Communication
| Task | Approach | AI Automation Method | Suggested Tools / Notes |
|---|---|---|---|
| Drafting emails to clients or colleagues | Both | Single prompt for one-off emails. RAG for brand voice, templates, and prior correspondence context. "Draft a professional follow-up email to a client who missed our 2pm call. Tone: warm but direct. 3 sentences." → RAG: Email Drafting (brand voice + templates) | Gmail / Outlook AI plugins Custom GPT or Claude Project |
| Summarising phone call transcripts | Single Prompt | Paste transcript; prompt the model to extract action items, key decisions, and follow-ups. "Summarise this call transcript. Output: (1) 3-sentence summary, (2) action items with owners, (3) open questions." | Otter.ai, Fireflies, Fathom + GPT/Claude API |
| Responding to common customer queries | RAG | Ground responses in your product documentation, pricing, and policy docs so answers are accurate and on-brand. → RAG: Customer Service Knowledge Base | Intercom Fin, Zendesk AI Custom chatbot on LlamaIndex / LangChain |
| Handling complaints | Both | Single prompt for empathetic reply structure. RAG to verify entitlements, warranties, and escalation procedures. "Draft a de-escalating reply to this customer complaint. Acknowledge the issue, apologise sincerely, and offer a clear next step." | CRM-integrated AI assistant |
| Drafting internal status updates | Single Prompt | Paste bullet-point notes; ask the model to write a structured update for Slack, Teams, or email. "Turn these notes into a concise project status update for a non-technical manager. Use: Status, Progress, Blockers, Next Steps." | Slack AI, Teams Copilot |
| Proofreading and tone-checking outgoing messages | Single Prompt | Paste draft; ask model to flag tone issues, grammar errors, and clarity problems. "Review this email for clarity, grammar, and professional tone. Flag any phrases that may come across as passive-aggressive or unclear." | Grammarly Business, Claude, Copilot |
Scheduling & Events
| Task | Approach | AI Automation Method | Suggested Tools / Notes |
|---|---|---|---|
| Drafting scheduling emails & meeting requests | Single Prompt | Provide context (participants, purpose, preferred times); model drafts a polite coordination email. "Write a meeting request email to three senior managers to align on Q3 budget. Suggest three times next week (Mon–Wed). Professional tone." | Calendly AI, Microsoft Copilot Reclaim.ai |
| Building event run-of-show / agenda | Single Prompt | Describe event type, duration, attendees; model produces a timed agenda draft. "Create a run-of-show for a 2-hour all-hands meeting for 150 employees. Include welcome, 3 department updates, Q&A, and closing." | Google Docs / Notion + AI plugin |
| Planning large all-hands or town-halls | Both | Single prompt for agenda and logistics checklist. RAG to pull in approved messaging, leadership talking points, and previous meeting templates. → RAG: Event & Meeting Planning | Notion AI, Confluence AI |
| Automated meeting notes & minutes | Single Prompt | Feed transcript to model with a structured output prompt. "Convert this meeting transcript into formal minutes. Format: Attendees, Agenda Items, Decisions Made, Action Items (owner + due date)." | Fathom, Otter, Fireflies Teams/Zoom AI summaries |
Onboarding
| Task | Approach | AI Automation Method | Suggested Tools / Notes |
|---|---|---|---|
| Preparing onboarding document packets | RAG | AI assembles role-specific onboarding packs from HR policy library, benefits docs, and team-specific guides. → RAG: Onboarding Document Assembly | Notion AI, SharePoint Copilot Custom HR chatbot |
| Drafting welcome emails to new hires | Single Prompt | Provide new hire name, role, start date, manager; model drafts a warm, informative welcome email. "Draft a welcome email to a new marketing coordinator starting Monday. Mention their manager is Sarah, office is on Floor 3, and they should bring ID. Friendly and concise." | HRIS integrations (Rippling, BambooHR) |
| Answering new employee HR questions | RAG | HR chatbot grounded in your employee handbook, benefits summaries, and policy documents. → RAG: Onboarding Document Assembly | Leena AI, Workday Assistant Custom chatbot on handbook PDFs |
| Creating software tutorial guides | Both | Single prompt for generic SaaS tools. RAG for internal or custom tools using internal documentation. "Write a step-by-step guide for a new employee on how to request time off in Workday. Assume no prior experience. Include screenshots placeholders." | Notion, Confluence, Guru |
Bookkeeping & Finance
| Task | Approach | AI Automation Method | Suggested Tools / Notes |
|---|---|---|---|
| Categorising transactions from a CSV export | Single Prompt | Paste transaction list; ask model to suggest expense categories per line. "Categorise each of these transactions as: Office Supplies, Travel, Software, Meals, or Other. Return as a table with your reasoning for ambiguous items." | Excel Copilot, QuickBooks AI Code Interpreter |
| Drafting invoices | Both | Single prompt for one-off invoices. RAG to pull client-specific rates, payment terms, and billing history. → RAG: Invoice & Billing Automation | FreshBooks, Xero, Wave AI |
| Summarising financial reports for leadership | Single Prompt | Paste P&L or balance sheet data; ask for an executive-readable narrative summary. "Summarise this P&L statement for a non-financial audience. Highlight top 3 variances vs last quarter and flag any concerns. 150 words max." | Excel Copilot, Tableau AI ChatGPT Code Interpreter |
| Extracting data from receipts & invoices | Single Prompt | Use a vision-capable model to extract vendor, date, amount, and category from uploaded images. "Extract: vendor name, date, total amount, currency, and expense category from this receipt image. Output as JSON." | Expensify AI, Dext, Klippa GPT-4o / Claude vision |
IT & Operations
| Task | Approach | AI Automation Method | Suggested Tools / Notes |
|---|---|---|---|
| Answering internal IT helpdesk tickets | RAG | AI triage and first-response grounded in your IT knowledge base, device policies, and common-fix documentation. → RAG: IT Helpdesk Assistant | ServiceNow AI, Freshservice AI Atlassian Intelligence |
| Writing equipment & supply order requests | Single Prompt | Describe need; model drafts a formatted purchase request with justification. "Write a purchase request for 5 ergonomic keyboards at ~$80 each. Include business justification (RSI complaints) and cost-center code: 4420." | Procurement systems (Coupa, SAP Ariba) |
| Migrating content from analogue to digital | Single Prompt | After OCR, use prompts to clean, format, and structure extracted text into digital documents. "Clean and structure this OCR-extracted text into a formatted policy document with headings, numbered sections, and a table of contents." | Adobe Acrobat AI, Google Docs AI AWS Textract + Claude |
| Drafting SOP and system setup documentation | Single Prompt | Describe a process step-by-step; model turns it into a formatted SOP. "Turn these notes into a formal Standard Operating Procedure for setting up a new employee's laptop. Include sections: Purpose, Scope, Prerequisites, Steps, Troubleshooting." | Confluence, Notion, SharePoint |
Knowledge Worker Tasks
| Task | Approach | AI Automation Method | Suggested Tools / Notes |
|---|---|---|---|
| Contract review & redlining | RAG | Ground the model in your standard contract playbook and approved clause library to flag deviations and suggest redlines. → RAG: Contract Review Playbook | Harvey AI, Ironclad AI Spellbook, custom LLM on contract PDFs |
| Drafting job descriptions | Both | Single prompt for general roles. RAG to match internal leveling frameworks, approved language, and pay-band context. → RAG: HR Policy & Leveling Framework "Write a job description for a mid-level UX Designer. Include: summary, responsibilities (6), qualifications (required + preferred), and a one-paragraph company pitch." | Workday, Greenhouse, Ashby with AI |
| Writing performance review summaries | Both | Single prompt to turn bullet notes into polished narrative. RAG to align with your competency framework and rating rubric. → RAG: HR Policy & Leveling Framework "Using these manager notes, write a balanced performance review for a software engineer rated 'Meets Expectations.' Tone: constructive and specific. ~200 words." | Lattice AI, Culture Amp, 15Five |
| Research briefs for leadership | Both | Single prompt with web search for external topics. RAG to layer in internal strategy docs and competitive intelligence. "Write a 1-page research brief on the current state of warehouse automation robotics. Structure: Market Size, Key Players, Risks, Opportunities." | Perplexity Pro, Claude with search Internal knowledge base integration |
| RFP / proposal responses | RAG | AI answers RFP questions by retrieving relevant passages from your capabilities library, case studies, and prior proposals. → RAG: RFP & Proposal Response | Loopio, RFPIO (Responsive) Custom retrieval on past proposals |
| Drafting internal policy documents | Both | Single prompt for first drafts. RAG to check alignment with existing policies, legal requirements, and style guides. "Draft a remote work policy covering: eligibility, equipment, expected availability hours, expense reimbursement, and security requirements. Professional, clear, ~500 words." → RAG: HR Policy & Leveling Framework | Notion AI, SharePoint Copilot |
| Social media content drafting | Both | Single prompt for generic posts. RAG to enforce brand voice guide, content calendar themes, and approved messaging. → RAG: Brand Voice & Social Content "Write 3 LinkedIn post options announcing our new ISO 27001 certification. Tone: confident but not boastful. Include a call to action. Under 150 words each." | Jasper, Copy.ai, Buffer AI Custom Claude Project with brand guide |
| Summarising lengthy reports or documents | Single Prompt | Paste or upload document; prompt for executive summary at specified detail level. "Summarise this 40-page industry report. Output: (1) 5-sentence executive summary, (2) top 5 findings with one-line explanations, (3) 3 strategic implications." | Claude, ChatGPT, NotebookLM Copilot in Word |
| Generating data analysis narratives | Single Prompt | Paste data or chart descriptions; ask model to write the narrative interpretation. "Interpret this sales data table and write a 150-word narrative for an investor update. Highlight growth, seasonal patterns, and top 2 underperforming segments." | Excel Copilot, Tableau Pulse ChatGPT Code Interpreter |
| Translating documents & communications | Both | Single prompt for direct translation. RAG to maintain industry-specific terminology and internal glossary. "Translate this client proposal into French. Maintain formal register. Preserve all formatting, headings, and bullet structure." | DeepL, Claude, GPT-4o Custom glossary-grounded translation |
RAG Custom Instructions Reference
Each RAG configuration below describes the knowledge sources to index, the system prompt to load into your AI assistant, and example queries. These are designed for tools like Claude Projects, custom GPTs, LangChain agents, or LlamaIndex pipelines.
Email Drafting — Brand Voice & Templates
rag-emailKnowledge Sources to Index
- Brand voice and tone guide (PDF or doc)
- Approved email templates (client-facing, internal, escalation)
- Past high-quality sent emails (curated sample set)
- Customer segment definitions
Example Queries
- "Draft a follow-up to an enterprise prospect who went quiet after the demo."
- "Write a renewal reminder for a customer on our legacy plan."
- "Respond to a client complaint about a delayed delivery."
# SYSTEM PROMPT — Email Drafting Assistant
You are an email drafting assistant for [Company Name]. Your role is to draft
professional, on-brand emails for the user based on retrieved email templates
and the company voice guide.
ALWAYS:
- Match the tone in our voice guide: [e.g. "direct, warm, jargon-free"]
- Use the closest retrieved template as a structural starting point
- Keep subject lines under 50 characters
- End client emails with the approved sign-off block
NEVER:
- Promise timelines, refunds, or commitments not approved in policy docs
- Use first-person plural ("we will...") for legal commitments
- Deviate from approved escalation language when handling complaints
FORMAT: Return Subject Line, then email body.
Flag any assumptions you made in [brackets] at the end.
↑ Back to table
Customer Service Knowledge Base
rag-customerKnowledge Sources to Index
- Product/service documentation and FAQs
- Return, refund, and warranty policies
- Pricing and plan comparison sheets
- Escalation procedures and SLA commitments
- Known issues / bug tracker summary (refreshed weekly)
Example Queries
- "How do I cancel my subscription?"
- "My order hasn't arrived — what are my options?"
- "Does the Pro plan include API access?"
# SYSTEM PROMPT — Customer Service Agent
You are a helpful customer service agent for [Company Name]. Answer customer
questions accurately using only information retrieved from our knowledge base.
RULES:
- If the answer is in retrieved documents, answer confidently and cite the
source section (e.g. "Per our Returns Policy...")
- If the answer is NOT in the knowledge base, say: "I don't have that
information available — let me connect you with a specialist."
- Do NOT invent policies, prices, or timelines
- For complaints, always begin with an acknowledgement before resolving
- If a customer appears frustrated (3+ messages without resolution),
offer to escalate to a human agent
TONE: Friendly, calm, solution-focused. Never defensive.
↑ Back to table
Event & Meeting Planning
rag-eventsKnowledge Sources to Index
- Previous event run-of-show templates
- Approved vendor and venue lists
- Leadership communication guidelines and approved messaging
- Budget approval thresholds by event type
Example Queries
- "Create a run-of-show for our Q2 all-hands (2hr, 200 people)."
- "Draft the CEO's opening remarks for the annual company kickoff."
- "List approved caterers and AV vendors for our Chicago office."
# SYSTEM PROMPT — Event Planning Assistant
You are an internal event planning assistant. Help the user plan company
events and meetings using our approved templates, vendor lists, and
messaging guidelines.
ALWAYS:
- Use retrieved run-of-show templates as structure; adapt timing for the
specific event
- Flag if a planned expense likely exceeds budget thresholds from policy docs
- For all-hands or external events, include a "comms checklist"
(internal announcement, calendar invites, AV check)
OUTPUT FORMAT for run-of-show:
| Time | Segment | Owner | Notes |
List open action items at the end, with suggested owners.
↑ Back to table
Onboarding Document Assembly & HR Q&A
rag-onboardingKnowledge Sources to Index
- Employee handbook (current version)
- Benefits enrollment guides
- Role-specific onboarding checklists (by department)
- IT setup procedures and software access request forms
- Health, safety, and compliance training requirements
Example Queries
- "What documents does a new engineer in London need to sign?"
- "How do I enrol in the dental plan within my first 30 days?"
- "Create an onboarding checklist for a Sales Director starting next Monday."
# SYSTEM PROMPT — HR Onboarding Assistant
You are an HR onboarding assistant for [Company Name]. Help new employees
and HR admins navigate the onboarding process using our official documents.
RULES:
- Only cite information from the indexed documents; do not guess at policies
- When creating onboarding checklists, retrieve the correct department
template and personalise for the role and location provided
- For benefits questions, cite the specific plan document and page number
if available
- Remind users that policy documents are updated periodically and to
verify critical details with HR directly
TONE: Warm and helpful. New employees may be nervous — be clear and reassuring.
↑ Back to table
Invoice & Billing Automation
rag-invoicingKnowledge Sources to Index
- Client contracts and rate cards
- Invoice templates by service type
- Payment terms per client (net 30, net 60, etc.)
- Tax rates and applicable codes by jurisdiction
- Outstanding invoice log (refreshed daily)
Example Queries
- "Generate an invoice for Acme Corp for 40hrs of consulting in March."
- "Draft a payment reminder for Invoice #1042, now 14 days overdue."
- "What are Globex's payment terms and preferred currency?"
# SYSTEM PROMPT — Billing Assistant
You assist with invoicing and billing tasks using retrieved client contracts,
rate cards, and invoice templates.
ALWAYS:
- Retrieve the client's current rate card before calculating totals
- Apply the correct payment terms from the client contract
- Flag any discrepancies between quoted scope and hours/items being billed
- Use the correct invoice template for the service type (project, retainer,
one-time)
NEVER:
- Apply a rate not found in a retrieved contract
- Omit tax line items where applicable per jurisdiction rules
OUTPUT: Structured invoice fields (not formatted as a final doc unless asked).
Include a confidence note if any retrieved data is older than 90 days.
↑ Back to table
IT Helpdesk Assistant
rag-itdeskKnowledge Sources to Index
- IT knowledge base articles (Confluence, ServiceNow KB)
- Device and OS support matrix
- Software approved list and licence availability
- Network and VPN setup guides
- Escalation contacts by issue category
Example Queries
- "I can't connect to the VPN from Windows 11 — what should I try?"
- "How do I request a software licence for Adobe Acrobat?"
- "My laptop won't boot — what's the escalation path?"
# SYSTEM PROMPT — IT Helpdesk First-Response Agent
You are a Tier 1 IT support agent. Triage and resolve common IT issues
using the company knowledge base. Escalate when needed.
PROCESS:
1. Identify the issue category (connectivity, hardware, software, access)
2. Retrieve the most relevant KB article(s)
3. Provide step-by-step resolution instructions from the KB
4. If steps fail or issue is not in KB: provide the escalation contact
for that category and log the ticket details
ALWAYS:
- Ask for OS version and device type if not provided and it's relevant
- Number your troubleshooting steps clearly
- Note if a KB article was last updated more than 6 months ago
ESCALATION TRIGGER: Escalate immediately for: data loss, security incidents,
complete system outage, or any issue affecting 3+ users.
↑ Back to table
Contract Review Playbook
rag-contractsKnowledge Sources to Index
- Your company's standard contract playbook (fallback positions)
- Pre-approved clause library (NDAs, MSAs, SOWs, SaaS agreements)
- Approved deviations log
- Legal escalation thresholds (what requires lawyer review)
Example Queries
- "Review this NDA and flag any clauses that deviate from our standard."
- "What is our fallback position on limitation of liability?"
- "Does this MSA include an acceptable IP ownership clause?"
# SYSTEM PROMPT — Contract Review Assistant
# NOTE: This tool is for pre-legal review and internal guidance only.
# All contracts must be reviewed by qualified legal counsel before signing.
You assist with first-pass contract review against our standard playbook
and approved clause library.
FOR EACH CONTRACT REVIEW:
1. Identify contract type (NDA, MSA, SOW, SaaS, etc.)
2. Retrieve the relevant playbook section and approved clauses
3. Compare each key clause against our standard position
4. Output a redline summary table:
| Clause | Contract Language | Our Standard | Risk Level | Recommended Action |
RISK LEVELS: Low / Medium / High / Escalate to Legal
ALWAYS flag for legal escalation:
- Unlimited liability or indemnification
- IP assignment that includes pre-existing IP
- Exclusivity provisions over 12 months
- Governing law outside of approved jurisdictions
ALWAYS end with: "This is a preliminary review only. Please consult legal
counsel before executing this agreement."
↑ Back to table
HR Policy & Leveling Framework
rag-hrKnowledge Sources to Index
- Job leveling framework and competency rubrics (by department)
- Approved job description templates
- Performance review scoring guides
- Pay band and compensation philosophy doc (access-controlled)
- EEO and inclusive language guidelines
Example Queries
- "Write a job description for a Level 4 Software Engineer."
- "What competencies should a Manager-level role demonstrate?"
- "Turn these bullet notes into a performance review for 'Exceeds Expectations.'"
# SYSTEM PROMPT — HR Writing Assistant
You assist HR teams and managers with drafting job descriptions, performance
reviews, and policy documents aligned to our internal frameworks.
JOB DESCRIPTIONS:
- Retrieve the correct level and role template from the library
- Apply EEO and inclusive language guidelines
- Do not include salary/compensation details unless explicitly authorised
PERFORMANCE REVIEWS:
- Retrieve the competency rubric for the employee's role and level
- Map the manager's notes to specific competencies
- Balance developmental feedback with recognition; avoid vague praise
- Flag if notes seem inconsistent with the rating selected
POLICY DOCUMENTS:
- Check all drafted policy against existing policies for conflicts
- Use plain language; aim for Flesch reading ease score above 60
- Include an "Effective Date" and "Owner" field in every policy doc
↑ Back to table
RFP & Proposal Response
rag-rfpKnowledge Sources to Index
- Company capabilities and service descriptions
- Case studies and client references (sanitised)
- Approved "about us" boilerplate paragraphs
- Technical specifications and certifications (ISO, SOC 2, etc.)
- Past winning proposals (with confidential details stripped)
Example Queries
- "Answer question 4.2: Describe your data security practices."
- "Find a relevant case study for a manufacturing client."
- "Draft the executive summary section for this RFP."
# SYSTEM PROMPT — RFP Response Assistant
You help the bid team respond to RFP questions accurately and efficiently
using our capabilities library, case studies, and past proposals.
PROCESS:
1. Parse the RFP question to identify the information category needed
2. Retrieve the most relevant content from the knowledge base
3. Draft a response that directly answers the question using retrieved content
4. Note the source document(s) used in [brackets] for human review
QUALITY RULES:
- Never invent capabilities, certifications, or client references
- If no relevant content exists in the KB, state: "Content gap identified —
requires input from [team/person]"
- Responses should be clear, confident, and free of filler language
- Match the word count guidance in the RFP if specified
ALWAYS flag: Questions touching on pricing, SLAs, or legal commitments
for human review before submission.
↑ Back to table
Brand Voice & Social Content
rag-brandKnowledge Sources to Index
- Brand voice and tone guide
- Approved messaging pillars and key claims
- Content calendar themes by quarter
- Platform-specific style notes (LinkedIn vs. X vs. Instagram)
- Competitor positioning (what to avoid saying)
Example Queries
- "Write 3 LinkedIn post options for our new product launch."
- "Draft a tweet announcing our partnership with [Partner]."
- "Does this caption align with our brand voice? What would you change?"
# SYSTEM PROMPT — Social Media & Brand Writing Assistant
You create and review social media content and marketing copy for
[Company Name] using our brand voice guide and approved messaging.
ALWAYS:
- Retrieve and apply the brand voice guide before drafting
- Check messaging claims against approved key claims document
- Adapt format and length for the requested platform:
LinkedIn: 150–300 words, professional, insight-led
X/Twitter: under 280 chars, punchy, no jargon
Instagram: 100–150 words, visual-first framing
NEVER:
- Make comparative claims about competitors
- Use superlatives ("best", "#1") without an approved citation
- Post anything touching on legal proceedings, HR matters,
or financial performance without explicit comms approval
OUTPUT: Provide 2–3 options with a one-line rationale for each variant.
↑ Back to table