The Analyst is the agent you talk to by default. You type a question about your data in your own words; it works out what to count, reads the data, and answers. Say you want to know whether last month’s visits really went up — you ask, and you get a figure plus a plain account of how that figure was counted, so you can decide whether to rely on it.
What you’ll need
- Some data to ask about — see Datasets for how to pick what a conversation reads.
Asking a question
- If you want Crunch to work harder on this one, set the effort first with the brain button (see Thinking and effort below).
- Type your question in the message box, in your business words. One question per turn works best.
- Send it, and read the answer from the top down — its parts are explained next.
Reading an answer
An answer can carry these parts, always in this order:
- The lead — the answer in a sentence or two. After each figure there’s a ↗; click it to open that figure’s table or chart in the side panel.
- ⚑ Check before relying on this — anything the data could not confirm. It sits right under the lead, so you see it before you act on the figure.
- How I read your question — when a word could mean two things, this says which reading Crunch took, for example “I read visits as sessions, not page views”. Crunch never stops to ask you; it picks the likelier reading and tells you, so you can reply “answer it the other way” if that isn’t what you meant.
- How this was counted — the definitions it held and the rules it used for edge cases, written from what it actually ran.
- Not established — any part of your question it could not answer, with why and what would answer it. Reply “yes” to have it try.
- Where this sits — what is still open, or the natural next question.
Thinking and effort
The brain button beside the microphone sets how hard Crunch works on your next question:
- Thinking (on by default) — Crunch reasons step by step before answering.
- Low effort (the default) answers directly. Medium breaks the question down and checks its work. High does the same with more reasoning, and uses more credits — the units your usage is counted in.
- Time limit — by default 5, 8 or 20 minutes for low, medium or high effort. When the time is up, Crunch asks “Do you want me to keep going on what’s still open?” Reply “keep going” to let it continue. No time limit never pauses, which also means it can use many credits.
- Show every step — shows each step of the work, instead of one live status line.
These settings are kept per conversation.
How to tell it worked
You get a lead with a figure and its ↗, and a How this was counted section that matches what you meant. If the reading or the counting isn’t what you had in mind, say so in your next message.
Good to know
- Tip: Say what a value means once, up front (“a trial is anyone on the free plan”). Crunch keeps it for the rest of the conversation, so you don’t have to repeat yourself.
- Tip: Keep to one question per turn. A turn that asks one thing is far easier to check than one that asks three.
- You can also ask a product question, such as “how do I upload a file?”. Crunch answers it from these help pages.
Related
- Solve a problem in a conversation — how to phrase and follow up a question.
- Charts, tables and artifacts — what you can do with each ↗.
- Plan agent — for a question that is really several questions.
- Choosing a model — how the model and effort affect your usage.