Design and document an AI-assisted workflow for a local business

Goal

Choose a real local business, ideally one you know well, and design an AI-assisted workflow that takes a repeated, text-heavy task off its team's hands. Document it well enough that the owner could understand it and a developer could build it. You won't build it in a live system, and you won't use any real customer data.

Steps

1. Choose the process. Pick a repeated task that involves text, such as enquiry emails, booking requests, replies to reviews, invoice questions or weekly reports. Explain why it's a good fit, using the checks from "What AI is good at, and where it fails".
2. Map the current process step by step: who does what, with which tools, and roughly how long it takes. If you can talk to someone at the business, use their numbers; otherwise mark your estimates clearly as assumptions.
3. Design the new workflow. Draw it with the trigger, actions, conditions, AI steps, human review points, logging and alerts, and explain what happens when each step fails.
4. Write the approved information the AI steps will rely on, such as prices, policies and common answers, taken from public sources or clearly marked as drafts.
5. Write and test the prompts. Test each prompt on at least ten realistic example inputs that you write yourself. Record the results and how you improved the prompts.
6. Cover privacy and fairness: what data flows through the workflow, which data rules apply, what you've kept out of the AI steps, and any risk of unfair outcomes.
7. Compare two ways to build it, with their pricing models, set-up effort, review time and maintenance. Don't quote specific prices.
8. Write a one-page summary for the owner in plain language: what it does, what it won't do, where people stay in control, and what to measure after launch.

Rules

• Use no real customer or personal data. Write realistic examples yourself.
• Don't contact the business's customers. If you talk to the owner or staff, explain that this is a training exercise, and don't promise them anything.
• Don't claim hours saved or accuracy you haven't measured. Label estimates as estimates.

How to submit

Put everything in one shared document or folder, give Skillhancer Labs staff view access, and submit the link below with any notes for the reviewer.

What reviewers look for

• The process is a genuine fit for AI, and the reasons are clear
• Human review sits wherever mistakes would be costly
• Prompts are specific, tested and improved based on the results
• Privacy, fairness and failure handling are addressed concretely
• The cost comparison covers pricing models and hidden costs, without invented figures
• The owner's summary is honest, clear and free of hype