Prompting Iris
How to ask Iris questions that get precise answers and canvases that show what you meant.
Iris works best when a question names the measure, the scope, the grouping, and the time range.
| Part | Examples |
|---|---|
| Measure | cost, amortized cost, savings, excess per day, coverage |
| Scope | AWS only, prod accounts, the payments team, Bedrock models |
| Grouping | by team, by service, by account, by model |
| Time range | last 30 days, this quarter, since Sep 1, vs previous month |
Weak and strong prompts
| Instead of | Try |
|---|---|
| Show me costs | Amortized cost by provider, last 30 days, vs previous period |
| Why is AWS expensive? | Which AWS services grew most this month vs last, and which accounts drove it? |
| AI spend | Monthly AI cost by model and account, this year |
| Savings | Open savings over $5K/mo with no owner, grouped by team |
Prompts by job
- Brief me on spend this quarter.
- Build a canvas of cost by team this month vs last month.
- What share of spend is covered by commitments?
- Why did AWS spike on Sep 1?
- What changed in EKS cost this week?
- Which resources are behind ANM-175?
- What share of spend has no owner tag, by account?
- Top 20 untagged resources by monthly cost.
- What are the three biggest levers to cut cost?
- Quick wins over $1K/mo that nobody owns.
Tips
- Use High depth for why questions, and Low for lookups.
- Follow up in the same chat. Iris keeps the context.
- Pin answers you'll want again to a canvas.
- If Iris picked the wrong scope, say so: only production accounts.