Ask about any dataset

Your revenue looks flat this quarter, and you suspect the real story is underneath: more customers spending less, or fewer spending more. You don’t need website tracking to find out. Crunch isn’t only for clickstream — the trail of clicks and page views a website records. An orders export, a listings feed or an operations log works the same way: you upload it, say what you need to know, and Crunch works out which columns can answer.

This guide walks you from a file on your computer to a first useful answer.

What you’ll need

• A file with rows and columns — Excel, CSV or Parquet. If you can open it as a spreadsheet, it will almost certainly work.
• A BumbleB account. The Free plan is enough to start — sign up if you haven’t.

That’s all. There’s no tag to install and no connector to set up. Each file you upload becomes a dataset: a table that lives in your own workspace.

Step by step

1. Export the data from wherever it lives today — your store’s admin, your booking system, a shared spreadsheet. Most tools have an export or download option that gives you a CSV.
2. Upload it from the datasets pill in the message box, with Upload a file…. Uploading files covers size limits and what happens when you upload the same file again.
3. Make sure the dataset is ticked in the pill, then ask your question in plain words, the way you’d ask a colleague who has the file open.
4. Tell Crunch what the file can’t tell it (see the next section). Do this once, at the start — Crunch keeps it for the rest of the conversation.
5. Ask the next small question. One step at a time gets you further than one big question — see Solve a problem in a conversation.

Tell Crunch what it can’t see

A file carries values, not meaning. A rating column full of 0s doesn’t say whether 0 means “terrible” or “not rated”. So say once, at the start, anything a stranger opening your file would get wrong:

• What one row is — “each row is one order line, not one order”. This changes every count.
• Special values — “a rating of 0 means not rated”, “-1 means unknown”. Left unsaid, they look like real numbers and get averaged in.
• Codes and currencies — “country code 1 is India; amounts are in rupees”.

How to tell it worked

When the upload finishes loading, the dataset is selected for the conversation and you can ask straight away. Your first answer leads with the result and then shows its working, including How I read your question and How this was counted. If both match what you meant, you’re set up. If not, say so in one line and ask again — Read an answer explains each part.

An example

A store’s revenue looks flat, and the owner doesn’t believe it’s that simple. They have one orders export and nothing else.

A flat total was hiding two moves in opposite directions. That points straight at the next question: why did orders get smaller? From there, split it the way your business is organized — by region, by channel, by product — one small step at a time.

Good to know

• Crunch reads what the file holds. If the columns can’t answer a question, it says so plainly and tells you what would answer it, rather than guessing.
• Newer numbers mean a fresh upload. An uploaded file is a snapshot. When you want this week’s orders, export and upload again, and ask in a new conversation — a conversation keeps reading the version it started with.
• Two files answer together when they share a column — an order ID, a customer ID, a date. See Ask one question across two datasets.

Tip: if a column name only makes sense to your team, explain it in your first message. It costs one sentence and saves a wrong answer.

Related

• Uploading files — formats, size limits and versions.
• Solve a problem in a conversation — how to break a big question into small ones.
• Read an answer — what each part of an answer tells you.
• Ask one question across two datasets — when one file isn’t enough.