Prompt library

A short guide to writing prompts that get consistent, accurate results from your AI assistant when it's working with Xvantage data. Most "the AI got it wrong" issues are actually prompt issues — this page helps you fix them.

The four habits that matter most

1. Be specific

The more context you give, the less the AI has to guess. Names, dates, IDs, SKUs — all help.

Vague Show me my invoices
Specific Show me unpaid invoices from the last 60 days, grouped by customer, with totals

2. Add time frames

Without a date range, the AI may default to "all time" and get a slower, noisier response. Always bound your queries to a relevant window.

No frame What orders are open?
Scoped What orders placed in the last 30 days are still open?

3. Break complex tasks into steps

When you're asking the AI to do three things at once, help it plan. “First…, then…, finally…” works well.

All at once Summarize my top customers and find renewals and draft email reminders
Stepped First, list my top 5 customers by revenue this quarter. Then, for each one, pull their upcoming renewals. Finally, draft a short reminder email for each.

4. Ask for structured output

Tables are easier to read than paragraphs for lookups. Specify the columns you want.

Prose Tell me about my overdue invoices
Structured Return overdue invoices as a table with columns: customer, invoice number, amount, days overdue

Useful patterns

  • Force a tool. If the AI is answering from memory instead of calling Xvantage, say: “Use the Xvantage invoice tool to find…” or “Check Xvantage for…”
  • Chain tools. “Pull the quote, then check product availability for each line item” tells the AI to make two tool calls in sequence.
  • Ask for reasoning. “Explain how you got this answer” when you want to understand or verify the AI's work.
  • Reset context. If the AI gets confused partway through, start a new conversation. Stale context is a common source of wrong answers.