — roughly 4 characters each. 200,000 tokens is about 150,000 words or 500 pages. Think of it as Claude's working memory. If you give Claude a 600-page document, it can't hold it all at once — you'd need to break it into chunks. Key takeaway: Claude can handle a lot, but there are limits.
30
A customer doesn't understand why Claude 'forgot' something from earlier in the conversation. How do you explain it?
Claude doesn't have memory across conversations by default — each new conversation starts fresh. Even within a long conversation, if it exceeds the context window, earlier parts 'fall off.' It's like a whiteboard that can only hold so much — when full, oldest notes get erased. For important information, use Projects to store key context, or keep critical information in recent messages.
31
How would you explain prompt engineering without using jargon?
Prompt engineering is the art of asking Claude the right way. Think about asking a new employee to do something — the clearer and more specific you are, the better the result. With Claude: be explicit about what you want, give examples of good output, break complex tasks into steps, provide relevant context upfront. It's communication skills applied to AI.
32
A customer says 'AI hallucinates too much, we can't trust it.' How do you respond?
Valid concern. Claude is designed to address this — it's 3-4 times more likely to say 'I don't know' than make something up. But no AI is perfect. The solution is designing workflows that account for this: use Claude for drafts not final sources, build in human review for high-stakes decisions, verify critical facts. Claude is a powerful tool that works best when humans stay in the loop.
33
How do you build confidence with a customer who's nervous about AI safety?
Acknowledge their concerns are legitimate. Explain Anthropic's approach: founded specifically to build safe AI, Constitutional AI makes values explicit and inspectable, extensive safety testing before release, transparent research publications, data privacy commitments (no training on Enterprise data). Offer to walk through their specific concerns — what exactly worries them?
34
How would you explain the difference between training data and context window?
Training data is what Claude learned from — billions of documents from the internet, books, articles. It's like Claude's education. The context window is what Claude can see right now in this conversation — your prompt, uploaded documents, conversation history. It's like Claude's desk. Education gives general knowledge; what's on the desk is what it's working with now.
35
A nonprofit executive asks: 'Can Claude access our donor database?' How do you explain the options?
Claude doesn't automatically access anything — it only sees what you provide. Options: 1) Copy and paste (simple but manual), 2) File uploads (upload exports or reports), 3) API integrations (custom connection, requires development), 4) MCP connectors (pre-built integrations), 5) Enterprise features (native integrations with GitHub, Google Drive). Right approach depends on technical capacity and security requirements.
36
How would you explain streaming responses to a non-technical stakeholder?
Streaming means you see Claude's response as it's being generated — word by word — instead of waiting for the whole thing. It makes Claude feel more responsive and lets you stop it early if going wrong direction. Like watching someone write versus getting a finished letter. For most conversational use cases, streaming makes the experience feel much more natural.
37
A customer asks what 'temperature' means in AI. How do you explain it?
Temperature controls how creative or predictable Claude is. Low temperature (closer to 0) means more consistent, predictable answers — good for factual tasks. Higher temperature means more creative, varied responses — good for brainstorming. Like asking a team member to 'stick to the facts' versus 'think outside the box.' Most enterprise uses want lower temperature for consistency.
38
How would you explain RAG (Retrieval-Augmented Generation) to a program officer?
RAG lets Claude search through your documents to find relevant information before answering. Instead of just relying on training, Claude first retrieves specific information from your knowledge base, then uses that to generate a response. For a nonprofit, Claude can answer questions about your specific programs, policies, and history — not just general knowledge. Like giving Claude access to your organizational memory.
39
A customer is confused about why Claude gave different answers to the same question. How do you explain it?
AI models like Claude are inherently non-deterministic — there's some randomness built in, which makes them creative and natural-sounding. The same question might get slightly different phrasing or examples. For tasks needing consistency: lower temperature settings, more specific prompts, structured output formats. Some variation is actually a feature — it's why Claude doesn't sound robotic.
40
How would you explain fine-tuning vs. prompting to a non-technical audience?
Prompting is giving Claude instructions at the start of a conversation — 'act like a grant writer, focus on these priorities.' Fine-tuning would be retraining Claude on your specific data to permanently change how it behaves. Most organizations don't need fine-tuning — good prompting gets you 90% of the way there. Fine-tuning is expensive and complex; prompting is something anyone can do.
41
A nonprofit wants to use Claude to help with grant writing. Walk me through how you'd approach that.
1) Discovery: What types of grants? Success rate? Time bottlenecks? 2) Quick wins: Use Claude to research funders, draft initial sections, review for clarity. 3) Build infrastructure: Create Project with past successful grants, style guide, boilerplate. 4) Workflow integration: Define when Claude helps vs. humans lead. 5) Measure impact: Track time saved, quality improvements, submission rates. Goal isn't automating grant writing — it's giving team capacity to apply for more grants.
42
A public health org wants to use Claude for patient triage. What questions would you ask?
High-stakes use case. Ask: What's the clinical context? Emergency vs. general intake? What decisions would Claude make or support? What's the human oversight model? Regulatory environment? HIPAA? State laws? What happens if Claude makes a mistake? Clinical validation requirements? How handle edge cases or emergencies? Recommend Claude for documentation, research, triage preparation — not making clinical decisions directly.
43
An education nonprofit wants to use Claude for tutoring. What considerations would you raise?
Ask about: Age of students (COPPA concerns? Content appropriateness?), Subject areas (Claude excels at some more than others), Equity (Who has access? Help or widen divides?), Teacher role (complement, not replace), Assessment (How know if working?), Guardrails (What shouldn't Claude discuss with students?). Note that Anthropic has AI Fluency for Students course.
44
A customer wants to deploy Claude for a use case that seems risky. How do you handle it?
Seek to understand first — what are they actually trying to accomplish? Sometimes underlying need is valid but approach is wrong. If truly problematic: be direct but non-judgmental about concerns, explain Anthropic's values and usage policies, explore alternative approaches achieving legitimate goals, if necessary decline and escalate internally. The relationship matters, but not more than doing the right thing.
45
How would you help a customer think through what data to include in their prompts?
Frame around relevance and necessity: What does Claude actually need for this task? What's the minimum information required? Is there sensitive data that shouldn't be included? Can you anonymize or summarize instead of raw data? Also help understand context window limits — you can't include everything. Prioritize what matters most for the specific task.
46
A customer is frustrated that Claude's outputs aren't consistent. What would you suggest?
Techniques: Lower temperature (more deterministic), Structured output formats (JSON, templates), Explicit instructions ('Always include X, never include Y'), Examples (show exactly what good output looks like), System prompts (lock in consistent behavior). Also dig into why consistency matters for their use case — sometimes variation is actually fine.
47
How would you help a customer build a proof of concept vs. a production deployment?
For POC: Focus on proving value proposition, use Claude.ai or simple API calls, manual processes are fine, test with real but limited scope, define 'success' before starting. For production: Build for scale, reliability, error handling; implement proper security and access controls; design for edge cases and failures; monitor usage and costs; plan for updates and maintenance. POC is about learning; production is about reliability.
48
A nonprofit has limited technical resources. How would you help them adopt Claude?
Start with zero-code options: Claude.ai Pro/Team (no development needed), Projects (organize context without coding), Native integrations (Google Drive, GitHub, Slack), MCP connectors (pre-built for common tools). Then identify if there's a technical champion who could learn basic API usage. Anthropic has free courses on their Academy. Find highest-impact, lowest-barrier starting points.
49
How do you help a customer set realistic expectations about what Claude can do?
Be upfront about capabilities and limitations: Claude is incredibly capable but not infallible, works best with human oversight, it's a tool not a replacement for expertise, some tasks are better suited than others, requires learning and iteration. Set expectations about adoption curve — takes time to learn effective prompting and workflow integration. Quick wins build confidence.
50
A customer wants Claude to integrate with their existing systems. What questions do you ask?
Need to understand: What systems? (CRM, document management, custom software), Integration goal? (Search? Write? Update?), Technical capacity? (Do they have developers?), Security model? (SSO, data residency?), Timeline and budget?, Existing APIs to leverage? This helps match them with right approach — native integrations, MCP connectors, custom API development, or specialist partners.
51
How would you help a customer think about data privacy when using Claude?
Key points: What goes in comes out (only share what's appropriate), Data retention (understand Anthropic's policies — no training on Enterprise data), PII handling (anonymize where possible), Regulatory compliance (HIPAA, GDPR, state laws), Access controls (who can see conversations?), Audit trails (Enterprise offers logging). For sensitive sectors, walk through specific compliance requirements.
52
A nonprofit wants to use Claude for donor communications. What would you recommend?
Great use case. Suggest: Personalization at scale (draft individualized thank-you notes), Consistency (templates Claude can customize), Voice (train on brand voice with examples), Segmentation (different messaging for donor types), Review process (human approval before sending). Caution against fully automated sending — donors can tell. Claude should accelerate and enhance, not replace personal touch.
53
How would you help a nonprofit use Claude for program evaluation?
Claude can help across evaluation lifecycle: Design (evaluation frameworks, logic models), Data collection (analyze surveys, interviews, feedback), Analysis (find patterns, themes in qualitative data), Reporting (draft evaluation reports, translate for audiences), Learning (synthesize findings into recommendations). Key: Claude as analytical partner, with humans making judgment calls about meaning and action.
54
A customer asks about using Claude for volunteer management. What possibilities would you explore?
Options: Matching (analyze skills and interests against opportunities), Communication (personalized outreach, reminders, thank-yous), Training (create materials, answer common questions), Scheduling (coordination and logistics), Recognition (track contributions, draft certificates, identify milestones), Feedback (analyze volunteer surveys). Start with their biggest pain point and build from there.
55
How would you approach a nonprofit worried about job displacement from AI?
Acknowledge it's a real concern, then reframe: Claude handles time-consuming tasks, not relationship-building. Most nonprofits are understaffed, not overstaffed. Goal is capacity, not headcount reduction. Staff can focus on higher-value work. AI creates new roles. Suggest involving staff in adoption — they know what's tedious and where AI could help. When people feel ownership, fear decreases.
56
A customer wants to use Claude for advocacy and lobbying. What would you consider?
Requires careful navigation: Compliance (lobbying has legal reporting requirements), Accuracy (policy positions must be factually correct), Voice (advocacy is deeply organizational — Claude shouldn't override), Targets (research legislators, track bills, analyze policy), Materials (draft testimony, talking points, action alerts). Claude great for research, drafting, analysis. Positions and strategy should be human-driven.
57
How would you help a federated nonprofit use Claude across local chapters?
Federated orgs have unique challenges: Consistency (maintain brand while allowing local customization), Access (who gets what level?), Cost (how is usage allocated or shared?), Knowledge (share learnings across chapters), Governance (who sets rules?). Recommend starting with pilot in few chapters, developing shared best practices, creating community of practice. Enterprise plan offers needed admin controls.
58
A customer wants Claude to handle their website chat. What would you advise?
Considerations: Scope (what to answer vs. escalate?), Brand voice (ensure consistency), Escalation (how hand off to humans?), Data (what information can it access?), Testing (catch bad responses before live), Monitoring (track quality over time). Recommend starting narrow — maybe FAQs only — and expanding as you learn. Always have clear path to human support.
59
How would you help a nonprofit use Claude for crisis communications?
Crisis comms require speed and accuracy: Preparation (pre-draft response templates for likely scenarios), Real-time (Claude helps draft rapid responses, adapt messaging), Consistency (ensure all channels have aligned messaging), Monitoring (analyze incoming feedback and media coverage), Review (always human approval before release in crisis). Strongly recommend building infrastructure before crisis hits.
60
A nonprofit is nervous about AI bias affecting their work. How do you address this?
Valid concern. Explain: Anthropic actively works on reducing bias through Constitutional AI, Claude trained to be helpful to everyone, but no AI is perfectly unbiased — reflects patterns in training data. Mitigation: Test outputs across diverse scenarios, have diverse reviewers check work, be explicit in prompts about inclusive language, monitor for patterns, report concerns to Anthropic. Bias is reason for thoughtful implementation, not avoiding AI.
61
What's missing from Claude that nonprofits would need?
Based on experience: Better CRM integrations (native Salesforce Nonprofit connector), Grant database search (Foundation Directory or Candid connection), Compliance templates (pre-built for common requirements), Multi-user collaboration (better real-time features), Cost controls (nonprofit-friendly pricing or credits), Offline mode (for field work with limited connectivity). Would want to learn from actual users what's top of their list.
62
If you could add one feature to Claude for beneficial deployments, what would it be?
A 'Nonprofit Knowledge Pack' — pre-built Projects with: Grant writing best practices, Nonprofit finance and compliance templates, Board management resources, Program evaluation frameworks, Fundraising and donor relations guidance. This would dramatically reduce time-to-value for nonprofits and ensure sector-relevant guidance out of the box.
63
How would you prioritize product feedback from nonprofit customers vs. enterprise customers?
Think about: Impact (how many organizations benefit?), Mission alignment (does it advance beneficial deployments?), Feasibility (how hard to build?), Revenue (can we sustain serving nonprofits without this?), Strategic value (does it differentiate us?). Nonprofit feedback might reveal needs that also exist in enterprise. Look for patterns that serve both.
64
What do you see as the biggest barrier to AI adoption for nonprofits?
Several, but primarily: Cost sensitivity (every dollar not going to mission feels like trade-off), Capacity (no time to learn new tools when understaffed), Trust (skepticism about AI safety and reliability), Technical capacity (many lack IT resources), Change management (staff resistance or fear). Solution isn't just better technology — it's meeting nonprofits where they are with appropriate pricing, support, and change management.
65
How would you measure whether a Claude deployment is successful for a nonprofit?
Both leading and lagging indicators. Leading (early signals): Time saved on specific tasks, User adoption and engagement, Quality scores on outputs, Staff satisfaction. Lagging (mission impact): Grants submitted/won, Donors retained/upgraded, Programs delivered, Beneficiaries served. Ultimate measure: did this help them achieve their mission better? But need leading indicators to know if on track.
66
What verticals within beneficial deployments have the most potential?
Based on where AI can have outsized impact: Education (tutoring, curriculum, teacher support), Public health (health literacy, outreach, research synthesis), Social services (case management, benefits navigation, intake), Environmental (data analysis, grant writing, advocacy), Humanitarian (translation, logistics, coordination). Would want to validate with data on where current customers are finding success.
67
How would you think about pricing and packaging Claude for nonprofits?
Models to consider: Discounted tiers (X% off standard enterprise pricing), Usage-based caps (lower limits at lower prices), Mission grants (credits for qualifying organizations), Shared infrastructure (multi-tenant for smaller orgs), Freemium (basic free, premium paid). Key is sustainable unit economics while removing barriers. Would want to understand Anthropic's current approach.
68
What's a trend in AI that nonprofits should be paying attention to?
Agentic AI — systems that can take actions, not just generate text. This is where we're heading: Claude that can actually update your CRM, AI that manages multi-step workflows autonomously, systems that monitor, alert, and act. For nonprofits, this means thinking beyond 'AI as writing assistant' to 'AI as workflow participant.' Organizations that prepare for this will have huge advantages.
69
How do you see Claude fitting into the nonprofit tech stack in 2-3 years?
Claude becoming a layer that connects everything: Your CRM knows about donors; Claude helps communicate with them. Your program database tracks outcomes; Claude helps report on them. Your documents contain knowledge; Claude makes it accessible. Your team has questions; Claude provides answers. Claude won't replace specialized tools but will be the connective tissue making the whole stack more powerful.
70
How would you build a community of practice for nonprofit Claude users?
Key elements: Shared learning (case studies, templates, best practices), Peer support (users helping users with implementation), Feedback channel (direct line to product team), Events (webinars, meetups, annual conference), Champions program (power users who mentor others). The community becomes a flywheel — users help each other succeed, driving adoption and providing feedback for product improvement.
71
A nonprofit board member asks: 'How do we know this AI thing isn't a fad?' What do you say?
Point to: AI is already embedded in tools they use daily (email, search, banking), Major funders are investing in nonprofit AI capacity, Underlying technology is maturing rapidly not peaking, Competitive pressure — other nonprofits are adopting, Efficiency gains are measurable and real. Also acknowledge: some AI applications will fail, landscape will change, not every use case makes sense. Question isn't 'should we use AI?' but 'how do we use it wisely?'
72
How would you think about Claude's role in nonprofit digital transformation?
AI can be both tool and catalyst: Tool (Claude solves immediate problems — drafting, analysis, research), Catalyst (adopting Claude forces better data practices, clearer processes, documented knowledge). Organizations that succeed will use Claude adoption as forcing function for broader transformation — not just adding AI to broken processes but rethinking workflows with AI as partner.
73
What would a successful beneficial deployments team look like in 3 years?
Vision: Scale (hundreds of nonprofit customers actively using Claude), Impact measurement (clear data on mission outcomes enabled), Community (thriving peer network sharing best practices), Product influence (nonprofit needs shaping roadmap), Thought leadership (Anthropic recognized as leader in AI for good), Sustainability (business model that works for Anthropic and nonprofits). Team would be connective tissue between nonprofit needs and Anthropic's capabilities.
74
How would you balance supporting existing customers vs. acquiring new ones?
In early stages, retention and expansion are everything: Happy customers refer others, Deep case studies are marketing gold, Learning from current customers improves product, Churn undermines everything. Focus on making current customers wildly successful, documenting wins, using them to attract similar organizations. Growth without retention is a leaky bucket.
75
What metrics would you want on a beneficial deployments dashboard?
Mix of: Health (active users, engagement, NPS, churn), Impact (customer-reported outcomes, time saved, mission metrics), Growth (new customers, expansion, pipeline), Efficiency (support tickets, time-to-value, self-service rate), Product (feature adoption, feedback themes, success patterns). Dashboard should tell the story: are we helping nonprofits succeed, and is it sustainable?
76
Tell me about a time you helped a customer navigate a technical challenge.
Use Be The Match or Planned Parenthood examples from Asana. Structure as: Situation, Task, Action, Result. Focus on how you diagnosed the problem, translated technical concepts, and drove to resolution.
77
Give me an example of when you had to learn a technical product quickly.
Draw from experience ramping up on Asana's technical capabilities. Emphasize learning process, resources used, how you built expertise, and how quickly you became effective.
78
Tell me about a time you translated something complex for a non-technical stakeholder.
Use an example explaining Asana's API, integrations, or technical features to a nonprofit executive. Focus on techniques — analogies, visuals, iterative simplification.
79
Describe a situation where a customer wanted to use a product in a way it wasn't designed for. How did you handle it?
Look for example where you redirected customer toward appropriate use case while validating underlying need. Show balancing being helpful with being honest.
80
Tell me about a time you worked with a product team to advocate for customer needs.
Share example gathering customer feedback, synthesizing into actionable insights, and influencing product decisions. Bonus if it was for nonprofit customers.
81
Give an example of how you stayed current on a rapidly evolving product.
Describe learning habits — documentation, release notes, internal training, experimenting with product, peer learning. Show you're self-directed in staying up-to-date.
82
Tell me about a technical mistake you made with a customer and how you recovered.
Be honest about a real mistake. Focus on what you learned and how you rebuilt trust. Show accountability and growth, not perfection.
83
Describe a time when you had to say 'no' to a customer. How did you handle it?
Find example where you declined a request but offered alternatives. Show maintaining relationship while being honest about limitations.
84
Tell me about a complex implementation you helped a customer complete.
Use example showing project management skills — planning, stakeholder coordination, troubleshooting, successful delivery. Quantify outcome if possible.
85
Give an example of how you helped a customer measure the impact of a product.
Connect to thinking about leading vs. lagging indicators. Share specific example defining success metrics and tracking them.
86
A customer is upset that Claude gave incorrect information that caused a problem. How do you handle it?
First, acknowledge impact and apologize. Then: 1) Understand exactly what happened — prompt, response, consequence. 2) Assess: model limitation, prompting issue, or workflow design problem? 3) Immediate fix: address the problem. 4) Prevention: work on guardrails — review processes, prompt improvements. 5) Feedback: document and share with product team. Focus on preventing recurrence, not blame.
87
A competitor offers a nonprofit free AI access. How do you respond?
Focus on value, not just price: What's total cost of ownership? (Implementation, support, learning curve) Safety and privacy implications? Track record with nonprofits? Actually free, or hidden costs? Then make case for Claude: Anthropic's mission alignment, safety and reliability, investment in their success. Free is only free if it actually works.
88
How would you handle a customer who wants to use Claude for something ethically questionable?
Seek to understand first — what are they trying to accomplish, and why? Sometimes what seems questionable has legitimate purpose. If truly problematic: Be direct but non-judgmental about concerns, explain Anthropic's values and usage policies, explore alternative approaches, if necessary decline and escalate internally. Relationship matters, but not more than doing the right thing.
89
A customer is comparing Claude's benchmark scores to competitors and Claude looks worse in some areas. What do you say?
Benchmarks are useful but incomplete. Acknowledge specific results honestly, explain what benchmarks actually measure and don't, ask about their actual use case — what matters for their work?, point to benchmarks where Claude excels, offer proof of concept with their real tasks. Question isn't 'which model wins benchmarks?' but 'which model solves your problem?'
90
A nonprofit is ready to scale Claude usage but their IT team is blocking it. How do you help?
IT concerns are usually legitimate. Understand specific objections (security? privacy? compliance? control?), provide documentation addressing each concern, offer to meet directly with IT, share Enterprise security features (SSO, audit logs, admin controls), propose limited pilot with defined guardrails, connect them with similar organizations that went through this. Sometimes IT needs to be heard, not just convinced.
91
How would you help a nonprofit that's had a bad experience with AI elsewhere?
Start by listening — what happened and what did they learn? Then: Acknowledge not all AI tools or implementations are equal, explain how Claude differs (Constitutional AI, safety focus), address specific concerns directly, start small with low-risk use case, build in safeguards they're comfortable with, be patient — trust rebuilds slowly. Their skepticism is an asset for thoughtful implementation.
92
A customer asks you to help with something outside your expertise. What do you do?
Be honest about limits of knowledge then figure out how to help: Acknowledge what I don't know, see if I can find the answer through research, identify right internal resource and make introduction, follow up to ensure they got what they needed. Goal is solving their problem, not looking like I know everything. Being honest about limitations builds trust.
93
How would you handle a situation where Claude's values conflict with a customer's expectations?
This is a feature, not a bug. Explain why Claude behaves the way it does — Constitutional AI, safety principles. Validate that their underlying need is legitimate (if it is). Explore alternative approaches within Claude's guidelines. If they disagree with principles, share Anthropic's reasoning. Escalate if there's genuine product question. Claude having values is what makes it trustworthy.
94
A nonprofit executive asks: 'Should we build our own AI or use Claude?'
For 99% of nonprofits, use Claude: Building AI requires enormous resources and expertise, pace of improvement means your build will be outdated quickly, your mission isn't AI development — it's your cause, differentiation comes from how you apply AI, not building it. Exceptions might be very large orgs with specific needs no existing tool meets. Even then, prove the use case with Claude first.
95
What's a question I should have asked you but didn't?
'How do you think about ethical implications of AI in the nonprofit sector?' My answer: AI in nonprofits carries unique responsibilities — serving vulnerable populations, limited resources to recover from mistakes, operating on trust. The question isn't just 'can AI help?' but 'will it help equitably, safely, strengthening rather than undermining human relationships?' That's why I'm drawn to Anthropic and beneficial deployments.
96
United Way's API costs for 2-1-1 are tracking $15K over budget ($65K vs $50K). How would you address this with the CTO?
Approach as partnership conversation: 1) Acknowledge concern — budget overruns are real problems for nonprofits. 2) Understand drivers — volume? model selection? inefficient prompts? 3) Identify optimizations — using Haiku where appropriate? prompt caching? batch processing for non-real-time? 4) Reframe around value — cost per call handled vs. human agent cost? 5) Project forward — if we optimize, what's realistic budget? Turn cost concern into cost optimization with clear ROI.
97
The Chief Impact Officer wants quantitative evidence, not anecdotal success stories. How do you build a measurement framework?
Design framework with her, not for her. Efficiency Metrics (Leading): Time saved per task, volume of outputs, user adoption. Quality Metrics (Intermediate): Grant success rates, donor response rates, call resolution rates. Mission Metrics (Lagging): People served through 2-1-1, funds raised, program reach. Propose controlled comparison where possible — chapters using Claude vs. similar chapters not using it. This gives her the evidence-based approach she values.
98
Seattle chapter has 15% seat utilization and hasn't received adequate onboarding. How do you turn this around?
Seattle needs intervention: 1) Diagnose first — call the ED. Is it onboarding? Use case fit? Staff resistance? Leadership buy-in? 2) Leverage Atlanta — connect Seattle ED with Atlanta ED for peer learning. 3) Focused re-onboarding — not generic; identify 2-3 specific pain points and show Claude solving them. 4) Quick win sprint — 30-day focus on one high-impact use case. 5) Success metric — get to 50% utilization in 60 days. Sometimes low adoption means we failed at enablement.
99
The Atlanta chapter ED won a $500K grant with Claude assistance and wants to present at the national conference. How do you leverage this?
Atlanta is gold. Document the case study with specifics (40% faster, $500K won, methodology). Support her presentation with slides, talking points, data. Create replicable assets — her prompt library, training videos packaged for other chapters. Connect her to struggling chapters — peer influence > vendor influence. Internal advocacy — get her in front of SVP of Programs, maybe board. External visibility — feature in Anthropic's beneficial deployments materials. She's a force multiplier.
100
Finance and Legal departments have <15% adoption due to compliance concerns. How do you address this?
Their concerns are legitimate. Don't push adoption; address underlying issues. For Finance: Claude isn't replacing audited financials; it's helping with narrative sections. Show use cases without compliance risk. Propose pilot with clear human review requirements. For Legal: General Counsel's concerns are valid. Offer to help develop AI usage policies. Identify low-risk use cases (research, first drafts, internal docs). Sometimes right answer is slower adoption with proper guardrails.
101
The 2-1-1 Director is impatient with the pace of API implementation. How do you manage expectations?
Validate her urgency while being realistic: 1) Acknowledge stakes — 2-1-1 serves 16M+ people; faster = more people helped. 2) Share progress — 15% of call volume is real progress in 60 days. 3) Address blockers — What's slowing things? CTO concerns? Error rates? Costs? 4) Set realistic milestones — what's achievable in 30/60/90 days? 5) Connect to impact — every percentage point is X more people served. Channel her impatience into advocacy for resources, not frustration.
102
The SVP of Programs needs measurable outcomes before the board meeting in 90 days. What's your 90-day plan?
Days 1-30: Consolidate and Document — quantify existing wins (Atlanta grant, 40% time savings, 2-1-1 results), build measurement framework with Chief Impact Officer, identify 2-3 additional quick wins. Days 31-60: Expand and Measure — activate Foundation Relations team, intensive support for Seattle/Miami, begin tracking metrics. Days 61-90: Package and Present — compile data into board-ready materials, ROI analysis for $2M grant, roadmap for phase 2. Board presentation writes itself if we execute well.
103
How would you help United Way demonstrate ROI on their $2M innovation grant?
Foundation partners want impact, not just activity. Investment: $2M grant + $450K Claude contract = $2.45M. Returns to quantify: Time savings monetized, grant revenue influenced ($500K Atlanta + others), 2-1-1 capacity increase, staff capacity redirected to mission. Story to tell: Before (manual processes limiting reach), After (AI-augmented teams serving more), Future (scalable model for entire 1,100-chapter network). Build one-pager CFO can share with foundation in Q2 2026.
104
The grants-assistant-pilot has minimal usage despite Foundation Relations interest. How do you activate this?
Interest without adoption usually means friction: 1) Talk to team — what's blocking? Time? Training? Trust? 2) Find the moment — when's their next big grant deadline? Meet them there. 3) Show, don't tell — take a real grant in progress and demonstrate value. 4) Connect to Atlanta — they won $500K; can Atlanta's prompt library help? 5) Remove barriers — is pilot hard to access or confusing? Foundation Relations is perfect use case — high-value, measurable. If interested but not using, it's our enablement problem.
105
How would you structure training for a federated nonprofit network with different tech stacks and capacity levels?
Tiered, flexible enablement: Tier 1 Core (Everyone): Basic Claude capabilities, 3-5 universal use cases, on-demand video. Tier 2 Role-Specific (By function): Grant writing for development, program documentation for service teams. Tier 3 Chapter-Led (Peer learning): Atlanta creates content for others, monthly community calls, shared prompt library. Tier 4 Hands-On (Struggling chapters): Dedicated sessions for Seattle/Miami with CSM involvement. Scalable self-service for most, intensive support where needed.
106
The CTO is worried about PII handling in the 2-1-1 helpline use case. How do you address this?
PII in helplines is serious. Validate the concern — 2-1-1 handles sensitive information; caution is appropriate. Review current safeguards — Solutions Engineer implemented data filtering; is it sufficient? Document architecture — what data touches Claude? What's filtered? Share Anthropic's commitments — Enterprise data not used for training, encryption, compliance. Propose additional controls — can we further minimize PII? Audit logging? Offer technical deep-dive with Anthropic security team if needed.
107
General Counsel has concerns about AI-generated content in donor communications. How do you help develop appropriate policies?
Position Claude as tool that fits within their governance: 1) Understand specific concerns — accuracy? disclosure? liability? brand voice? 2) Review what others do — share how similar nonprofits handle AI in communications. 3) Propose policy framework: human review for external comms, disclosure guidelines, prohibited uses, accountability. 4) Offer Anthropic resources — trust & safety documentation. 5) Start with internal — AI for internal docs first, external later. Helping develop good policies makes me a partner.
108
You need concrete outcomes before February 2026 board presentation. What metrics would you prioritize?
Board members want business impact. Must-Have: Time savings quantified in hours and dollars, grant revenue influenced by Claude, 2-1-1 efficiency gains, adoption trajectory showing momentum. Nice-to-Have: User satisfaction, quality improvements (donor response rates, grant success), projected ROI if expanded. The Story Arc: Phase 1 (current) proof of concept with wins, Phase 2 (proposed) expand to 500+ seats, Phase 3 (vision) AI-enabled network serving more communities. Show $2M bet is paying off.
109
How would you create a case study from United Way's success for other federated nonprofits?
Case study should be replicable: Structure: Challenge (federated network, limited resources, need to scale), Solution (Claude Enterprise + API), Implementation (phased rollout with champions), Results (quantified outcomes), Lessons (what worked, what they'd do differently). Key Elements: Atlanta as hero narrative, 2-1-1 as API proof point, measurement framework as rigor, change management approach. Target Audiences: Habitat for Humanity, Boys & Girls Clubs, foundation partners, Anthropic portfolio. Becomes recruiting tool for entire sector.
110
The contract renewal is in 10 months. What does a strong renewal story look like?
10 Months From Now Need: 70%+ seat utilization (from 39%), 2-1-1 API in full production (from 15% pilot), at least 3 pilot chapters at high adoption, measurable ROI documented, executive sponsors advocating, expansion conversation already happening. Renewal Conversation: Not 'should we continue?' but 'how do we expand?' Board has seen results; CFO can justify investment. Risk Factors to Address Now: Seattle/Miami adoption, API cost concerns, impact metrics framework.
111
How would you handle tension between the CTO's skepticism about Claude Enterprise value and the Programs team's enthusiasm?
Classic value perception gap. Understand CTO's perspective: he sees API value (2-1-1 is technical, measurable), doesn't see Enterprise value (seems like 'just chatting'), team is stretched thin. Bridge the gap: Show Enterprise through his lens — productivity gain per dollar. Connect to his goals — less ad-hoc support requests from Programs. Find technical use case for his team — documentation, troubleshooting. Tactical: Invite him to see Atlanta's implementation, have Programs share specific outcomes, ask what would make him a believer.
112
The SVP mentioned potential for 500+ additional seats if pilot chapters succeed. How do you build toward this expansion?
Phase 1 Pilot Success (Now-Month 6): Get 4 of 5 pilots to 60%+ adoption, document replicable playbook from Atlanta, show measurable outcomes for board. Phase 2 Expansion Planning (Month 4-8): Identify next wave of chapters (10-15 high-potential), develop self-service onboarding for scale, build chapter-to-chapter support network. Phase 3 Expansion Proposal (Month 6-9): Business case with pilot data, pricing discussion for volume, implementation plan that doesn't overwhelm CTO. Phase 4 Contract (Month 9-10): Renewal + expansion as single conversation. The 500 seats are earned.
113
How do you balance supporting struggling chapters (Seattle, Miami) vs. doubling down on successful ones (Atlanta)?
Both matter for different reasons. Invest in Atlanta because: Champions create momentum, her success enables expansion, she can help other chapters (force multiplier). Invest in Seattle/Miami because: Failure undermines pilot narrative, can't claim success with 3/5 chapters, reveals what doesn't work. Resource Allocation: Atlanta — enable her to help others (not more CSM time). Seattle/Miami — direct CSM intervention. Chicago/Denver — monitor, light touch. The math: 3/5 success is okay; 4-5/5 is the expansion story.
114
What's your enablement strategy for users who only use Claude 1-2x per month?
Low-frequency users have a habit problem, not value problem. Diagnose Why: Don't know what to use it for? (use case gap) Forget it exists? (visibility) Takes too long? (friction) Tried it, didn't work? (experience) Interventions: Workflow integration (where does Claude fit in routine?), Trigger moments ('Every time you start a donor letter...'), Templates (pre-built prompts), Peer examples, Manager accountability. Success Metric: Move from 1-2x/month to 2-3x/week.
115
The Marketing team wants Salesforce integration for personalized donor communications. How do you respond?
Great sign — they're thinking about scale. Immediate: Validate the need, clarify what they mean (full integration or workaround?). Current Options: MCP connectors may have Salesforce capability, export/import workflows as interim, custom API integration if they have developers. If True Integration Needed: Document request formally, share with Anthropic product team, set expectations on timeline. Reframe: Even without integration, Claude can draft personalized templates. Don't over-promise; do show we're listening.
116
How would you help United Way think about AI governance for a 1,100-chapter network?
Balance consistency and autonomy. Central Governance: Core policies (data privacy, prohibited uses), approved vendors, training requirements, incident reporting. Chapter Flexibility: Choose whether to adopt, customize use cases for local needs, set local priorities. Governance Structure: AI steering committee with chapter representation, shared prompt libraries with local additions, community of practice, regular policy review. Start Small: Pilot chapters inform design, General Counsel leads policy, scale governance with adoption.
117
The CFO needs to report on innovation grant ROI to foundation partner in Q2 2026. What should that report include?
Structure: 1) Investment Summary — $2M grant allocation, $450K Claude contract, staff time. 2) Quantified Returns — time savings monetized, revenue influenced, cost avoidance, capacity created. 3) Mission Impact — people served through 2-1-1, programs improved, chapters enabled. 4) Qualitative Wins — Atlanta case study, staff testimonials, partner feedback. 5) Path Forward — expansion plan, sustainability beyond grant, lessons for other nonprofits. Report should make foundation proud and eager to share success.
118
How do you handle Atlanta creating training materials that may not align with Anthropic's best practices?
Champion-created content is valuable but needs quality control. First, Appreciate: Atlanta taking initiative is exactly what we want; peer content resonates better. Then, Partner: Ask to review materials collaboratively (not audit), offer to enhance with Anthropic best practices, co-brand if appropriate. Quality Considerations: Are prompts following safety guidelines? Guidance accurate about capabilities/limitations? Appropriate expectations? Outcome: Atlanta's content becomes 'certified' for network use, she feels valued, quality maintained without stifling innovation.
119
The 2-1-1 helpline serves 16M+ people annually. What's the potential impact if the API integration succeeds at scale?
Crown jewel of implementation. Current: 15% of call volume with Claude assistance. Potential at Scale: Reduced wait times (critical for people in crisis), more consistent resource matching (better outcomes), staff capacity for complex cases (AI handles routine), extended hours coverage, better data for community needs assessment. Quantified Impact: If 50% volume with AI assistance, 20% handle time reduction, 15% match accuracy improvement = thousands more people served annually. Every second saved on routine calls is available for someone in crisis. This is AI's highest purpose.
120
If United Way succeeds, how would you replicate this playbook for other federated nonprofits?
United Way becomes the template. What's Replicable: Phased rollout (central + pilot first), champion identification/enablement, measurement framework for federated impact, peer-to-peer learning model, governance balancing central/local. What's Customizable: Specific use cases, tech stack integration, chapter autonomy levels, pricing. Playbook Components: Assessment template, champion criteria, pilot selection process, training curriculum, measurement framework, governance template, case study format. Go-to-Market: United Way executive speaks at conferences, joint webinars, foundation partners spreading word. This becomes proof point for entire sector.
121
A nonprofit has a $20K annual budget for AI. What would you recommend?
At $20K/year, I'd recommend Claude for Enterprise Team plan or a hybrid approach. Option 1: Enterprise seats — at nonprofit pricing (~$1,800/seat), that's ~11 seats for the team. Option 2: API-first — $20K in API credits goes far with Sonnet ($3/$15) or Haiku ($1/$5). At Sonnet rates, that's roughly 1.3B input tokens or 1.3M output tokens annually. Option 3: Hybrid — 5 Enterprise seats (~$9K) + $11K API budget for custom integrations. Recommend starting with Enterprise for adoption, add API for specific automation use cases.
122
A customer says Claude is too expensive. How do you respond?
First, understand the comparison — expensive vs. what? Then reframe around value and optimization: 1) Model selection: Are they using Opus when Sonnet would suffice? Haiku for high-volume? 2) Cost optimization: Prompt caching can save 90% on repeated contexts. Batch processing saves 50%. 3) Total cost: Factor in time saved, iteration cycles reduced, quality improvements. 4) ROI framing: If Claude saves 10 hours/week at $50/hour, that's $26K/year in productivity vs. maybe $5K in API costs. 5) Nonprofit pricing: Enterprise has nonprofit discounts. The cheapest AI is the one that actually works.
123
Walk me through how you'd help a customer estimate their API costs.
Step-by-step: 1) Identify use cases — what tasks, how often, what volume? 2) Estimate tokens per task — input (prompt + context) and output (response). Rule of thumb: 1 token ≈ 4 characters or 0.75 words. 3) Select appropriate model — Haiku for simple/high-volume, Sonnet for balanced, Opus for complex. 4) Calculate monthly volume — tasks × frequency × tokens. 5) Apply base pricing — e.g., 10M input + 2M output on Sonnet = $30 + $30 = $60/month. 6) Factor optimizations — prompt caching, batch processing. 7) Add buffer — typically 20-30% for growth and variance. Offer to revisit after 30 days with actual usage data.
124
How does batch processing work and when should a nonprofit use it?
Batch API processes requests asynchronously with 50% discount on all tokens. How it works: Submit batch of requests, get results within 24 hours (usually faster). When to use: Processing historical data (years of donor records), bulk content generation (thank-you letters for all donors), document analysis (reviewing all grant reports), any task where real-time isn't needed. When NOT to use: Interactive applications, time-sensitive tasks, anything needing immediate response. For nonprofits, batch processing is perfect for year-end reporting, annual donor communications, or processing backlogs.
125
A customer asks about Claude's context window. Explain it and why it matters.
Context window = everything Claude can 'see' at once. Claude offers: Standard: 200K tokens (~150K words, ~500 pages). Enterprise: 500K tokens. Sonnet 4.5 beta: 1M tokens (at premium pricing). Why it matters for nonprofits: Analyze entire grant applications in one go. Process full annual reports without chunking. Review complete policy manuals. Handle long conversation histories. Compare: GPT-4 is 128K, Gemini is 1M but at premium tiers. Claude's 200K standard is generous for most nonprofit use cases without extra cost.
126
What's extended thinking mode and when would a nonprofit use it?
Extended thinking lets Claude reason step-by-step internally before responding. How it works: Claude generates a 'thinking' content block showing its reasoning process, then provides the final answer. Tokens used for thinking are billed as output tokens. When to use: Complex grant strategy questions, multi-factor program evaluation, budget analysis with many variables, any decision requiring careful reasoning. Minimum budget: 1,024 tokens. Start there and increase if needed. For nonprofits, this is valuable when you need Claude to really think through a problem, not just generate quick content.
127
Explain Claude's vision capabilities for document processing.
Claude can process images, PDFs, charts, and diagrams natively. How it works: Upload image or PDF, Claude 'sees' it as a human would — understands layout, formatting, tables, even handwriting. PDF processing: Claude sees each page as an image, understanding structure and context. For nonprofits: Process scanned grant applications, extract data from infographics in reports, analyze photos from field work, read legacy documents, review signed contracts. Key advantage: No OCR preprocessing needed — Claude handles it natively. Combined with long context, you can process entire document packages.
128
What is MCP (Model Context Protocol) and how would a nonprofit use it?
MCP is a standard protocol for connecting Claude to external data sources and tools — like USB for AI. How it works: Pre-built connectors let Claude access Google Drive, Slack, GitHub, databases, and more. Claude can search, read, and potentially write to these systems. For nonprofits: Connect to Salesforce for donor data. Access Google Drive for organizational documents. Search Slack for institutional knowledge. Pull from program databases. Why it matters: Instead of copy-pasting data into Claude, Claude accesses it directly. Reduces friction, increases accuracy, enables more sophisticated workflows.
129
What's Claude Code and who would use it?
Claude Code is a command-line tool for developers — AI-powered coding assistance directly in the terminal. What it does: Understands entire codebases, helps with complex coding tasks, debugging, documentation, onboarding. Who uses it: Development teams building custom applications, IT staff maintaining systems, technical nonprofits with engineering resources. For nonprofits: If you have developers building custom tools (like a 2-1-1 system), Claude Code accelerates their work. It's bundled with Team and Enterprise plans — no extra cost. Not relevant for non-technical staff, but valuable if you have technical capacity.
130
What's the difference between Claude Pro, Max, Team, and Enterprise plans?
Pro ($20/month): Individual user, 5x more usage than free, access to all models including Opus. Max ($100-200/month): Power users, highest individual limits, extended features. Team ($30/seat/month, 5 min): Small teams, shared workspace, basic admin controls. Enterprise (custom pricing): Large organizations, SSO, SCIM, audit logs, 500K context, admin dashboard, compliance features, no training on data. For nonprofits: Start with Team for small groups. Move to Enterprise when you need SSO, compliance, or scale. Nonprofit discounts typically available on Enterprise.
131
A customer is evaluating Claude vs. ChatGPT vs. Gemini. How do you position Claude?
Position based on their priorities: For safety-conscious orgs: Claude's Constitutional AI is the differentiator — values are explicit, auditable, built by a company whose mission is safe AI. For document-heavy work: Claude's 200K context standard beats GPT's 128K. Process more without chunking. For enterprise needs: Claude's admin controls, audit logs, and data privacy commitments. No training on Enterprise data. For quality: Claude excels at nuanced instructions, complex reasoning, saying 'I don't know' rather than hallucinating. For cost: Comparable to GPT-4, cheaper than GPT-4o on some tasks. Haiku competitive with budget options. Ask what matters most to them, then show how Claude wins on that dimension.
132
What's Claude's weakness compared to GPT-4?
Be honest: Speed: GPT-4o is faster for real-time applications. Claude optimizes for quality over speed. Ecosystem: OpenAI has broader third-party integrations and a larger developer community. Brand recognition: More people have heard of ChatGPT. May be easier internal sell. Multimodal: GPT-4 has more advanced image generation and voice capabilities. Price perception: On paper, some GPT models look cheaper (though total cost often comparable). How to respond: Acknowledge these, then redirect to where Claude wins — safety, context length, instruction following, enterprise features. The right model depends on the use case.
133
What's Claude's weakness compared to Gemini?
Be honest: Price: Gemini Flash is cheaper for high-volume, simple tasks. Google integration: If you're deep in Google Workspace, Gemini has native advantages. Context window: Gemini offers 1M tokens (though at premium pricing). Distribution: Gemini is embedded in Google Search, reaching massive scale. How to respond: Gemini is a strong product, especially for Google-native orgs. Claude wins on: safety and alignment, instruction following, enterprise controls, and for organizations that want a dedicated AI partner rather than a feature of their productivity suite.
134
A nonprofit had a bad experience with ChatGPT hallucinating. How do you rebuild trust in AI?
Validate first: 'That's a real problem, and it's why many organizations are cautious.' Then differentiate: Claude is specifically designed to reduce hallucinations — it's 3-4x more likely to say 'I don't know' than make something up. Constitutional AI trains Claude to be honest. But no AI is perfect. The solution isn't avoiding AI — it's designing workflows that account for limitations: Use Claude for drafts, not final sources of truth. Build in human review for anything high-stakes. Start with low-risk use cases to build confidence. Verify critical facts independently. Offer a pilot: Let them test Claude on real tasks with low stakes. Experience builds trust faster than promises.
135
A customer is worried about vendor lock-in with Claude. How do you respond?
Legitimate concern. Address it directly: API standards: Claude's API follows similar patterns to other LLMs. Prompts and integrations are largely portable. Data ownership: Your data is yours. Export anytime. No training on Enterprise data. Multi-vendor strategy: Many organizations use multiple AI providers for different tasks. Claude for some, others for others. Switching costs: Mainly prompt optimization and integration work — not insignificant but not prohibitive. Anthropic's incentive: We want you to stay because Claude is valuable, not because you're trapped. Best practice: Document your prompts and use cases. This makes any future migration easier and is good practice regardless.
136
A deployment is failing to get adoption. Walk me through your diagnosis.
Systematic diagnosis: 1) Usage data: Who's using it? Who isn't? How often? What for? 2) User interviews: Talk to adopters AND non-adopters. What's working? What's not? 3) Common failure modes: Access issues (can't log in, don't know how), Value unclear (don't see the point), Friction (takes too long, too complicated), Trust (worried about accuracy), Time (too busy to learn), Leadership (manager doesn't support). 4) Root cause: Is it training? Use case fit? Change management? Technical barriers? 5) Intervention: Match solution to problem. Training won't fix access issues. Cheerleading won't fix real friction. 6) Quick wins: Find one thing that works and amplify it.
137
API costs came in 2x higher than projected. How do you handle this with the customer?
Acknowledge, diagnose, optimize, prevent: 1) Acknowledge: 'I understand this is frustrating. Let's figure out what happened and fix it.' 2) Diagnose: Pull usage data. What's driving costs? Model selection? Volume? Output length? Unexpected use cases? 3) Quick wins: Switch to smaller models where appropriate. Implement prompt caching. Use batch processing for non-urgent tasks. Shorten prompts and outputs. 4) Optimize architecture: Are they sending unnecessary context? Can they filter before sending to Claude? 5) Set guardrails: Usage alerts, spending caps, model restrictions by use case. 6) Prevent recurrence: Monthly cost reviews, usage dashboards, clear ownership. Turn the crisis into a partnership moment — you're solving this together.
138
Claude gave a customer wrong information that caused a problem. How do you handle it?
Handle with care: 1) Acknowledge impact: 'I'm sorry this happened. Let's understand what went wrong and make sure it doesn't happen again.' 2) Don't be defensive: AI can make mistakes. That's reality. 3) Investigate: What was the prompt? What was the response? What was the context? Was this a hallucination, outdated information, or misinterpretation? 4) Immediate fix: Address the specific problem they encountered. 5) Prevent recurrence: Better prompting? Human review process? Different use case boundaries? 6) Document and share: If it's a pattern, report to Anthropic product team. 7) Rebuild trust: Follow up, check in, show you care about their success. The relationship matters more than being right.
139
A customer wants to use Claude for medical advice. How do you handle this?
High-stakes use case requiring careful navigation: Clarify scope: What exactly do they want? Patient triage? Health education? Clinical documentation? Administrative? Risk assessment: Direct medical advice = high risk. Documentation = lower risk. Draw clear lines: Claude should NOT make clinical diagnoses, prescribe treatments, or replace medical professionals. Claude CAN help with: health literacy content, administrative tasks, research synthesis, documentation assistance, appointment preparation. Recommend safeguards: Always human review for anything patient-facing. Clear disclaimers. Clinician oversight. Compliance check: HIPAA requirements, state regulations, liability considerations. Connect to experts: Anthropic has healthcare-specific guidance. Offer to connect them with the right resources.
140
A customer wants to use Claude for financial advice. How do you handle this?
Similar to medical — high stakes, requires care: Clarify scope: Portfolio management? Financial literacy? Accounting? Compliance? Risk assessment: Direct investment advice = regulated, high risk. Education = lower risk. Draw clear lines: Claude should NOT recommend specific investments, predict market movements, or replace licensed advisors. Claude CAN help with: financial literacy content, research and analysis, report drafting, budgeting tools, explaining concepts. Recommend safeguards: Human review for anything client-facing. Clear disclaimers ('not financial advice'). Compliance officer sign-off. Regulatory awareness: SEC, FINRA, state regulations may apply. Best practice: Position Claude as a tool that helps financial professionals be more efficient, not a replacement for licensed advice.
141
A customer wants Claude to make automated decisions without human review. What do you advise?
Proceed with caution: Assess the stakes: What decisions? What's the impact of errors? Who's affected? Low-stakes automation (fine): Categorizing support tickets. Drafting initial responses for human review. Sorting documents. High-stakes automation (caution): Anything affecting people's benefits, employment, health, finances. Recommend human-in-the-loop: For consequential decisions, Claude should inform and support human decision-makers, not replace them. Why: AI can be wrong. Accountability matters. Trust is hard to rebuild. Bias can compound. Practical middle ground: Claude can do 90% of the work and flag for human approval. Efficiency gains without full automation risk. Frame as 'augmentation, not replacement.'
142
How do you handle a customer whose use case might violate Anthropic's usage policies?
Navigate carefully: 1) Understand intent: What are they actually trying to accomplish? Sometimes the underlying need is legitimate even if the stated approach isn't. 2) Know the policies: Anthropic's acceptable use policy prohibits certain applications (weapons, harassment, deception, etc.). 3) Be direct but constructive: 'I want to help you achieve your goal, but this specific approach raises concerns. Let me explain why and suggest alternatives.' 4) Find alternatives: Often there's a way to meet the legitimate need within guidelines. 5) Escalate if needed: If uncertain, loop in Anthropic's trust and safety team. 6) Document: Keep records of the conversation and guidance given. Maintain the relationship while maintaining integrity.
143
What are Anthropic's content policies that a CSM should know?
Key policies to know: Prohibited: Weapons/violence, CSAM, harassment, deception/fraud, malware, unauthorized access. Restricted: Medical/legal/financial advice (with appropriate caveats), political content, adult content. Enterprise considerations: Customers are responsible for their end users' compliance. Admin controls help enforce policies. Audit logs track usage. Safety by design: Constitutional AI builds values into the model itself. Claude will refuse harmful requests. For beneficial deployments: Most nonprofit use cases are well within guidelines. Focus on the positive — Claude is designed for helpful, harmless, honest interactions.
144
A customer asks about Claude's training data. What do you tell them?
What we know: Claude is trained on a diverse dataset including web content, books, articles, and code. Training data has a cutoff — Claude doesn't know about very recent events without web search. Anthropic doesn't use Enterprise customer data for training (key privacy commitment). What we don't share: Specific training data composition isn't public (common across AI companies). How to respond: Focus on what matters to them — privacy (their data isn't used), recency (knowledge cutoff exists, web search helps), quality (extensive curation and safety training). If they need specifics, offer to connect them with Anthropic's documentation or technical team.
145
A customer asks if Claude will remember their previous conversations. What do you explain?
By default, no: Each conversation starts fresh. Claude doesn't have persistent memory across sessions. Within a conversation: Claude remembers everything in the current context window (up to 200K tokens). Long conversations may eventually exceed this. Projects feature: In Claude.ai, Projects let you store persistent context that's loaded into each conversation. Enterprise features: Some memory and personalization features may be available. API considerations: Developers can build memory by storing and re-injecting conversation history. For nonprofits: Projects are the key feature for maintaining organizational context. Set up Projects with key documents, guidelines, and background for consistent Claude interactions.
146
What's the difference between Claude.ai and the API?
Claude.ai: Chat interface, accessible via browser or app. For end users having conversations. Projects, Artifacts, built-in features. Subscription-based (Pro, Max, Team, Enterprise). API: Programmatic access for developers. Build Claude into applications, automate workflows, custom integrations. Pay-per-token pricing. Requires technical implementation. For nonprofits: Most staff use Claude.ai for daily work. API is for building custom tools (like a 2-1-1 integration). Many organizations use both — Claude.ai for humans, API for systems. Enterprise plan includes both.
147
How does Claude handle data privacy for Enterprise customers?
Key commitments: No training on customer data: Enterprise conversations aren't used to improve Claude. Data encryption: In transit and at rest. Data retention: Configurable retention policies. SOC 2 Type II: Anthropic is certified. Regional options: Data residency considerations for some deployments. Admin controls: Who can access what, audit logs, SSO. For nonprofits: This matters for donor data, beneficiary information, and compliance requirements. Enterprise gives you the controls you need to meet your obligations. Be ready to connect customers with Anthropic's security documentation or trust team for detailed questions.
148
A customer needs to meet specific compliance requirements (HIPAA, FERPA, etc.). What do you say?
Don't overpromise — compliance is complex: HIPAA (healthcare): Anthropic offers BAA (Business Associate Agreement) for Enterprise customers. Required for PHI. FERPA (education): Student data protections. Enterprise controls help, but implementation matters. SOC 2: Anthropic is SOC 2 Type II certified. GDPR: Data processing agreements available. Regional considerations. General guidance: 1) Understand their specific requirements. 2) Connect them with Anthropic's compliance documentation. 3) Involve their legal/compliance team. 4) Compliance is shared responsibility — Anthropic provides controls, customer implements properly. Offer to facilitate conversations with Anthropic's security team for detailed requirements.
149
What integrations does Claude Enterprise offer?
Native integrations: Google Drive, Gmail, Google Calendar. Microsoft 365 (in development). Slack. GitHub. MCP connectors: Extensible protocol for connecting to additional data sources. Custom options: API for building custom integrations. What this means for nonprofits: Connect Claude to your document repositories. Access organizational knowledge without copy-pasting. Enable Claude to help with real data. If they need an integration that doesn't exist: Document the request. Share with Anthropic product team. Explore MCP or API alternatives.
150
A nonprofit executive asks: 'What happens to our data if we stop using Claude?'
Clear answer: Your data is yours: Export anytime. No training on Enterprise data means no trace in the model. Upon termination: Data is deleted according to your retention settings and Anthropic's policies. No lock-in: Your prompts, documents, and workflows can be moved to other platforms. Practical steps: Export any saved Projects or content before ending subscription. Download conversation history if needed. Ensure local copies of any important outputs. Reassurance: Anthropic's business model is providing valuable AI services, not monetizing your data. Your data leaving when you leave is a feature, not a bug.
151
Explain the difference between input tokens and output tokens.
Input tokens: What you send to Claude — your prompt, uploaded documents, conversation history, system instructions. You're charged when Claude 'reads' them. Output tokens: What Claude generates — the response, code, analysis. Typically more expensive (often 5x input price). Why it matters: A long document with a short summary = high input, low output. A short prompt with a long essay = low input, high output. Optimization implications: Reduce input by trimming unnecessary context. Reduce output by asking for concise responses. Cache inputs that repeat. Choose models based on which side dominates your use case.
152
How do you explain tokens to someone who's never heard of them?
Simple analogy: Tokens are the 'words' AI uses to read and write. But they're not exactly words — they're word pieces. Rule of thumb: 1 token ≈ 4 characters or ¾ of a word. Examples: 'Hello' = 1 token. 'Anthropic' = 2-3 tokens. A page of text ≈ 400-500 tokens. Claude's 200K context ≈ 500 pages. Why it matters: Tokens determine what Claude can 'see' at once (context window) and what you pay (API costs). For nonprofits: You probably don't need to think about tokens day-to-day with Claude.ai subscriptions. It matters more for API usage and understanding limits.
153
A customer wants to know Claude's uptime and reliability.
What to share: Anthropic maintains high availability for production services. Status page available for real-time information. Enterprise SLAs may include uptime commitments. Be honest about limitations: Like any cloud service, outages can happen. API rate limits exist to ensure stability. Heavy usage periods may see slowdowns. Practical guidance: Build workflows that gracefully handle unavailability. Have backup processes for critical functions. Don't make Claude a single point of failure for mission-critical operations. For Enterprise: Discuss SLA terms during contract negotiation. Understand what's guaranteed vs. best effort.
154
What should a customer do if Claude isn't performing well on their use case?
Systematic troubleshooting: 1) Check the prompt: Is it clear? Specific? Does it include enough context? Examples of good output? 2) Try a different model: Maybe Opus for complex reasoning, Sonnet for balanced tasks, Haiku if speed matters. 3) Adjust temperature: Lower for consistency, higher for creativity. 4) Add structure: Break complex tasks into steps. Use clear formatting. 5) Provide examples: Show Claude what good looks like. 6) Iterate: Prompt engineering is iterative. Small changes can have big impacts. If still struggling: Share the use case with me. We can workshop it together. Sometimes a fresh perspective helps. Document and share feedback with Anthropic.
155
How do you help a customer who's overwhelmed by Claude's capabilities?
Start simple: 1) Identify ONE high-value, low-risk use case. Not ten. One. 2) Get a quick win: Something that shows value in the first week. 3) Build from there: Once they trust it, expand to adjacent use cases. 4) Resist the urge to show everything: Features are great, but overwhelm kills adoption. Practical approach: 'What's taking too much of your time right now?' Start there. Show Claude helping with that specific thing. Let them experience value before exploring more. The goal is building a habit, not demonstrating features.
156
A customer says their team doesn't have time to learn a new tool. How do you respond?
Empathize, then reframe: 'I hear that — everyone's stretched thin. But here's the thing: if Claude can save you 5 hours a week, isn't it worth investing 2 hours to learn it?' Make it easy: Offer focused, 30-minute training on ONE use case. Provide templates they can use immediately. Show, don't tell — do a live demo with their actual work. Peer pressure (positive): 'Your colleague in Marketing is saving 3 hours a week on donor communications. Want me to connect you?' Reduce friction: Can we get them started in 15 minutes, not 2 hours? What's the absolute minimum they need to know? Sometimes the answer is: Start with one person who has time, let them become the internal champion, then spread organically.
157
How would you run a pilot program for Claude at a nonprofit?
Structure for success: 1) Define scope: Which team? Which use cases? What timeframe? (Recommend 30-60 days). 2) Set success metrics: Before starting, agree on what 'success' looks like. Time saved? Tasks completed? User satisfaction? 3) Identify champions: 2-3 enthusiastic early adopters who will drive it. 4) Enable properly: Training, documentation, prompt templates, easy access. 5) Check in regularly: Weekly touchpoints during pilot. What's working? What's not? 6) Gather data: Usage metrics, user feedback, outcome measures. 7) Decision point: At end of pilot, evaluate against success metrics. Go/no-go on expansion. 8) Document learnings: What worked, what didn't, what to do differently at scale.
158
A customer's IT team is blocking Claude adoption. How do you help?
Understand their concerns: Security? Privacy? Cost? Control? Shadow IT? All legitimate. Address specifically: Security — share Anthropic's security documentation, SOC 2, encryption. Privacy — no training on data, configurable retention. Cost — help model and control API spend. Control — Enterprise admin features, audit logs, SSO. Compliance — connect with Anthropic's compliance resources. Tactical moves: Offer to present directly to IT. Provide a security questionnaire response. Propose a limited pilot with extra monitoring. Find an IT champion who sees the value. Sometimes IT needs to feel heard. Involve them early, not as a blocker to overcome but as a partner to enable.
159
How do you handle a customer who wants features that don't exist yet?
Validate, document, set expectations: 1) Understand deeply: What specifically do they need? Why? What would it enable? 2) Check alternatives: Can we achieve the goal with existing features differently? 3) Be honest: 'That feature doesn't exist today. Here's what we can do instead.' 4) Document the request: Formally capture it for the product team. 5) Share context: 'Other customers have asked for similar things. I'll advocate for this.' 6) Set expectations: 'I can't promise a timeline, but your feedback matters.' 7) Follow up: If the feature ships, let them know. Shows you were listening. Never promise features you can't deliver. But showing you care about their needs builds trust.
160
How do you prepare for an Executive Business Review with a nonprofit?
Preparation: 1) Pull all data: Usage metrics, adoption rates, feature utilization. 2) Gather outcomes: Time saved, tasks completed, qualitative wins. 3) Identify risks: Low adoption areas, concerns raised, blockers. 4) Prepare recommendations: What should they do next? More users? New use cases? 5) Know your audience: Who's in the room? What do they care about? Structure: Executive summary (1 page). Adoption metrics with trends. Outcomes and ROI. Risks and mitigation. Recommendations. Q&A. Key principle: Lead with value delivered, not features used. Executives care about outcomes, not activity.
161
A nonprofit is considering canceling. How do you save the account?
Understand before solving: 1) Why? Budget? Adoption? Value not realized? Champion left? Leadership change? 2) Listen fully. Don't interrupt with solutions. For each scenario: Budget: Can we right-size? Fewer seats? Different plan? Adoption: What blocked it? Can we re-onboard? Value: Did they give it a real chance? Can we show ROI? Champion left: Who else can we enable? Leadership change: Can we re-pitch the value? If it's truly not working: Sometimes the answer is 'not right now.' Leave the door open. Stay professional. A good departure can lead to a return later. Never: Beg, discount desperately, or make promises you can't keep. Maintain the relationship even if you lose the deal.
162
What makes a successful nonprofit Claude deployment?
Key success factors: 1) Executive sponsorship: Leadership believes in and supports adoption. 2) Clear use cases: Specific problems Claude is solving, not vague 'AI adoption.' 3) Enabled champions: Internal advocates who drive peer adoption. 4) Proper training: Not just features, but workflows and best practices. 5) Quick wins: Early value demonstrated within first 2 weeks. 6) Measurement: Tracking outcomes, not just usage. 7) Iteration: Willingness to adjust based on what works. 8) Realistic expectations: Claude augments, doesn't replace. Needs human partnership. 9) Integration: Fits into existing workflows, not a separate tool. The CSM's job is to orchestrate all of these, not just handle support tickets.
163
How do you build a business case for Claude at a nonprofit?
Structure: 1) Current state: How are they doing things today? What's the cost (time, money, opportunity)? 2) Proposed solution: How would Claude help? Which use cases? 3) Investment: What does Claude cost? Implementation time? Training? 4) Returns: Time saved (quantified in hours and dollars). Capacity created (what could they do with that time?). Quality improvements. Revenue influenced (grants, donations). 5) Timeline: When do benefits start? When is break-even? 6) Risk mitigation: What if it doesn't work? How do we know? Make it their language: Connect to their strategic priorities. Use their metrics. Reference their constraints. The best business case is one they could present to their board.
164
How would you handle a customer who's using Claude for something you disagree with but isn't against policy?
Navigate thoughtfully: 1) Separate personal from professional: Your job is to help them succeed within Anthropic's guidelines. 2) If it's against policy: That's clear — redirect to compliant alternatives. 3) If it's allowed but you're uncomfortable: Reflect on why. Is this a genuine ethical concern or personal preference? 4) If genuine ethical concern: You can share your perspective thoughtfully, then respect their decision. 'I want to share a consideration...' 5) If just preference: Keep it to yourself. Not your call. 6) Escalate if needed: If you're genuinely uncertain about policy, ask internally. The line is: You serve the customer within Anthropic's guidelines. You're not the policy maker, but you're also not a robot. Bring your judgment, but know your role.
165
What's something about Claude that surprises customers?
Common surprises: 1) How good it is at nuance: Claude follows complex instructions better than expected. 2) Long context actually works: Processing 100 pages at once without quality degrading. 3) It admits uncertainty: 'I don't know' is refreshing versus confident hallucinations. 4) Vision capabilities: People don't expect it to understand charts and documents visually. 5) Constitutional AI in practice: Claude declines harmful requests in principled ways. 6) Speed of improvement: Models get better frequently. What didn't work 6 months ago might work now. Use these as teaching moments: 'Try this — I think you'll be surprised.' Positive surprises build enthusiasm and adoption.
166
What's your philosophy on being a great CSM?
Core beliefs: 1) Outcomes over activity: Success is customer results, not hours logged or features demo'd. 2) Partnership over vendor: I succeed when they succeed. Same team. 3) Honesty over comfort: Tell them what they need to hear, not what they want to hear. 4) Proactive over reactive: Anticipate problems, don't just fix them. 5) Enablement over dependency: Build their capability, not reliance on me. 6) Long-term over short-term: The relationship matters more than any single deal. 7) Curiosity over assumptions: Every customer is different. Keep asking questions. For beneficial deployments specifically: The mission matters. Helping nonprofits use AI well isn't just business — it's impact.
Anthropic Interview
flashcards
Decks in class (4)
# Cards
