Log inStart free
RUEN

Three ways to put AI next to your data

AI for spreadsheets, no code · Lesson 2 / 22

Choosing where the model sits

There are exactly three ways to connect a model to a spreadsheet, and each carries a different cost when it goes wrong. Understanding the difference removes half the frustration.

1. Chat beside the sheet

Two windows open: the spreadsheet and the chat. You describe the task, take the formula back, paste it into a cell.

  • Plus: the model sees no data unless you paste it. Maximum control.
  • Plus: you can argue, clarify and ask for explanations.
  • Minus: manual copying, and it is easy to mix up ranges.

2. The built-in assistant

Copilot in Excel, Gemini in Google Sheets, smart panels inside SaaS tools. The assistant sees the whole sheet and can insert formulas and columns on its own.

  • Plus: it knows your structure and fills in ranges itself.
  • Minus: your data goes to the vendor's cloud. For personal data and trade secrets that is a decision, not a detail.
  • Minus: changes land instantly, so it is easy to miss that a column was rewritten.

3. A model call inside a cell

Functions like AI in Google Sheets, or add-ins that call a model from a formula. Each row is processed separately: categorising comments, pulling a city out of an address, scoring sentiment.

  • Plus: ideal for "label 500 rows of text" work.
  • Minus: the output is non-deterministic. A recalculation can assign a different category to the same row.
  • Minus: slow, and it hits request limits.
Choosing:
formula or logic          -> chat beside the sheet
routine on your structure -> built-in assistant
row-by-row text labelling -> model call in a cell
anything sensitive        -> chat only, no data pasted
Insight. Output from an in-cell model call has to be frozen straight away: copy the column and paste it back as values. Otherwise yesterday's labels stop matching today's and your report numbers drift apart for no visible reason.
Common mistake. Letting the assistant work in a file with no copy. It edits the sheet directly, and people usually remember version history right after saving over the original.
Pro tip. Keep a separate sheet called Scratch and park every assistant-generated intermediate formula there. Your main sheet stays clean and you can still see where each number came from.

Cheat sheet

  • Chat: most control, least convenience.
  • Assistant: convenient, but your data leaves the building.
  • In-cell calls: for text, and freeze the result as values.
  • Copy the file before letting an assistant into it.
1. What is the main risk of an in-cell model call?
2. What does chat beside the sheet win on?
3. What should you do before letting an assistant into a working file?
Task — checked by AI

Take your own working spreadsheet and assign three real tasks from it to the three ways of working with AI: one to chat beside the sheet, one to a built-in assistant, one to row-by-row labelling. For each, write what the task is, why that channel fits, what risk the channel carries, and how you close that risk. Separately, name one task of yours that must never go through a cloud assistant, and explain why.

← Back

🔒 Answer the question correctly to move on to the next lesson.

Three ways to put AI next to your data — AI for spreadsheets, no code — Skilvy