1. Is the problem specific?
Turn “we want AI efficiency” into one observable action, such as classifying inquiries or preparing an internal draft.
Practical checklist
Eight checks before
you build AI for trade.
Whether an automation idea is ready for development usually depends on process, data and permissions before it depends on a model name. Use this checklist for a first internal review.
Start here: If the team cannot describe the trigger, inputs, recipient and approval point, document the process before choosing a model, plugin or MCP integration. Passing this checklist only means the project is ready for further evaluation; it is not a commercial outcome.
The eight checks
2. Is the trigger clear?
Write when the action starts, who starts it and which situations must stay out of scope.
3. Are the inputs available?
List emails, spreadsheets, manuals, history or system fields, including versions, formats and gaps.
4. Who receives the output?
Name the salesperson, manager, service team, warehouse or system that will use the result.
5. Which actions need approval?
Keep pricing, delivery promises, contracts, claims and external sends behind an explicit review step.
6. Is access sufficient?
Check APIs, accounts, read/write scope, data location and network conditions. Missing access changes the plan.
7. How will it be tested?
Choose reviewable indicators such as field error rate, handling time, edit rate or knowledge coverage.
8. Who maintains it?
Agree on updates, exceptions, logs, rollback and future costs so the workflow has an owner after launch.
How should a trade business start before commissioning AI?
Choose one observable workflow and describe its inputs, recipient and approval points. With a redacted one-page brief, you can use WeChat or WhatsApp to discuss scope, current systems, access and acceptance. The contact entry starts a conversation; it is not a quote, project approval or outcome guarantee.
Prepare a redacted one-page brief
Write the business context, single first workflow, trigger, inputs, output format, recipient, approvals, system boundary and acceptance indicators. Remove unnecessary names, order numbers and secrets. Pricing, delivery scope, data location and maintenance are confirmed case by case.
A useful first scope can answer
- Which single step is in scope first?
- Which real or redacted materials will be tested?
- Who approves the result and at what point?
- What manual path handles failure?
- What evidence allows the scope to expand?
What this checklist does not replace
It does not replace security, compliance, procurement, contract or commercial reviews, and it cannot guarantee better conversion or revenue. Customer data, cross-border transfer, external models and third-party systems need review by the responsible owner under the real business conditions.
Maintained by Chenfeng AI. This is an owned practical resource, not legal, financial or security advice. Facts and scope are reviewed as the business changes.