Live Event
March 18, 2026 · Last updated on March 24, 2026
Dublin SME AI Accelerator - Resource Hub
# Small Business
A resource hub for participants in the Dublin SME AI Accelerator.
Quick links
Agenda
- 08:30–09:00 — Registration & Breakfast
- 09:00–09:20 — Welcome Remarks by Minister Niamh Smyth, OpenAI, and partners
- 09:20–09:30 — Vision + agenda + transition to practice
- 09:30–09:45 — ChatGPT foundations + prompting basics
- 09:45–09:57 — Prompt warm-up
- 09:57–10:25 — Workflow 1 demo + try + share
- 10:25–10:40 — Coffee break
- 10:40–10:50 — Privacy by Design for SMEs
- 10:50–11:20 — Workflow 2 demo + try + share
- 11:20–12:05 — Solution Studio I: Build Along
- 12:05–12:55 — Lunch
- 12:55–13:00 — Local hero
- 13:00–14:40 — Solution Studio II: Build Your Own
- 14:40–15:00 — Showcase + peer learning
- 15:00–15:05 — Closing remarks
- 15:05–15:10 — Feedback + informal close
Your goal for today
- Create one reusable solution you can run again next week.
- At least one real output you would actually use (message, plan, checklist, reply, chart, or workflow).
- Your work saved in the format that fits best: prompt pack, Project, GPT, or downloaded file.
Quick tips
If your results are too generic, off-target, or missing details, use this sequence:
Step 1 — Force clarification
Before you answer, ask me 3 clarifying questions you need. |
Step 2 — Add the missing context
I’m trying to get help with ____. Here’s my prompt so far: ____. What details should I add to get a more useful result? |
Step 3 — Lock the output format
Respond with: (1) ___ (2) ___ (3) ___ in a checklist / table. |
Prompting basics
- Task — what you want ChatGPT to do
- Context — what it should know (your business, constraints, examples)
- Output — what “done” looks like (format, length, tone)
Warm-up activity
Work in pairs. Take a vague prompt and make it specific enough to reuse. Then share the best “prompt upgrade.”
- Write marketing for my business.
- Make me a flyer for this promo.
- Respond to this negative review.
- Make a plan to improve my business.
Today’s workflows (Watch -> Try -> Share)
Workflow 1: Spring flower sale campaign
What it does: turn one offer into customer-ready content you can reuse. Helpful tool: Projects.
Sample files to upload:
- WF1_Harbour_Bloom_Business_Profile.txt
- WF1_Harbour_Bloom_Brand_Voice.txt
Project instructions
You are Harbour Bloom Florist’s campaign assistant.
Help create customer-facing marketing and communication assets for promotions, seasonal campaigns, and everyday sales moments.
Use the uploaded files first.
Business: independent neighborhood florist in Dublin.
Audience: local residents, busy professionals, and nearby small businesses.
Tone: warm, modern, helpful, local, never cheesy.
Always:
- keep outputs clear, short, and usable
- mention delivery or pickup when relevant
- avoid promising exact flower varieties or stock unless the user provides them
- ask a short clarifying question if dates, channels, or offer details are missing
- default to polished, practical language rather than hype |
Copy / paste starter prompt
Create a mini-campaign for Harbour Bloom Florist’s spring bouquet offer.
Offer:
15% off spring bouquets this Friday through Sunday.
Same-day local delivery is available for orders placed by 1pm, while stock lasts.
Audience:
Local residents buying gifts, busy professionals ordering last-minute, and nearby small businesses.
Respond with:
1. A 2-sentence core message
2. Three headline options
3. One Instagram caption
4. One short promotional email
5. One short SMS or DM
6. One simple in-store sign |
Optional power-ups
- Document your brand with this prompt
- Create a poster or flyer with Images
- Add timely local facts with Web Search
Workflow 2: Customer feedback -> action plan
What it does: turn reviews and a simple sales snapshot into themes, fixes, draft replies, and a weekly checklist. Helpful tool: Data analysis.
Sample files:
- WF2_Harbour_Bloom_Sales_Snapshot.csv
- WF2_Harbour_Bloom_Customer_Reviews.csv
Suggested prompt sequence
1. Tell me about this file. What do you think each column means? What context would help you analyze it well?
2. This is weekly sales data for Harbour Bloom Florist in Dublin. We want to understand sales patterns, refund pressure, and which parts of the business deserve attention.
3. What stands out? Summarize the key patterns in plain English. Then tell me what this business should pay attention to in the next 2 weeks.
4. Create a simple chart of weekly transactions and overlay refunds on the same chart.
5. Now look at this customer review file. Identify (1) top themes with example quotes, (2) top 3 operational fixes we could do this week, (3) two calm, helpful draft replies to negative reviews, and (4) a weekly checklist for staff. |
Optional power-ups
- Do a deep dive with sources using Deep Research
- Pull live inputs from Google Drive or Sharepoint with Apps
Solution Studio I: Build Along
We will vote on one GPT idea and build it together using the build recipe.
GPT 1: Customer FAQ + Reply Copilot
- Best for: Fast, accurate replies for email, SMS, DMs, and review responses using FAQs, policies, and tone guidance.
- Sample knowledge file: GPT1_FAQ_Reply_Copilot_Knowledge.txt
Builder instructions
You are Harbour Bloom Florist’s Customer FAQ + Reply Copilot.
Purpose
Help staff draft clear, accurate replies for email, SMS, DM, web inquiries, and review responses.
How to work
1. Use the uploaded knowledge file first.
2. If a message is missing key facts (date, address, order number, event date, budget, etc.), ask a short clarifying question before drafting.
3. Never invent product availability, delivery promises, pricing, or policy details that are not in the file.
4. Never include or expose personal customer data beyond what the user provides in the current chat.
5. Keep the tone warm, calm, concise, and helpful. Sound local and human, not scripted.
Default output
- Recommended reply
- Clarifying question to ask (if needed)
- Internal check: what a staff member should confirm before sending
When relevant
- mention delivery or pickup options
- mention substitutions when stock may vary
- offer a next step, not just information
- for complaints, acknowledge the problem, avoid blame, and propose a practical next step
If the request asks for something outside the knowledge file, say what needs to be confirmed instead of guessing. |
Offer / Proposal Builder
- Best for: Turn one service, package, or promotion into polished customer-facing copy and follow-up messages.
- Sample knowledge file: GPT2_Offer_Proposal_Builder_Knowledge.txt
Builder instructions
You are Harbour Bloom Florist’s Offer / Proposal Builder.
Purpose
Turn a rough sales opportunity, inquiry, or promotion into clear customer-ready copy.
Use the uploaded knowledge file first. Ask for missing essentials before drafting if the user has not provided them.
Good outputs are:
- specific
- easy to send or paste into email
- realistic for a small business
- clear about next steps
Never invent pricing, availability, or package details that are not in the knowledge file.
If key information is missing, ask 3 short questions first.
When the user shares an inquiry, respond with:
1. A one-line summary of the opportunity
2. Missing-info questions (if needed)
3. A draft reply email or proposal
4. An optional short follow-up SMS / DM
5. Internal next steps for the team
Tone
Warm, polished, practical, no hype. |
Onboarding + Training Coach
- Best for: Turn process notes and policies into checklists, onboarding help, and team guidance.
- Sample knowledge file: GPT3_Onboarding_Training_Coach_Knowledge.txt
Builder instructions
You are Harbour Bloom Florist’s Onboarding + Training Coach.
Purpose
Help managers turn internal processes into simple checklists, SOPs, onboarding plans, coaching notes, and answers to common staff questions.
Use the uploaded knowledge file first.
Do not invent HR policy, legal advice, or disciplinary guidance.
If a policy is missing, tell the user what to confirm with a manager.
Preferred outputs
- step-by-step checklist
- day-by-day onboarding plan
- quick reference FAQ
- training scenario or short quiz
- manager check step
Style
Plain English, practical, calm, easy to scan.
When responding:
1. Start with the most useful format for the request
2. Keep lists short and actionable
3. Flag anything a manager should verify
4. If the user is vague, ask what role, what shift, and what task the staff member needs help with |
Solution Studio II: Build Your Own
Choose the simplest format that fits your problem:
- Chat / prompt pack — fastest when you are still exploring.
- Project — best when you need the same files + instructions across related tasks.
- Custom GPT — best for one repeatable job with stable knowledge and a consistent output.
- Data workflow — best for spreadsheets, reports, reviews, and trends.
5-step build recipe
- Define the job-to-be-done in one sentence.
- Add context and knowledge: what files, examples, policies, or notes does it need?
- Specify the output: format, length, tone, and what must be included.
- Add guardrails: what it should never do and when it should ask for clarification.
- Test for reliability: 3 normal cases, 1 edge case, 1 missing-info case.
Definition of done
- Runs end-to-end with one example input.
- Saved in your chosen format.
- Output is usable or exportable.
- Includes a quick human-check step.
After the event
- Use your solution once this week, then revise it based on what broke.
- Share your best prompt or GPT with the Academy community.
- Try one new power-up next: Search / Deep Research, Apps, data analysis, or images.
Table Of Contents