Choosing a Model and Writing Good Prompts
Which AI model to use with Flourish, and how to ask for what you need: plain-language tips for operators working in Claude or ChatGPT.
You don't need special training to use AI with Flourish. If you can explain a task to a coworker, you can explain it to Claude or ChatGPT. This article covers which model to pick and a few habits that get better results.
Which model should I use?
Most AI clients let you choose a model. Bigger models are more capable, but slower and they use up your plan's usage faster.
Task | Claude | Examples |
Quick lookups | Haiku | "How much Gelato flower do we have?" "What's the status of order 1042?" |
Everyday work (start here) | Sonnet | Drafting an order from an email, updating items, checking what ships tomorrow |
Long, multi-step jobs | Opus | Working through a week's production plan, a large spreadsheet, or many BOMs at once |
If the assistant seems to miss the point or gives answers you don't expect, try a larger model. ChatGPT offers similar tiers.
Tips for good results
Name the facility or license. Many companies have several. "In my Denver manufacturing license…" saves a round of questions.
Use the names you see in Flourish. Item, customer and vendor names as they appear in the app give the best matches. If the assistant can't find something, give it the exact item name or SKU.
Say what you want done, not how to click. "I want to make pre-rolls and eighths of Tropicana Truffle tomorrow. What do I have to work with?" works better than describing screens.
Give it the source. Paste the email, attach the PDF or photo, or share the spreadsheet. It will pull out the details and show you what it matched.
Work backwards. Ask outcome questions: "What's scheduled to ship tomorrow, and am I short on anything?" The assistant can combine orders, inventory and production to answer.
Typos are fine. The assistant is forgiving about spelling. It's precision about which item or customer that matters.
It will ask when it's unsure. If a request is ambiguous, a good assistant asks rather than guesses. Answer its question, and it continues.
Always review before you approve
For anything that changes data, the assistant shows you what it's about to do (for example, a table of order lines or item changes) and waits for your approval. That preview is your control. Check the customer, items, quantities and facility before you say yes. If something is off, tell it what to change and it will show you the updated version.
Save your process
If your team always does something a certain way, write it down once and give it to the assistant, for example: "Always use the Packaging Line 2 machine and the Morning shift for pre-roll runs, and put the batch number in ref field 1." In Claude, you can save instructions like these in a project so they apply every time.
How did we do?
Manufacturing: Build BOMs and Work Orders with an LLM
AI Security & Admin Checklist
