Wilmar AI Cafe
← Back to menu

Week 4 · Create Your AI Assistant

We open with a quick Workflow Copilot warm-up — a daily agenda summary at the start of your day — then build a chatbot connected to company knowledge and adapt it into a useful business assistant for a real team scenario.

📅 Fri 7 Aug · 4–5 PM
📍 Wilmar office
🧭 Warm-up + 3 guided builds
⏱️ 60 min
Host · Xinran Liu

📦 Class materials

Download these before we start. The two Sugar SOPs are the knowledge source for the agent in section 2 — drop them into your own SharePoint or OneDrive folder and point your agent at that folder. The practice file is the upload for section 3.

📄Sugar Trade Process SOPpdf 📄Sugar Limits Setting SOPpdf 📊Sugar risk practice filexlsx

Menu

Follow along, click to expand, and build as you go.

1
10 min

Warm-up · Today's Agenda Summary with Workflow Copilot

Start the day the automated way: a Workflow Copilot flow that runs each weekday morning, gathers today's remaining calendar events, and sends you a plain-text agenda summary in Teams.

What is Workflow Copilot?
  • Instead of manually picking triggers and actions, you describe the workflow in plain language and Copilot drafts the steps.
  • Your role is to review the drafted steps and adjust them.
  • Every workflow has a trigger (when it starts), an action (what happens), and an output (the useful result) — this prompt covers all three at once.
The prompt · copy into Workflow Copilot
Create a workflow that runs every weekday at 8:30 AM. Find my calendar events for today that have not yet ended, and take only the title, start time, and end time — no meeting links, join URLs, or online meeting data. Send me a Teams message in Chat with Flow bot listing each event on its own line in plain text. If there are none, send: You have no upcoming events today. My timezone is Asia/Singapore.
Test on your own account and check the run history before trusting it — then reuse the same pattern for reminders, follow-ups, or an end-of-day wrap-up.
2
15 min

Build a Copilot Agent Connected to Company Knowledge

Watch a live demo of building a Copilot agent that answers questions from the two Sugar SOPs above — the kind of question you would otherwise go hunting through a PDF for.

What you will see
  • How to put the two Sugar SOPs in a SharePoint or OneDrive folder and connect Copilot to that folder.
  • How the agent searches those files and answers based only on what the SOPs actually say.
  • How to test it with the real questions below and check the answer against the source.
Why Copilot for company knowledge
  • It searches only your connected documents, not the whole internet.
  • Answers are grounded in your actual SOPs, so they stay relevant and controlled.
  • Pointing it at one specific folder narrows the search space and improves accuracy — the agent focuses on the SOPs instead of everything you have access to.
Practice questions

Try these against the two SOPs, then check each answer back in the PDF.

What is the approval process for sugar trade limits?
How do we handle washout transactions?
What are the payment terms for each customer class category?
How is the country risk rating decided, and how would I work out the rating for a client?
Always check the answer against the SOP before acting on it — the agent points you to the right clause quickly, but you stay responsible for the decision.
3
15 min

Build a Copilot Agent for Document Generation

Upload the sugar risk practice file and have the agent turn a month-end risk pack into a written review memo, in exactly the format you ask for.

The practice file

Download Sugar risk practice file from Class materials above. It is a fictional month-end risk pack with four sheets:

  • Counterparty_Exposure — credit limits, outstanding exposure, utilisation, payment terms, customer class and country risk rating.
  • Open_Positions — physical and futures positions by delivery month, with mark-to-market.
  • Market_Prices — month-on-month moves in the main sugar benchmarks and freight.
  • Incidents_Log — washouts, late payments and limit breaches logged during the month.
Sample instruction
Using the uploaded sugar risk workbook, write a one-page Monthly Sugar Risk Review in this format: 1. Headline — three sentences on the overall risk position this month. 2. Credit risk — counterparties above 90% limit utilisation or with overdue days, in a table: counterparty, country, exposure, utilisation %, overdue days, action needed. 3. Market risk — net long/short by product, total mark-to-market, and what the price moves did to it. 4. Issues this month — each incident in one line, with the amount and current status. 5. Recommended actions — maximum five bullets, each naming an owner and a deadline. Use only figures from the workbook and state the sheet each number came from. Flag anything that looks like a limit breach in bold. Keep the tone professional and concise.
Tips
  • Be specific about the exact structure you want — numbered sections beat "write a summary".
  • Say what to include and what to avoid, and ask it to cite the sheet so you can check the numbers.
  • Re-run with a small change to one instruction and compare — that is the fastest way to see what wording actually does.
The data is invented for this session — no real counterparties or positions. Check any generated figure against the sheet before you trust the pattern on your own data.
4
20 min

Build Your Own Agent

Use the second half of the session to build something yourself on the spot, test it, and ask questions as you go.

What to build

Choose a simple use case that is relevant to your team and try building it during the session.

Ideas by department
  • HR: onboarding assistant, policy Q&A bot, training recommender
  • Sales: sales FAQ bot, proposal drafting assistant, customer meeting prep
  • Finance: expense policy assistant, invoice or PO query bot, budget explainer
  • IT: helpdesk assistant, troubleshooting guide, system access helper
  • Analytics: dashboard explainer, SQL helper, data dictionary bot
Good starting point
Pick one repeated question your team gets often, gather the documents that answer it, and build a simple agent around that use case.
Tips
  • Start narrow so your agent feels focused and useful.
  • Bring one or two good documents instead of too many files.
  • Test with realistic questions and refine the instructions based on the answers.
  • Use the live Q&A time to troubleshoot and improve your build.