Everyday business. New possibilities.
A practical, three-session series for SME owners and their teams.
Discover how to use AI for everyday business tasks—from understanding your data to creating a campaign and supporting your customers.
Save your work after each session so you can build on it throughout the sprint.
Before we begin
A little preparation will help you make the most of the hands-on time.
Try one part of the workflow on your own business data between sessions, using only data you’re permitted to share. Keep your output and note what helped, what you’d change and any questions that come up.
Each workshop includes a dedicated 30-minute Q&A with the OpenAI team to answer your questions.
11am–noon · WorkshopNoon–12:30pm · Q&A with OpenAISingapore time (SGT)
Explore business data with AI, discover useful questions to ask, and identify patterns in sales, customers and performance. Practise checking the findings and turning them into a clear next step.
A business analysis, a clear visual and a repeatable approach to turning data into decisions.
If Zoom asks for a passcode, use the one in your session follow-up email.
Meet Little Lantern Gifts, a fictional Singapore online gifting business. We’ll use its synthetic data to explore sales and customers, then carry a checked insight into a campaign and customer-support exercise.
The participant starter pack contains the 132-transaction CSV, checked fallback table and matching exercise guide.
132 transactions from March–August 2026, with sales in Singapore dollars. This CSV is the dataset used in Session 1.
Download transaction register (CSV)Verified customer-segment totals and definitions, matching the CSV. Use this to check your results or follow along if uploads or analysis are unavailable.
Download checked fallback table (TXT)Business context, prompts and practice steps for the transaction register. Ask for ideas, choose an analysis, check the findings and save your work.
Read the guide and copy the promptsAll business details and records in these files are fictional and created for learning; they do not represent any real business, customer or transaction. You do not need to bring or upload real customer information.
Work through the steps below. Copy a prompt or dictate it into ChatGPT, then check the results before moving on.
Use Little_Lantern_Gifts_Transaction_Register.csv. Upload the CSV, not the ZIP.
All 132 records are SYNTHETIC and cover 1 March–31 August 2026.
Copy or dictate each prompt. If dictating, check the transcription before sending.
Keep the same conversation for steps 1–7 so ChatGPT has the file and earlier results.
Imagine you run a small gifting business. Each row is one completed transaction for one product type; quantity counts gift sets. Sales are in Singapore dollars and equal quantity times unit price. Tax and delivery are excluded; there are no discounts or refunds. Costs, live stock, marketing consent and reasons for buying are not supplied. Sales do not establish profitability or future demand.
Upload the transaction register, then use:
I run Little Lantern Gifts, a fictional gifting business. I've uploaded our synthetic sales register. I want to know where to focus next month. What does this file contain, what dates does it cover, and what needs clarifying?
Check: 132 transactions, March–August 2026, three customer segments and three products. If asked which month you mean, specify September 2026 for this fictional exercise: it is the month immediately after the data. Do not treat the register as current business performance.
What three useful analyses could we do with this transaction register? For each, explain what business decision it could help me make and what the data cannot tell us. Wait for me to choose.
Read the suggestions before choosing. You do not need to arrive with the analysis already planned. If the first response has already identified corporate gifting as the largest segment, the next step checks and quantifies that finding.
Using this file, compare total sales by customer segment in a table. Use Singapore dollars. What are two observations and one question worth investigating?
Check against Checked_Fallback_Table.txt:
Corporate gifting: S$36,276
Event & community: S$12,024
Personal gifting: S$6,120
Total: S$54,420
Read the table before moving on. The chart helps communicate the size of the differences; it does not prove why those differences exist.
Now turn those totals into a simple bar chart. Put customer groups along the bottom and sales in Singapore dollars up the side. Show the value on each bar.
Check all three labels and values against the table. If chart generation is unavailable, keep the table or use Excel or Google Sheets to chart the checked totals.
Show one sales calculation I can check. Do the segment totals add up to total sales? Suggest one action to test and what we'd need to confirm first.
Open the original CSV and check a row: quantity multiplied by unit price should equal sales amount. Check that the three segment totals sum to S$54,420. Asking ChatGPT to check itself is useful, but also compare with the source and the prepared fallback table.
This follow-up turns the question raised in the session into a reusable prompt:
Is corporate gifting the largest customer segment in every month? Compare total sales by customer segment for each month. Show a table in Singapore dollars and corporate gifting's share of each month's sales. Tell me whether the pattern is consistent.
Check: corporate gifting is the largest segment in all six months. Its monthly share ranges from about 57.9% in April to 70.5% in August. Being largest every month does not rule out individual large transactions driving results.
Optional follow-up visualization:
Turn the monthly sales table into a line chart, with months along the bottom and sales in Singapore dollars up the side. Use one line for each customer segment.
Use this follow-up prompt to create the summary card discussed in the session:
Create a short business insight card with four parts: the finding, the numbers supporting it, one action worth testing, and what we need to check before acting. Use only what this register supports. Keep it clear and concise.
Save the card with your table or chart. Treat the action as a proposal, not a proven business recommendation. Check costs, margins, capacity and demand before making a commitment.
Choose a plugin available in your account and review its setup requirements. If it includes a data-source app, connect only an approved account and review its permissions. A plugin supplies a workflow and/or tools; a connected app supplies access to a service. Installing a plugin does not automatically authorize file access. You can also analyze the uploaded CSV in ordinary ChatGPT without installing a plugin.
If you have connected Google Drive and stored the practice CSV there, try:
Find Little_Lantern_Gifts_Transaction_Register.csv in my connected Google Drive. Confirm the file name and date range before we analyze it. If there is more than one match, ask me which to use.
Confirm the correct file and that its contents are accessible before continuing. Connecting Gmail does not by itself connect Google Drive.
Then use this specific practice task:
Look only at corporate gifting. Which product has the highest total sales? Show sales by product in a table, highest first, in Singapore dollars. Check that the total matches our earlier corporate sales figure. Suggest one product to explore for a campaign and what we should confirm first.
Check: Drinkware S$18,972; Desk essentials S$10,752; Notebook gift sets S$6,552. Total S$36,276.
Optional visualization:
Turn those product totals into a simple bar chart. Show sales in Singapore dollars and label each bar with its value.
Free accounts support file uploads and data analysis, with separate usage limits. Plugin and app availability also varies by account and workspace.
If upload or analysis is unavailable, open Checked_Fallback_Table.txt, paste its table and definitions into chat, then use:
Using this checked summary of fictional sales, tell me two things you notice, one question it can't answer, and one action worth testing. Keep it brief.
If only the chart fails, use the table. If chat is unavailable, review the checked table on your own or with a partner.
Set up a plugin available in your account and connect an approved data source if needed. Choose a business question and a small file you are authorized to use. Remove unnecessary personal or confidential information. If approval is unclear, keep practising with the synthetic register.
Help me analyze my business data. First, ask about my business, my decision, and the file I can use. Then suggest three analyses and wait for my choice. After that, show a chart or table and help me check one result.
Bring one chart, one checked insight and one possible next step. Note any access or setup issues so we can address them.
Practice files: https://sme-ai-sprint-preread.vercel.app/#practice-files
Plugin setup: https://help.openai.com/en/articles/20001256-plugins-in-chatgpt
Free account guidance: https://help.openai.com/en/articles/9275245-chatgpt-free-tier-faq
Data controls: https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt
Business privacy: https://openai.com/enterprise-privacy/
SESSION 02 · RECORDING · 5 NOVEMBER 2026
If Zoom asks for a passcode, use the one in your session follow-up email.
Watch Session 2 recordingChoose where to focus, create an Instagram post and campaign site, then plan a performance test.
Fictional workshop materials. Use this Session 2 snapshot separately from Session 1’s data.
Download overview (PDF) · Download sales data (CSV)
Fictional workshop material | Singapore | All amounts in Singapore dollars
All business details, figures and customer comments in this overview are invented for the workshop. They are not real customer research or business results.
Little Lanterns is a small Singapore business selling thoughtful gifts and everyday home accessories through an online shop and occasional pop-up markets.
Customers buy presents for friends and colleagues, as well as items for themselves. The brand’s voice is warm, thoughtful and straightforward.
Sales have fallen over the past three months. The owner wants help understanding what is happening and deciding where to focus their marketing effort.
| Product | Price (SGD) |
|---|---|
| Notebook | S$18 |
| Mug | S$24 |
| Candle | S$32 |
| Tote bag | S$22 |
| Zip pouch | S$16 |
| Gift set: notebook, mug and gift box | S$48 |
Three hours a week and up to S$300 a month for marketing. The owner does not have to spend the full amount.
An Instagram account with 1,200 followers. Posting has been irregular.
A list of 300 previous customers who have opted into marketing emails.
An online shop linked from the Instagram profile. No real shop URL is supplied for this fictional business.
No previous paid advertising experience.
No approved discount, special promotion or limited-stock offer.
Product margins and sales attribution by marketing channel are unavailable.
These are three consecutive fictional months. Orders include online purchases and pop-up sales.
| Measure | Month 1 | Month 2 | Month 3 |
|---|---|---|---|
| Total orders | 240 | 210 | 180 |
| First-time customer orders | 180 | 150 | 120 |
| Returning-customer orders | 60 | 60 | 60 |
| Average order value (SGD) | S$40 | S$40 | S$40 |
| Revenue (SGD) | S$9,600 | S$8,400 | S$7,200 |
Definitions: first-time customer orders are orders placed by someone who had not purchased before. Returning-customer orders are orders placed by someone who had purchased before. These are order counts, not a customer retention rate. Average order value is revenue divided by total orders.
The separate sales CSV contains exactly these figures. Do not add the CSV totals to the table totals.
Three fictional comments from recent customer conversations:
“I found you at a pop-up. I didn’t realise I could order online afterwards.”
“I liked the products, but choosing something for a colleague took me a while.”
“I noticed a mug in a friend’s Instagram story and wanted to know where it came from.”
These comments suggest possibilities, but we do not know how representative they are. They are research-style inputs for the exercise, not testimonials to put in customer-facing copy.
No detailed customer purchase history, product-level sales, margins, website traffic, purchase-conversion data, stock history or reliable marketing attribution is supplied. We do not know whether seasonality affected sales.
Ask your marketing adviser what the information suggests, where the owner should focus and what else would help them decide. Assess the recommendation before choosing a campaign.
No product reference photographs are included. Generated products should be treated as illustrative concepts. For a real business, supply approved product photographs and check any generated representation against the actual products.
Start a new Work chat. Upload the overview and sales CSV, or paste the full overview above. The CSV repeats the overview’s sales table. Stay in the same chat, send one prompt at a time and replace anything in square brackets with your choice.
Finish with a recommended business priority and a chosen campaign concept.
Act as Little Lanterns’ marketing adviser. Review the business overview and context and fictional sales and customer snapshot.
What does the information suggest is happening in the business? What should the owner focus on to improve sales, and what evidence supports your recommendation?
Distinguish supplied facts from hypotheses, and flag missing information that could change your recommendation.
Pause and review
Which figures support the recommendation? What is still unknown? Decide whether to accept it before continuing.
This follows the new-customer priority discussed in the workshop. If your chat recommended something else, first explain which priority you want to explore.
Based on your recommended priority, suggest three distinct campaign concepts for Little Lanterns to attract new customers.
For each, describe the intended customer, buying need, main message and products to feature. Recommend how to reach them, explain why the concept might fit and flag its main limitation.
Use supplied facts and label assumptions. Don’t invent offers, testimonials, urgency or product claims.
Recommend one concept within three hours a week and S$300 a month, then wait for our choice before creating content.
Make a choice
Compare the proposed concepts, their customer needs, suggested channels and limitations. Choose one concept. For the shared exercise, tell ChatGPT you want to develop it as an organic Instagram campaign before sending Prompt 3.
Agree a short campaign plan, create its image and caption, and build the working campaign site.
Replace [chosen concept] with the concept you selected.
Develop [chosen concept] into a practical campaign for Little Lanterns.
Explain who we’re trying to reach, their buying needs, our main message and the action we want them to take.
Recommend how to use Instagram to introduce the campaign and follow it up, within three hours a week and S$300 a month. Explain what the first image and caption should communicate.
Use the supplied facts, label assumptions and don’t invent offers or product claims. Keep the plan concise and wait for my feedback before creating the image.
Review and agree
Check the audience, message, customer action and follow-up. Give any feedback and confirm the plan before moving on. For example: “Keep the tote and pouch as the focus, and make the follow-up one simple product post. Use this plan for the first image.”
Create the first Instagram image and caption from our agreed campaign plan.
Follow the audience, message, products and visual direction we chose. Keep on-image text short and readable.
Use supplied product references if available; otherwise label the product designs as illustrative outside the artwork.
Write the caption separately, using confirmed facts and prices and a clear customer action.
Inspect the result
Check text, product appearance, prices and whether the image and caption fit your chosen idea. Without supplied product references, these are illustrations of fictional products.
Create a working campaign site with the agreed content and image. Use Sites when available; otherwise download the self-contained HTML file and open it in a browser. The workshop site is separate from Little Lanterns’ existing shop.
Keep using the same chat. If the site build cannot access your generated image, attach the downloaded image and ask it to use that file.
Create a working campaign site for Little Lanterns using Sites, our agreed plan and Instagram image.
Make it mobile-friendly. Include the campaign headline, a short introduction and featured products with confirmed prices. Add “Explore the products,” clickable product details and “Buy — demo only,” which confirms no order was placed.
Use supplied facts only. Label it “Fictional workshop prototype” and identify illustrative product images. Don’t collect customer information or take payments.
Deploy the site using Sites. Give me the live URL and check its buttons. If Sites is unavailable, provide a self-contained HTML file I can download and open in a browser, with styles, interactions and the image embedded.
Try the customer journey
Read the Instagram caption, then open the site preview or downloaded HTML file in a browser. Click “Explore the products,” inspect the product details and try the demonstration purchase action. Check that the message, image and prices match, and that the page is easy to use at a narrow width.
Improve the post and page, then plan a focused performance test.
My feedback is: [describe one specific change to the Instagram image, caption or campaign page, and why it matters].
Revise the affected materials. Keep the message, product facts and prices consistent across the post and page, preserve what already works, and briefly explain what changed. If you change the image, use the revised image on the page too.
Keep the site as a fictional workshop prototype with demo-only purchase actions. Deploy the revised site and give me the updated live URL, or provide an updated HTML download if we used that fallback.
Example feedback: “Move the products and prices directly below the campaign image, and make the caption’s invitation point to those products. Keep the headline, image and prices.”
Recheck the affected content and buttons after the revision.
Plan a small performance test for our agreed campaign, within three hours a week and S$300 a month.
What should we test first to learn whether it helps attract first-time customers? Recommend one variation, what to keep the same, and one main measure of success.
Explain how to collect the results using the information and tools we actually have, what would support keeping or changing the campaign, and when the evidence would be too limited to decide.
Distinguish first-time purchases from attention or clicks. Flag any setup needed; don’t invent benchmarks, results or tracking capabilities. Treat the website as a prototype until a real purchase destination is ready. Don’t launch ads, post to Instagram or spend money.
Review the test
Is there one clear variation and a business outcome to measure? Can the owner collect that information? What needs to be ready before testing, and what would make the result inconclusive?
Save your final image, copy the caption and test plan, and keep the live site URL or downloaded HTML file for review. The site uses fictional workshop content and demo-only purchase actions. Deploy the fictional site using Sites. Do not post to Instagram or spend money.
Use appropriate information about your own business to work through all three stages: understand and choose, plan and create, then improve and test.
Share your business context, relevant source data and realistic time or budget. Ask ChatGPT where to focus, check the evidence and choose one marketing goal.
Ask for three ideas and a recommendation. Use your customer knowledge to challenge the advice, then choose one approach and agree the message and customer action.
Create one piece of content, such as a social post with an image and caption or a customer email. Check the facts, product details and prices, then give specific feedback and improve it.
Plan one small test: choose one variation, what to keep the same, one business outcome to measure, how to collect the results and when to review them. Flag missing information or setup; do not invent tracking capabilities or results.
Bring back your content, why you chose the approach, what changed after your feedback, and the test plan. If you ran the test, share what you learned; otherwise, explain what needs resolving before you run it. Creating content and planning the test complete the homework. Publish or spend only when you are ready and authorised.
SESSION 03 · 12 NOVEMBER 2026
11am–noon workshop · Noon–12:30pm Q&A · Singapore time
Join Session 3 on Zoom ↗Objective: organise your knowledge, answer customers and save a reusable Skill.
Before the workshop
Before the workshop, download and unzip the practice files from this pre-read, then upload all five PDFs to a folder in your own Google Drive so you can follow along. Connect Google Drive in ChatGPT using the same account.
Fictional workshop materials. If Google Drive is unavailable, upload the five PDFs directly to ChatGPT and follow the same file review.
01 · Product Specifications (PDF)
03 · Order and Service Policies (PDF)
04 · September Delivery Policy (PDF)
05 · Promotional Offer Draft (PDF)
Download customer enquiries (ZIP)
Q1 is Mei’s gift enquiry. Open Q1_Mei_Gift_Enquiry.txt when Tara retrieves the same question in Gmail. Q2 is the shop-assistant Skill test. The bulk-order challenge and product-care enquiry are optional practice.
Start a new Work chat. Use the five PDFs in your connected Google Drive, or upload them here. Stay in the same chat through Steps 1–7. Send one prompt at a time and review the result.
Review all five files, approve a plan and organise copies in your own Drive.
Confirm the selected files and review the proposed plan before approving.
Review the five Little Lanterns PDFs (01–05) in Google Drive or uploaded here, as of 12 November 2026. Confirm the file names.
Identify current, outdated and unapproved files. Flag conflicting information and anything you need me to clarify.
Propose Current, Archive and Needs review folders; explain where each file belongs. Flag anything unclear. Wait for my approval before changing files.
Ask ChatGPT to create a new Drive folder and organise copies of the reviewed files where supported. Verify the result against your approved plan. If Drive actions are unavailable, continue with approved PDFs 01–03 in the chat.
I approve the plan: 01–03 current; 04 superseded; 05 unapproved.
Create a new private “Little Lanterns — Session 3” folder in my Google Drive; use a unique name if needed.
Add copies of the five reviewed files from Drive or this chat and organise them into Current, Archive and Needs review subfolders. Keep names and contents unchanged; leave originals and sharing unchanged. Delete nothing.
Show the new folder link and where each file went. Tell me if anything couldn’t be completed.
Use the approved sources to draft, check and improve one customer reply.
Tara · Gmail
Use Gmail to find the exact subject “[Workshop demo] Gift for a colleague”. Show Mei’s question; don’t draft yet. If multiple matches, let me choose.
Participants
Open Customer_Enquiries/Q1_Mei_Gift_Enquiry.txt. Paste the full question into your approved-source chat. Or use Q1 in Step 4. You do not need access to Tara’s inbox.
Use the same Q1 you just retrieved or opened.
Draft a warm reply under 120 words using approved sources. Flag anything to check; keep sources in a separate internal note. Draft only.
Q1 — Mei’s gift enquiry
Hi Little Lanterns,
I’m looking for a gift for a colleague. What is included in the S$48 gift set? Could you add a short thank-you message? What would the total be including standard delivery in Singapore?
Thanks, Mei
Send these as separate messages.
Check the draft against the source sections. Remove unsupported claims or promises.
Read the result and open a cited source.
Improve your response with feedback:
Ask ChatGPT to revise Mei’s reply:
Include the card message ‘Thank you for all your help!’ and say whether the mug is dishwasher safe. Keep it under 80 words. Keep sources and owner checks separate.
Create, inspect and install your Skill, then test it on a different question in a fresh chat.
Stay in the same chat so ChatGPT can use the folder you created earlier.
Create and save a private “Customer Support Responder” Skill for product and policy questions.
Save the method: understand the question, consult approved sources, draft, check, and flag missing or conflicting facts.
“Current” is the subfolder we created inside “Little Lanterns — Session 3” in Google Drive. It contains approved PDFs 01–03: product specifications, product care and service policies. Save its link as the source location. If inaccessible, ask me to upload those PDFs.
Write a warm email or spoken draft under 80 words. Keep sources and owner checks separate. Invent no commitments.
Save the instructions and source link. Keep changing prices and policies in the source files; leave Mei’s one-off details out of the Skill.
Show the saved Skill, or explain why saving failed.
Open the actual saved Skill and inspect what it retained. Complete any installation prompt and check privacy before testing in a fresh chat.
Open the saved Skill and check:
1. Purpose and sources: product/policy questions, exact Current folder and PDF fallback.
2. Output: a draft under 80 words, with separate sources and human checks.
3. Gaps or conflicts: ask or flag them. Invent no commitments.
Confirm the saved instructions, installation and privacy. Correct gaps before testing.
After changing instructions or sources, rerun Q2.
Select your saved Skill. Attach sources if asked, then check the answer.
Use my saved Skill “Customer Support Responder”.
A customer is standing in the shop with a 15-inch laptop. They’re asking whether our tote will fit it, protect it from knocks and keep it dry in the rain. What should I tell them?
Try this after Q2. Skip if time is short.
Use my saved Skill “Customer Support Responder”.
A customer wants eight gift sets by Friday and says they’ve seen a 10% promotional offer. Can I confirm availability, delivery and the discount? What should I tell them, and what needs checking?
Update the Skill with feedback:
After reviewing the answer, send this feedback to update your saved Skill and revise the answer.
Update my saved “Customer Support Responder” Skill: use “our customer service team will check” in the customer draft. Keep owner approval in the internal note for stock, delivery and discounts. Save, show the changed rule and revise the answer.
Organise your knowledge: Show the approved source structure. Explain one file you excluded.
Answer customers: Show your checked customer reply. Open one supporting source.
Save a reusable Skill: Show the saved Skill and Q2 result. Name what still needs checking.
Today’s session: Use the fictional Little Lanterns files and questions.
Your own business: Use approved information; check access and sharing.
No history unless saved; not used for training while temporary.
A copy may be retained for up to 30 days for safety.
Confirm current versions and approvals; flag old policies and draft offers.
Check facts, prices and promises. Keep internal notes separate.
Session 1 — Turn data into a decision
Ask useful questions and check the numbers. Separate evidence from assumptions.
Session 2 — Choose, create and improve
Compare ideas, then choose a direction. Give specific feedback and plan a small test.
Session 3 — Make the workflow repeatable
Organise approved sources and check answers. Save a Skill; test it on a new question.
Across all three: give context, check the work and use your judgment.
Choose a recurring customer question. Organise its source documents, check one useful response, save the method as a Skill and test a different question. Keep changing facts in the source documents.