Zamino

How to phrase a good question

Real examples of vague vs specific prompts — and why time windows, filters, and metric names make Zamino answer better.

Chat with your data4 min read

Zamino is happy to answer almost anything, but the quality of the answer scales directly with the specificity of the question. Here's what we've learned from thousands of real Chat sessions.

✅ Three things that make a question "good"

  1. A time window. "Last month" is always better than no window. "Q1 2026" is even better.
  2. The unit you care about. By customer, by SKU, by channel, by category — pick one (or more).
  3. The metric name. Say "net revenue" (or "gross margin", "MRR", "COGS") rather than leaving it implicit.

Zamino will fill in reasonable defaults if any of these are missing — and tell you what it picked — but you'll iterate less if you say it up front.

Side-by-side: less good vs good

💡 Try one of the suggested prompts

When the chat is empty, Zamino offers five suggestions you can click to fill the composer:

  • 📈 Analyze spending trends
  • 🚨 Review unusual transactions
  • 💰 Summarize cash flow
  • 📊 Compare monthly revenue
  • 🧾 Check top expenses

These are intentionally broad — Zamino will infer reasonable defaults (last 90 days, your default currency) and tell you what it picked so you can refine.

When you don't know the metric name

That's fine. Zamino will either ask, or pick a reasonable default and tell you what it picked. Look at the result and either confirm or say "use gross margin instead" — Zamino re-runs with the updated definition.

Follow-ups remember context

Once you've asked a question, subsequent messages in the same thread inherit context. So after asking about revenue by month, you can say "just the last 30 days" or "break it by region" and Zamino interprets correctly. No need to re-state the metric.

When Zamino pushes back

If Zamino can't answer — usually because a source isn't connected, the data doesn't exist, or the question is genuinely ambiguous — it says so plainly and suggests next steps. It won't silently invent a number. That's the most important rule the model lives by.

Where to go next