⚙️
Setting Up AI with Context
🟢 Beginner
Like a dedicated project folder in Perplexity. You set the rules, tone, topic, and data sources once — then every question you ask inside that Space already "knows" what you're working on. Perfect for ongoing projects like IRVINS sales ops, market research, or a new product launch. Set a system prompt like "You are a senior FMCG commercial analyst for a premium snack brand in Singapore" and every answer is automatically in the right context.
Claude's persistent workspace — upload files, write a Project Instruction (its system prompt), and every conversation inside inherits that context automatically. Ideal for deep document analysis, supplier contract reviews, or any task where Claude needs to thoroughly know your material. Claude Projects vs Perplexity Spaces: Use Claude Projects for deep document reasoning, multi-step analysis, and code generation with your own uploaded files. Use Perplexity Spaces for real-time web research with citations and team-shared knowledge bases with live Connectors. Both can hold IRVINS brand docs, product catalogues, and SOPs as their knowledge base.
Let AI read your actual files — Google Drive, Notion, Slack, and more. So instead of pasting documents into every chat, AI can directly reference YOUR company data, SOPs, and reports when answering questions. For IRVINS: connect the shared commercial Drive folder and AI instantly knows your latest P&L, trade terms, and product docs.
Tell the AI who it is before the conversation starts. E.g. "You are a senior FMCG commercial analyst in Singapore. You know IRVINS is a premium salted egg snack brand manufacturing in Singapore and selling across SG, SEA, China, and the US. Always respond in concise executive format with data-backed recommendations." This shapes every response to match your industry and role — without repeating yourself every session.
Set permanent preferences that apply to ALL your conversations — not just one Space. Examples: "Always respond in bullet points. Lead with the recommendation. Flag risks clearly. Keep responses under 300 words unless I ask for detail." Also add shorthand: "When I say 'quick take' = 3-bullet summary. When I say 'full brief' = structured 1-page document." These two instructions alone cut your prompt writing time by 60%.
Premium AI tools now remember context across sessions. Perplexity and Claude remember your preferences, past research, and decisions — so new conversations build on history, not starting from zero. Think of it as the AI's long-term memory: it knows you're at IRVINS, it remembers the strategy discussion you had last week, and it doesn't ask for the same context twice.
🧠
GenAI · Agentic AI · AI Automation
🟢 Beginner
The Content Creator. You prompt it, it generates. Creates text, images, code, slides, and more from natural language. Autonomy: Low — needs you every step. IRVINS examples: Draft a TikTok livestream script for a new flavour launch. Summarise last month's sales data in plain English. Write a job description for a new retail staff role. Generate 5 ad headline variants for a Shopee campaign.
The Autonomous Worker. You set the goal — AI figures out the steps, plans the approach, and acts across multi-step tasks independently. Autonomy: High. IRVINS examples: "Research our top 3 SG competitors — find their new products, pricing, and recent campaigns, then write a 1-page competitive brief." AI does the entire research → analysis → writing chain without you clicking anything.
The Workflow Engine. Trigger → AI processes → output delivered. Connects systems to run repeatable tasks 24/7 without you. Autonomy: Full. IRVINS examples: Daily stock alerts sent to Shawn every morning. Auto POS reconciliation when an MT purchase order email arrives. Weekly sales report compiled and sent to the team before Monday standup. These run whether you're in the office or not.
Think of it as a car analogy: GenAI = you drive, AI is the engine (you steer every step). Agentic AI = GPS-guided driving (you set the destination, AI navigates and makes decisions along the way). AI Automation = self-driving (you set it up once and it runs forever without you). IRVINS' 24-week plan uses all three: GenAI first (Weeks 1–4), Agentic AI next (Weeks 5–15), then full Automation always-on (Weeks 16–24).
📅
AI in Daily Work
🔵 Intermediate
Start every day with a 60-second AI briefing instead of scrolling news. Prompt: "What happened in FMCG, snack industry, Singapore retail, and Southeast Asia consumer markets in the last 24 hours? Give me 5 bullet points with the most commercially relevant developments for a premium snack brand." Takes 30 seconds to read; saves 20 minutes of scattered reading.
30 minutes before any meeting, ask: "I have a JBP review with NTUC FairPrice tomorrow. Give me: (1) 3 things to know about their current retail strategy, (2) 3 negotiation anchors for shelf space and promo support, (3) likely objections and how to handle them, (4) one question to ask that puts me in a stronger position." Walk in prepared every time.
The blank page is the productivity killer. Always generate a first draft with AI, then edit. This is faster for 95% of work. Email to a retailer, business proposal, budget justification, performance review, SOP update — give AI the context and ask for a draft. You refine in 5 minutes vs. building from zero in 45 minutes. Key rule: Tell AI the audience, purpose, and tone. It handles structure and language; you add judgment and specifics.
Before any important call, decision, or strategy session, run a quick background research session. Examples: "What is Lotus's current FMCG snack category strategy in Thailand?", "Summarise the salted egg food trend in Singapore and SEA — peak, current status, and what's coming next", "What are the most effective corporate gifting strategies for FMCG brands in Singapore?" Perplexity gives you cited, real-time answers in 60 seconds.
Before making any important call, lay out the decision to AI and ask for structured analysis. "I need to decide whether to expand IRVINS into direct D2C e-commerce in Malaysia or focus on Shopee/Lazada marketplace first. Give me: key decision criteria, pros/cons of each, what we'd need to be true for each to work, and your recommendation with reasoning." AI won't replace your judgment but will sharpen it dramatically.
The #1 thing that separates AI power users from casual users: habit. Before starting any task that takes more than 15 minutes, ask "Can AI do the first 80% of this?" If yes, start with AI. The goal is AI-first thinking: your default action is to reach for AI before reaching for a blank doc, Google, or a colleague. Build the reflex in weeks, and your output doubles.
🟣
Advanced AI Use Cases
🟣 Advanced
Dedicated AI email tools go far beyond basic drafting — they learn your writing style, auto-triage your inbox, and draft replies in your voice. Fyxer: Connects to Gmail/Outlook; learns how you write from past emails; drafts replies; auto-sorts inbox into "reply", "FYI", and "marketing noise"; takes meeting notes and writes follow-ups. Users save 1+ hr/day. Google-verified and Microsoft-verified. Superhuman AI: Premium email client; Split Inbox auto-categorises by importance; AI drafts replies matching your personal style; users process inbox 2x faster and save ~4 hrs/week; write 59% more emails per hour. SaneBox: AI filters low-priority mail into separate folders automatically — inbox shows only what truly needs your attention. No rule-setting required. Claude/Perplexity in Gmail: Use browser extensions to run AI directly in your inbox — paste a thread, ask Claude to draft a contextual reply in your tone, copy and send. Choose by profile: Fyxer for auto-drafted first-draft replies; Superhuman for elite inbox speed and team collaboration; SaneBox for inbox zero discipline with minimal effort.
AI transforms sales ops from a reporting function to a decision engine. At IRVINS (Zack & Kami): Weekly sales reports auto-compiled before Monday sync (saving 4–6 hrs/week); AI-generated personalised B2B outreach emails for new corporate leads; promo ROI analysis that used to take a day now takes 10 minutes; competitor battlecards updated weekly from Perplexity research. Replace the manual Excel grind with AI-powered commercial intelligence.
AI generates campaign briefs, ad copy variants, social media calendars, and performance summaries in minutes. At IRVINS (Natasha, Calista, Wan Xing): 30 pieces of social content generated from one brief; TikTok hooks and Douyin variants created together; Shopee listing copy A/B tested with AI; weekly competitive intel brief delivered automatically. Feed IRVINS brand guidelines and tone of voice once — AI stays on-brand across every output.
AI processes unstructured data — meeting notes, customer feedback, survey responses, review comments — and turns them into structured insights and dashboards. For IRVINS leadership: weekly business performance digests compiled automatically; customer review themes surfaced weekly; meeting action items extracted and assigned without manual note-taking. AI transforms the messy signals in your business into clear decisions.
AI eliminates the manual grind of month-end. At IRVINS (Agnes): P&L sheets auto-generated from the central data hub; AI flags spend anomalies before they're posted to SAP; cash flow commentary drafted from raw numbers in 2 minutes; POS reconciliation reduced from 3 hours to 10 minutes. SAP journal workflows semi-automated — AI pre-checks data consistency before posting and highlights variances. Finance team shifts from data-wrangling to actual analysis.
AI multiplies design output without adding headcount. At IRVINS (Gordon Hoon & Calista): One campaign brief → 10 copy variants and all platform resizes in under 3 hours; China Douyin assets localised from SG originals without a separate briefing cycle; AI generates concept mood boards for R&D flavour launches. Tools: Canva AI (bulk resize, AI image generate, Magic Write), Adobe Firefly (on-brand image generation from text prompts), Claude (creative briefs and direction), Midjourney (concept ideation). One skilled designer with AI = output of three.
AI handles repetitive data entry and flags issues before they become problems. At IRVINS (Shawn & Assh): RPA scripts written by Claude via Vibe Coding auto-populate SAP/WMS entries from Google Sheets — cutting 5–8 hrs/week of manual keying; daily AI ops summaries flag stock discrepancies, SLA breaches, and pick errors each morning; low-stock alerts sent automatically so reorders happen before stockouts hit. Team focuses on fulfilment quality, not data entry.
AI removes admin load so HR can focus on people, not paperwork. At IRVINS (Filia): Employment contracts and offer letters auto-generated from a Google Form submission; CV screening summaries for frontline and tech-adjacent roles compiled instantly; pulse surveys deployed and AI-analysed weekly for department sentiment; 30-day personalised onboarding plans generated for every new hire. For 300 staff, saves 10+ hrs/month of document work. Tools: Claude (document drafting), Google Apps Script (automated doc generation from Sheets), AI chatbots for common policy questions.
AI optimises the factory floor without complex ERP upgrades. At IRVINS (Ken): Daily production priority lists generated from orders and current raw material stock; batch capacity calculated automatically — how many runs before ingredients run out; daily production reports auto-populated from shift logs with downtime events flagged; quality deviation tracker identifies recurring defect patterns before they hit a full batch. Vibe Coding prompt to start: "Build a daily production scheduler — I input today's orders and current raw material inventory, and it outputs a ranked batch schedule with urgent flags for ingredients running low."
AI brings real-time market intelligence into the R&D kitchen. At IRVINS (Assh, with Gordon as commercial interface): Weekly AI Snack Trend Radar (Perplexity Deep Research) scans ecom reviews, TikTok food content, Reddit, and food media for emerging flavours, textures, and formats across SG, China, and US — top 5 trends every Thursday morning. AI concept scoring ranks new SKU ideas by demand signals, manufacturing feasibility, and IRVINS brand fit before a single trial batch is made. Market-driven R&D pipeline instead of gut-feel innovation.
AI transforms reactive supply chain management into proactive planning. At IRVINS (Jolene): AI demand forecasting for hero SKUs — blending 12 months of sales history, upcoming campaign calendars, and seasonality to generate weekly reorder recommendations; planning risk alerts flag demand-supply mismatches and shipment delays before they cause stockouts; purchase order drafts auto-generated from forecast data for Jolene's review and approval. Planning cycle time cut by 30%. No more emergency reorders.
Kicks off a 20–30 minute autonomous research sprint — reads dozens of sources, synthesises findings, and produces a fully structured report with citations. Like hiring a research analyst for a single task, instantly. IRVINS use cases: Full competitor landscape for a new market entry; consumer sentiment analysis for a new flavour concept; supplier background check before onboarding; regulatory summary for an export market. Available in Perplexity Spaces with cited sources you can verify.
🎸
Vibe Coding & AI Agents
🔵 Intermediate
You describe what you want to build in plain English. AI writes all the code. You run it, test it, and paste errors back to AI to fix. No coding experience needed — just the ability to clearly describe what a tool should do. This is the new skill that makes every non-technical IRVINS team member a builder. If you can describe a problem, you can build a solution.
Step 1 — Define: State the problem in one sentence. Who uses it, what data goes in, what comes out. Step 2 — Prompt: Give Claude a detailed brief including tech stack (e.g. "Google Apps Script", "HTML dashboard", "Python script"), data source, and specific output format. Step 3 — Test: Run the code. If errors appear, paste the exact error text back to Claude: "I got this error: [paste]. Please fix." Iterate until it works. Step 4 — Deploy: Share with your team. Most IRVINS tools can be hosted on Google Sheets, Replit, or as a standalone HTML file. Zero-to-deployed in under 1 hour for most tools.
An "agent" is just a script that does multiple steps automatically. Vibe Coding lets you build agents without writing a single line yourself. IRVINS Vibe Coding prompts to try now: "Build a script that reads our Google Sheet of B2B leads and sends me an email listing anyone not contacted in over 5 days." / "Create a low-stock alert dashboard that reads our inventory file and flags any SKU under 200 units." / "Build a script that takes a customer name and product list and auto-generates a formatted quote PDF using our price list."
For more complex multi-step automations, frameworks like n8n (visual workflow builder, self-hostable, hundreds of integrations) and CrewAI (multiple AI agents collaborating on a task — one researches, one analyses, one writes) let you orchestrate entire workflows visually. n8n is ideal for connecting IRVINS systems: new Shopee order → AI summarises → Slack alert → CRM log → reorder check, all without code.
1. Define the job in one sentence: "This agent reads our daily sales Sheet and sends a Slack alert if any SKU's sales drop more than 30% vs the prior week." 2. Define inputs and outputs: Input = Google Sheet with [Date | SKU | Sales]. Output = Slack message with SKU name, this week's sales, prior week's sales, and % drop. 3. Prompt Claude: "Write a Google Apps Script that does exactly this. Explain each section." Paste into Apps Script, run, set a trigger. Done.
📊
Claude Artifacts & Perplexity Files
🔵 Intermediate
Claude creates fully interactive web tools, calculators, dashboards, and apps directly in the chat window. No deployment or coding needed — Claude builds it, you use it instantly. IRVINS examples: "Build me a promo ROI calculator — I input trade spend, baseline sales, and uplift sales, and it shows me ROI, incremental revenue, and break-even point." / "Create an interactive channel mix dashboard with a pie chart and trend table." Copy the HTML and share with your team. Refreshes in seconds when you ask for changes.
Perplexity can generate shareable interactive HTML dashboards from your data. Paste in a CSV or describe your data structure and ask: "Create an interactive dashboard with a revenue trend chart (last 8 weeks), a channel mix pie chart, and a SKU performance table with RAG status (red/amber/green)." The output is a shareable link — no login needed for viewers. Ready in under 5 minutes, shareable immediately with leadership.
Perplexity generates fully cited, structured research reports that you can export and share. Use for: competitor landscape analysis before a JBP, new market entry feasibility, ingredient or packaging supplier background checks, and industry trend reports. Every claim is cited with a source link — you can verify any point instantly. Unlike ChatGPT reports, Perplexity's are grounded in real-time web data, not training data that may be outdated.
Ask Perplexity to generate structured data directly as a downloadable Excel file. "Create a 30-day social media content calendar for IRVINS in Excel format: Date | Platform | Post Type | Copy | Hashtags | Assignee | Status." Or: "Build a promo tracking template with columns for: SKU, channel, promo mechanic, start date, end date, target uplift, actual uplift, ROI." Download and use immediately — no manual formatting.
Generate full presentation decks from a single prompt. "Create a 6-slide investor update deck for IRVINS covering: company overview, 2025 performance highlights, 2026 growth strategy, new market opportunities, team structure, and funding use. Boardroom tone, data-driven." The old way: 2–3 hours of blank-slide building. The AI way: 5 minutes to generate, 20 minutes to refine. Iterate with plain English: "Make slide 3 more visual. Add a chart placeholder. Change the tone to more optimistic."
🔴
Expert Automation & Workflows
🔴 Expert
Perplexity takes control of a browser or desktop. You describe a task in plain English ("Go to this site, extract the pricing table, save it as a CSV") and it executes step by step — clicking, reading, copying — just like a human would. This is RPA without writing RPA code. IRVINS use: "Go to Shopee Seller Centre, pull this week's sales data for all SKUs, and save it to a Google Sheet." Run on schedule for daily automated data pulls.
Chain multiple AI tasks end-to-end: Research → Summarise → Draft email → Schedule send → Log to CRM. Each step passes its output to the next. You can automate entire business processes from trigger to final action with zero code. IRVINS example chain: New B2B lead added → Perplexity researches the company → Claude writes personalised email → email saved as draft in Gmail → lead record in Sheets updated with research summary. Entire chain runs in under 2 minutes, automatically.
Google Apps Script is a free, built-in automation engine inside Google Workspace (Sheets, Docs, Gmail, Drive, Calendar). You describe what you want to Claude, Claude writes the JavaScript, you paste it into the Apps Script editor, and it runs on a schedule automatically. How to use it: (1) Describe your task to Claude: "Write a Google Apps Script that reads column D of my Sheets, sends a Gmail alert if any value drops below 100." (2) Claude writes the complete script. (3) Open Google Sheet → Extensions → Apps Script → paste code → Save → set a Time Trigger (e.g. daily 8AM). Done — zero coding. IRVINS examples already running: Modern Trade PO email → extract line items → log to Sheets → send DC email (3 hrs/week → 10 mins). Weekly Google Reviews across 10 stores auto-extracted to Sheets (1 hr → 5 mins). Advantage over n8n: Free, runs inside Google Workspace your team already uses, non-technical staff can maintain it with Claude's help.
Connect AI to live data sources via API — Shopee, Lazada, TikTok Shop, Google Sheets, SAP, your own database. AI can then read and write real data in real-time. IRVINS Phase 4 targets: Shopee/Lazada API → live GMV and inventory data pulled automatically; SAP API → AI-generated journal entries pushed directly; TikTok Shop API → daily sales and content performance pulled into the commercial cockpit. Claude writes the API connection code via Vibe Coding — you just provide the API documentation.
JSON is the universal language of data exchange between systems. When AI extracts data (e.g. from a PO email or Shopee API), it structures it as JSON — a clean, machine-readable format that any other system can read. Example: AI reads a PO email → extracts {SKU, quantity, price, retailer} as JSON → Apps Script writes each item as a new row in the Orders Sheet → triggers a Slack notification → auto-drafts the DC email. Each step passes the JSON forward. No manual touching of data anywhere in the chain.
Visual workflow builders where you connect AI nodes with business apps — no code needed. Trigger: "New order in Shopee" → AI summarises → Slack alert → CRM log → Reorder check. Hundreds of integrations. n8n is self-hostable (better for data privacy); Make.com is cloud-first (easier to start). When to use n8n vs Apps Script: Use Apps Script for workflows that live entirely within Google Workspace. Use n8n when you need to connect non-Google systems (Shopee, Slack, Notion, external APIs) in complex chains.
Write structured prompts as reusable templates with variables — {{customer_name}}, {{product_sku}}, {{channel}}, {{market_region}}. One master prompt handles thousands of personalised outputs automatically. IRVINS example: A single prompt template for B2B outreach emails takes customer name, company, industry, and recommended SKUs as variables — runs for 50 leads at once, each email personalised. This is the foundation of scalable AI-powered content, outreach, and analysis.